A systematic map of generative AI guidelines and reporting in ecology and evolutionary biology: towards the framework of AI disclosure for Improved Transparency (AIdIT)

A systematic map of generative AI guidelines and reporting in ecology and evolutionary biology: towards the framework of AI disclosure for Improved Transparency (AIdIT) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review A systematic map of generative AI guidelines and reporting in ecology and evolutionary biology: towards the framework of AI disclosure for Improved Transparency (AIdIT) Szymon M. Drobniak, Martyna Cendrowska-Pek, Agnieszka Gudowska, and 12 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9160721/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 9 You are reading this latest preprint version Abstract Generative artificial intelligence (AI) is rapidly becoming embedded across scientific workflows, yet mechanisms for transparently documenting its use remain fragmented and weakly enforced. Focusing on ecology and evolutionary biology as a model discipline, we systematically mapped AI-related journal policies across 230 journals and assessed article-level compliance using a large sample of recent publications. To provide a reporting background, we also synthesised author contribution guidelines. Nearly half of journals provided no guidance on AI use, and where policies existed, they were largely generic, publisher-driven, and poorly translated into reporting practice. While author contribution statements were widely adopted, explicit AI disclosures appeared in fewer than 6% of papers, even in journals with formal AI policies. Text-mining of 124 guideline documents revealed highly standardised, precautionary language emphasising responsibility and prohibitions, with minimal operational guidance on acceptable uses or disclosure formats. To address this gap, we introduce AIdIT (AI disclosure for Improved Transparency), a standardised, taxonomy-based framework for reporting AI use across all stages of the research lifecycle. AIdIT integrates structured categories of AI use, human oversight statements, and machine-readable outputs to support reproducibility, accountability, and comparability. Together, our systematic evidence synthesis and proposed framework highlight an urgent need to normalise AI transparency as a core component of open research practice. Artificial Intelligence generative AI LLM large language models transparent science open research practises reporting guidelines Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction The explosive rise of generative artificial intelligence (AI) in scientific workflows has ignited debates about the transparency of its use among researchers across disciplines. Generative AI is increasingly used not only in neutral applications (e.g., text editing, spellchecking, style improvement) but also, e.g., to generate analytical code, reframe mathematical models or screen/curate large datasets. One field that has seen a rapid proliferation of AI applications is ecology and evolutionary biology (EEB). EEB integrates field observations, experimental studies, large comparative datasets, and advanced statistical modelling, all of which converge within a highly collaborative team 1 , 2 . This interdisciplinary nature makes the field particularly well placed to benefit from emerging AI tools, but also especially vulnerable to opacity in how these tools are used and reported. Since the public release of accessible large language models (LLMs) in late 2022, generative AI has rapidly become a routine assistant across these tasks, supporting information generation (e.g., literature scoping, text drafting, generating code snippets), content refinement (e.g., text polishing, bibliography formatting, annotating data), and content comparisons (e.g., comparative literature reviews, summarising multiple data sources) 3 – 5 . A Nature survey revealed substantial uptake of such tools among researchers, many of whom already use them to brainstorm ideas, assist in research design and refine writing 6 . According to the survey, over 66% of researchers appreciated faster data processing with AI, 58% noted increased effectiveness in previously infeasible computations, and 55% admitted that AI can help save time and resources. On the other hand, 69% of respondents noted that AI can generate results without proper insight or understanding, and 53% expressed concerns about the potential dangers AI poses to scientific reproducibility 6 . This latter concern resonates well with a more recent poll 7 according to which, 70% of surveyed researchers accept using AI to edit papers (roughly half of this group only if AI use is properly disclosed), but only around 15% thinks the same about using AI in translations, summarising papers or generating initial drafts (additional 20–30% would allow for each of these uses conditional of proper disclosure of AI methods employed). Clearly, generative AI now touches multiple stages of the scientific process, from study conception through analysis and interpretation to writing and dissemination, as well as teaching and student assessment. There are multiple benefits to the rapid expansion of the use of generative AI tools in research. They can lower language barriers, especially for non-native speakers, accelerate repetitive or algorithmic tasks, and potentially free up time for higher-order reasoning, conceptual synthesis and creative study design 8 , 9 . In principle, such efficiency gains could strengthen, rather than weaken, the quality and inclusiveness of research in EEB. A question arises: do these benefits outweigh the potential costs associated with the unethical or opaque use of AI tools? Being relatively new, widely available and fast-evolving, AI algorithms still evade many attempts to codify their use in informal norms or formal policies 10 , 11 . At the same time, being widely available and democratised, AI tools’ use often falls through the cracks of existing policies 12 , 13 . Without transparent and accountable use, the risks are substantial: LLMs may fabricate facts and citations, encode and amplify social and scientific biases, and produce fluent yet incorrect prose. Such issues are difficult to detect, particularly for non-experts 8 . This can spread misinformation, undermine critical thinking skills and blur the boundaries between human and machine contributions. These risks are particularly acute when AI tools are used unsupervised or without disclosure, for example, in educational settings where developing independent reasoning, methodological skills, and a clear sense of authorship is central. At the moment, only exposing cases of AI misuse can provide any sense of how much it may be affecting the current scientific record. Notably, beyond text generation, AI-based image manipulation and analysis have already contributed to retractions and corrections, exposing verification gaps in current quality-control procedures 14 . At the same time, peer-review workflows face confidentiality and accountability challenges when manuscripts or reviews are shared with external AI systems, and multiple independent evaluations demonstrate that currently available AI-detection tools remain inconsistent and unreliable 15 , 16 . In response to the growing use of generative AI in research, many publishers have begun issuing policies on its use in manuscript preparation and peer review. Most allow limited use for language editing or copy-editing, if this is acknowledged, with strict bans usually associated with AI-assisted peer review or uploading manuscript drafts to AI agents (i.e., uses not directly associated with authoring a paper) 17 . However, policy development has been uneven: major publishers, such as Springer-Nature 18 , Elsevier 19 and Taylor & Francis 20 have issued portfolio-wide guidance on acceptable uses of generative AI, but audits report limited consensus on their scope and enforcement 17 . Community-level resources have so far lagged behind the needs: to date, only two comprehensive attempts to define a taxonomy of AI uses have been proposed 21 , 22 . Both are preliminary and do not go beyond the classification of uses to actual reporting guidelines. As a result, in practice, authors, editors, and reviewers often operate in a landscape of partial, fragmented, and sometimes conflicting expectations. In EEB, open data, reproducible workflows, and explicit authorship contribution statements (e.g., via CRediT 23 and MeRIT 24 ) are increasingly becoming the norm. Yet if the use of AI tools remains largely invisible in published articles, readers cannot assess which elements of a study reflect direct human intellectual input, which were assisted by AI and which tasks (e.g., language editing versus data analysis) were delegated to automated tools. At the same time, inconsistent or vague guidance may discourage honest disclosure and exacerbate inequalities between authors who – due to educational background or local research culture – feel comfortable disclosing the use of AI, and those who don’t 25–27 . However, in EEB, it is still unclear (i) what proportion of journals in the field currently have explicit polices on generative AI, (ii) where such policies are located, (iii) whether journals that have formalised authorship statements are also more likely to address AI usage, and (iv) to what extent these journals’ policies translate into article-level disclosure in practice. To address these gaps, we need a field-specific assessment of AI-related journal policies, their visibility and their alignment with open science practices related to transparent authorship crediting. We surveyed the webpages of 230 EEB journals (including multi-disciplinary journals that routinely publish EEB studies) and documented the presence, location, and within-publisher consistency of AI policies. AI reporting statements and recommendations were also extracted and text-mined to identify consistent patterns of semantic skew, uniformity, and language specificity. For the same journals, we also examined the co-occurrence of AI guidance with author-contribution practices (including their compliance with CRediT). Finally, we reviewed a sample of recently published articles to see whether they follow these journal guidelines. We examined whether journals that require contribution disclosures or AI-use statements have higher reporting rates, and we also noted cases where authors voluntarily disclosed AI use even when it was not required. Although the issues raised by the use of generative AI are universal, we use EEB as a model case. Being an increasingly collaborative, diverse and fast-moving subfield of biological sciences, EEB has recently embraced many practices from the open science movement 28 (transparent crediting, growing use of registered reports and pre-registrations). A discipline-wide approach to generative AI would therefore be a natural next step towards greater transparency, making EEB a well-defined subfield that can serve as a proxy to biological sciences in general. Based on our overview, we identify current problems and inconsistencies in AI reporting and propose a single, standardised framework to make reporting clear, consistent, and complete. Although our work focuses on EEB studies, we designed the framework to be broadly applicable, with the goal of encouraging better open-research practices for the use of generative AI in research and education. Materials & methods Journal selection and assessment of formal policies on author contributions and AI In line with recent transparency recommendations, below we employ the MeRIT contribution reporting guidelines 24 to describe our methods and contributions. A list of journals in the fields of Ecology and Evolutionary Biology was compiled by SMD, following Pottier et al. (2024) 29 . The list included all titles classified under these two subject categories in the Clarivate Journal Citation Reports. To ensure coverage of multidisciplinary venues that frequently publish relevant research, we supplemented this list with 13 additional high-profile journals, including Nature, Nature Communications, Nature Climate Change, Scientific Reports, Science, Science Advances, Communications Biology, Proceedings of the National Academy of Sciences, PLoS Biology, Biological Reviews, Current Biology, eLife, and Philosophical Transactions of the Royal Society B. In April 2025, we screened (SMD, JR, MCP, AG, KJ, PP, KS, MG, WO, FB, NB, AA) all 230 journals to identify formal policies regarding (i) reporting of author contributions and (ii) the use of AI. For each journal, we first searched the “Instructions for Authors” (or equivalent) on the journal’s website. If no mention of author contributions was found, we used keywords such as “contribution”, “CRediT”, and “statement”. If no mention of AI was found, we searched the site using keywords including “artificial”, “intelligence”, “AI”, “generative”, and “LLM”. For each journal, we used a standardised extraction template to record: Website address (link to the information for authors); Presence of guidelines on reporting author contributions (Yes, mandatory / Yes, suggestions / No); Presence of guidelines on AI use (Yes / No); if Yes: Are any types of AI use explicitly permitted or prohibited? (Yes / No); Are there guidelines on where or how AI use should be declared? (Yes / No); Is there a suggested format for declaring AI use? (Yes / No); Verbatim copy of any relevant guidelines. Following initial extraction, a second extractor (from the same pool of extractors) rechecked all records without guidelines to confirm that no relevant information had been overlooked. We also verified whether any AI-related policies were located in sections of the journal website other than the “Instructions for Authors” (e.g., Editorial Policies, Publishing Ethics) and recorded their locations and links when present. Article screening and data extraction For article-level screening, we searched (SMD, JR, MCP, AG, KJ, PP, KS, MG, WO, FB, NB, AA, MZN) the Web of Science (Core Collection) using the following standardised advanced search string: (((SO=(JOURNAL_TITLE)) AND DT=(Article)) NOT DT=(Proceedings Paper OR Publication with Expression of Concern OR Biographical-Item OR Excerpt OR Book Chapter OR Record Review OR Note OR Book Review OR Correction OR Correction, Addition OR Database Review OR Data Paper OR Editorial Material OR Software Review OR Withdrawn Publication OR Retracted Publication)) where JOURNAL_TITLE was iteratively replaced with each journal name. Search results were sorted by date (newest first). From each journal, we exported metadata for the 20–30 most recent articles (retaining all bibliographic fields) and used these as a screening pool. From this pool, we selected the first 10 accessible, full-text research articles for detailed data extraction. For each article, we recorded the following information using a standardised extraction template: Contribution present (Yes / No): whether a statement detailing the contributions of all co-authors was included; Contribution CRediT (Yes / No): whether contributions were reported using the exact version of the CRediT taxonomy; AI statement present (Yes / No): whether any statement on the use of AI tools was included; AI statement location: article section where the statement appeared (Title page / Methods / Acknowledgements / Ethical note / Separate section / Other); AI model (Yes / No): whether the statement specified the AI model used (e.g. ChatGPT, Llama, Gemini, Copilot); AI usage (Yes / No): whether the statement described how AI was used (e.g. writing assistance, language editing, data processing); Negative statement (Yes / No): whether the article explicitly stated that no AI tools were used. If an AI statement was present, its text was copied verbatim. To ensure that declarations were not missed due to version differences, at least one randomly chosen article without an AI statement from each journal was cross-checked between its online and PDF versions. Articles that could not be accessed were excluded, and the reason for exclusion was recorded. Our initial assessment was strictly limited to the author guidelines/instructions, i.e., a journal section that provides formatting, content, and related information for authors considering submitting a manuscript. After completing our survey, we realised that some journals provided additional/sole AI-related policies in locations other than the author guidelines. Thus, we repeated the search to identify other locations where AI usage guidelines might be available (e.g., on publisher websites, in separate ethics/editorial policy sections). These, if present, were recorded under “Other locations”. Text mining of journal AI guidelines All journal-level AI guidelines identified during policy screening were subjected to a structured text-mining analysis (done by SMD) to characterise their content, consistency, and semantic patterns. Analyses were conducted on the verbatim guideline texts extracted from journal websites and stored in a centralised database. Only records containing substantive guideline text (non-empty entries exceeding 20 characters) were retained, yielding 124 documents out of 230 journals. Prior to analysis, publisher names were standardised to resolve inconsistencies across journal websites. Publisher identity was cross-validated using both declared publisher information and journal website domains, with discrepancies resolved using automated normalisation rules followed by manual verification where necessary. Basic document statistics (character count and word count) were recorded to assess heterogeneity in guideline length. Text preprocessing and analyses were conducted in R using the tidytext , tm , and quanteda packages 30 – 32 . Guideline texts were converted to lowercase, tokenised into unigrams and bigrams, and stripped of punctuation, numbers, and English stopwords. Tokens shorter than three characters were excluded, and stemming was applied to unigrams using the Snowball algorithm ( SnowballC package 33 ). Token frequencies were calculated to identify the most common terms and phrases across guidelines. To identify vocabulary distinctive to individual journals or publishers, we computed term frequency–inverse document frequency (TF–IDF) scores at the document level 32 . Keywords-in-context (KWIC) analyses were performed for selected AI-related terms (e.g., AI , artificial intelligence , language model , ChatGPT ) to examine how these concepts were framed, particularly with respect to permissions, prohibitions, and disclosure requirements. Higher-level semantic structure was explored using unsupervised topic modelling. Latent Dirichlet Allocation (LDA) models were fitted to document–term matrices using the topicmodels package 34 , with the number of topics chosen based on interpretability. Word co-occurrence networks were constructed to visualise associations among frequently co-occurring concepts, and sentiment analysis was applied using a lexicon-based approach ( syuzhet 35 ) to provide a coarse characterization of guideline framing. Results Systematic mapping Of the 230 journals in the field of ecology and evolution, 104 (45%) had no guidelines, not even a mention, regarding the potential use of artificial intelligence in the articles they publish (Fig. 1 A). Journals that mentioned the use of AI usually included a statement on permitted use (119 out of 126) and on how it should be disclosed (123 out of 126). We also found that most journals with any mention of artificial intelligence policies had guidelines for author contribution statements (see Fig. 1 B): AI disclosure requirements were present in 74% of journals with mandatory authorship statements and 68% of journals with suggested authorship statements. Interestingly, even journals belonging to the same publisher differed in this respect. For instance, some Wiley and Springer journals provide AI guidance, while others do not mention it at all (Fig. 1 C). Among the major publishers, Elsevier appeared to be the only one with a consistent AI policy across all its journals. Where information was provided, it was usually found on the journal's website, the publisher's website, or another external site (Fig. 2 ). An overview of how the published papers comply with the journal's guidelines on author contribution statements shows that authors readily follow these guidelines and also include such statements even when they are not mandatory (Fig. 3 A). Overall, author statements were found in more than 67% of papers. Among them, 44% used the exact version of the CRediT standard. The situation is very different when it comes to declaring AI use. We found that approximately 94% of papers do not disclose the use of AI (Fig. 3 B). Unfortunately, assessing the accuracy of the non-disclosure cases (whether they represent genuine absences of AI involvement vs. false negatives) remains impossible under the current field-wide heterogeneity of AI-reporting recommendations. Text mining Text mining was conducted on 124 journal-level AI guideline documents, which varied widely in length and detail (median 252 words; range 149–11,766 words), indicating substantial heterogeneity in how journals address AI use (Extended Supplementary Results (ESM), Fig. S1). Most guidelines were short and embedded within broader publishing or ethics policies rather than presented as stand-alone AI documents. Across the full corpus, unigram and bigram frequency analyses revealed a highly standardised vocabulary dominated by generic policy language. The most frequent terms and stems were associated with authors, responsibility, content, manuscripts, tools, and use, with AI-specific terms (e.g. artificial intelligence, AI, language model) appearing far less frequently and often embedded within otherwise boilerplate text. Bigrams such as artificial intelligence, generative AI, and AI tools were common but typically appeared in formulaic statements rather than in detailed procedural guidance (Fig. 4 ). Term Frequency-Inverse Document Frequency (TF–IDF) analyses showed that most journals exhibited very low lexical distinctiveness, reflecting strong within-publisher reuse of near-identical guideline text. Distinctive terms were primarily associated with publisher-level policies rather than individual journals, indicating that AI guidance is largely standardised at the publisher level and rarely tailored to specific journal scopes or disciplinary practices (for details, please see the Text Mining section of the Electronic Supplementary Materials). Even though the overall sentiment score indicated a positive lexical incline in the tone of most guideline texts (ESM, Fig. S4), most extracted policies sounded close to neutral. Still, the keywords-in-context (KWIC) analyses demonstrated that references to AI were most often framed in restrictive or cautionary contexts, emphasising the authors' responsibility, accountability, and prohibition of AI tools (see ESM). Consistently, mentions of permissible AI use (e.g. for language editing or stylistic improvement) were typically brief and weakly specified, while explicit guidance on acceptable analytical or data-processing uses was rare (see Fig. 5 for the overall distribution of recommendations’ specificity). Statements explicitly requiring disclosure of AI use were uncommon and often vague about where or how such disclosure should be made. Topic modelling identified a small number of dominant thematic clusters, primarily centred on (i) authorship and responsibility, (ii) ethical conduct and integrity, and (iii) permitted versus prohibited uses of automated tools (ESM, Fig. 7 ). Topics directly addressing transparency, reporting standards, or disclosure formats formed a minor component of the overall semantic structure, indicating that these issues are not central in most current guidelines. Word co-occurrence networks further highlighted the close association between AI-related terms and concepts such as responsibility, accountability, and integrity, whereas links to methods, analysis, or data were weak or absent (ESM, Fig. S5). Sentiment analysis supported this pattern, with guidelines predominantly exhibiting neutral-to-cautionary framing rather than permissive or enabling language. In line with prior evidence from similar text-mining analyses 36 , 37 , the most specific recommendation texts were also the most negative and restrictive (Fig. 6 ). Notably, the largest players on the market (e.g., Springer Nature, Wiley, Elsevier) cluster in the region of average specificity and average sentiment score. There is one interesting outlier: MDPI shows the sentiment score close to zero (neutral wording) with an extremely high specificity index. This may be due to the exceedingly high density of AI-related words in MDPI guidelines (ESM, Fig. S17), which stems from the fact that the only included guidelines texts for MDPI were very short (< 50 words). Nonetheless, MDPI was represented by only 2 journals, largely precluding any generalisations for this publisher. Discussion Our systematic mapping of generative AI policies in ecology and evolutionary biology (EEB) highlights a growing structural risk: AI tools are rapidly becoming embedded across research workflows, while mechanisms for documenting their use remain sparse, fragmented, and weakly formalised as verifiable policies. Nearly half of the surveyed journals provide no guidance on AI use, and where guidance exists, it is typically generic, publisher-driven, and poorly translated into reporting practices. Moreover, negative AI usage statements (i.e., intentional indication of no generative AI use) are virtually non-existent (< 1% of the analysed sample of articles). Comparing the results of our survey with the reported incidence of AI use in recent survey 7 suggests considerable opacity in the disclosure of LLM use. As a result, AI use remains largely invisible in the published record, despite mounting evidence that it already affects how scientific knowledge is generated, analysed, and communicated 5 , 14 , 26 , 38 – 40 . The absence of clear AI disclosure standards has tangible consequences. Without knowing whether and how AI tools contributed to data processing, code generation, text mining, or interpretation, readers cannot fully assess the provenance of results 35 . This opacity complicates research reproducibility, hinders methodological evaluation, and limits reviewers' and meta-researchers' ability to detect systematic biases or error propagation 38 , 40 . Recent high-profile cases of retractions and corrections linked to AI-generated images 14 , 41 , fabricated citations 42 , 43 , or unverifiable analytical steps 44 – 46 illustrate that these risks are no longer hypothetical but already materialising in the scientific literature. Importantly, these problems are not caused by AI use alone, but also by the lack of a systemic AI disclosure practice that could help flag such problematic studies during peer review. When AI-assisted steps are not transparently reported, errors introduced upstream - whether through hallucinated references, inappropriate model assumptions, or subtle data transformations - become difficult or impossible to trace. Over time, this can erode trust in published findings, particularly in fields such as EEB that inform conservation policy, environmental management, and public decision-making. The situation can be seen as mirroring a similar erosion of trust that eventually became one of the driving forces behind the open data trend, now, for many publishers, hardwired into scientific publishing. It remains to be seen whether the current disconnected AI reporting policies will follow a similar trajectory and catalyse a field‑wide movement toward explicit, unified disclosure practices. Precautionary policies without operational guidance Our text-mining analyses show that existing AI guidelines are relatively uniform. At the same time, they are dominated by precautionary language emphasising responsibility, integrity, and prohibitions, most notably the exclusion of AI systems from authorship, with limited specificity regarding acceptable uses, reporting requirements, or integration into existing open science and transparency practices. While these policies serve an important symbolic function, they provide little operational guidance for authors navigating real research workflows. Mentions of AI use in methods development, data analysis, modelling, or visualisation are rare, despite the growing prevalence of AI-assisted tools in precisely these domains 7 . Such narrow framing reflects a broader pattern in which AI is treated primarily as an ethical risk rather than as a methodological variable 10 , 25 , 39 . Such an approach may help publishers manage liability, but it fails to support transparent science. Moreover, it risks becoming rapidly outdated as AI tools diversify and become increasingly integrated into standard analytical environments, often as implicit, seamless components of software suites used in research and writing. The contrast between widespread adoption of author contribution statements and the rarity of AI disclosures is instructive. Contribution statements are normalised and standardised and are often embedded in submission systems. In contrast, AI disclosures are optional, inconsistently worded, and rarely linked to specific sections of a manuscript. This difference suggests that, consistent with broader surveys 7 , low disclosure rates are not primarily driven by authors’ unwillingness or bad faith. Instead, they reflect the absence of clear expectations, standardised formats, and workflow integration. As noted below, adding another vaguely specified declaration to an already complex submission process is unlikely to improve transparency and may even discourage disclosure. Social dynamics may further suppress reporting. Particularly for students and early-career researchers, admitting to AI use may carry a perceived stigma or fear of being judged as less competent or less original 10 , 12 . In the absence of clear norms, silence can appear safer than transparency. The consequences of poor AI reporting extend beyond journals. In higher education, AI tools are increasingly used in theses, coursework, and student-led research, often without clear guidance on acceptable use or reporting standards 12 , 27 . This creates ambiguity for students, supervisors, and examiners alike. Without structured disclosure, it becomes difficult to distinguish between legitimate assistance, inappropriate delegation, and outright misconduct. Complicating these considerations, there is currently little empirical understanding of how sustained use of AI affects the development of core cognitive skills such as critical thinking, synthesis, and creative problem-solving. The existing evidence is, at best, fragmentary and heterogeneous 47 – 49 . In this context, transparent reporting is not merely a matter of compliance, but a pedagogical tool that encourages reflection on how AI is used, where human judgment remains essential, and where over-reliance may be harmful. Study limitations Our study has several limitations. It focuses on journals within ecology and evolutionary biology and may not capture governance patterns in other disciplines, some of which may be more advanced—or substantially less developed—in their approaches to AI disclosure. Article‑level reporting was assessed using a recent snapshot of publications, which provides a useful baseline but cannot reveal temporal trends, journal‑specific transitions, or the effects of newly introduced policies. Likewise, text‑mining approaches necessarily abstract away nuance in individual guidelines, editorial decisions, and informal practices that may shape AI reporting in ways not captured by our coding framework. Finally, the absence of AI disclosure does not imply actual AI use, and interpreting non‑disclosure remains inherently ambiguous: it may reflect genuine non‑use, a lack of awareness, explicit journal exemptions (e.g., allowing unreported use for language polishing), or insufficiently integrated submission workflows. Taken together, these limitations highlight the need for longitudinal, cross‑disciplinary, and mixed‑methods research on AI governance - work that can track policy evolution, incorporate qualitative insight from editors and authors, and evaluate whether emerging standards genuinely improve transparency and research integrity. The AIdIT framework: standardizing AI disclosure across research and education To address the gaps identified by our map and text mining, we propose a new framework: AIdIT ( AI d isclosure for I mproved T ransparency; see Fig. 7 and Table 1 for a schematic structure). It can serve as a standardised, non-punitive norm for reporting AI use across the research lifecycle. AIdIT is designed not to restrict innovation, but to make AI-assisted work interpretable, auditable, and comparable, much like author contribution statements or data availability declarations. The framework is grounded in a taxonomy of AI uses and research stages (Table 1 ) mimicking the CRediT authorship roles classification. As such, it recognises that AI can contribute in qualitatively different ways across conceptualisation, literature review, data curation, formal analysis, methods development, visualisation, and writing. Rather than asking whether AI was used, the framework asks how it was used (e.g. generation, refinement, comparison) and where it influenced the research process. Table 1 Taxonomy of AI uses, categorised into three areas of application. See also Fig. 7 for practical implementation of categories. Research cycle stage Area of AI use Content generation Content refinement Content comparisons Conceptualisation - Idea generation - Idea evaluation - Gap identification - Literature review Data curation - Finding new data sources - Data cleaning - Data annotation - Building relations between datasets Formal analysis - New code generation - Statistical model formulation - Mathematical calculations/modelling - Refining existing code - Comparing coding approaches Funding acquisition - Generating proposal sections - Grant text editing and corrections - Benchmarking against funding body guidelines Investigation - New data generation - Assets generation - Augmentation of existing data - Modification of existing assets - Text mining of collected data - Summarising multiple data sources Methods - Identifying appropriate methods - Protocol generation/review - Methods/protocols refinement - Comparing multiple approaches Validation - Identifying validation approaches - Data cross-checking - Error management - Comparing results of validations Visualisation - Code generation for visuals - Generative images - Refinement of visual aesthetics - Evaluation against guidelines and practice standards Writing - Identification of literature to cite - Text generation based on prompts - Bibliography management - Text editing for style/grammar/spelling - Identifying contexts for own writing - Evaluation against guidelines and policies A central feature of AIdIT is its emphasis on human oversight and verification. When disclosing the use of generative AI, authors are prompted to describe not only the role of AI, but also how outputs were checked, validated, or approved. This shifts the focus away from tool prohibition toward accountability and quality control, aligning AI disclosure with established scientific norms. The taxonomy of AI uses proposed above was inspired by the AI usage cards designed by the computer science community 50 . Building on the initial ideas, we made the usage categories more general, user-friendly, and generalisable across a wider array of disciplines. Our taxonomy is more uniform in dividing the research process into clear, well-defined stages; it also includes areas of AI use that may be relevant in specific circumstances as standalone classes (e.g., Visualisation, Validation, and Funding acquisition). Our proposal can also be seen as an extension of other existing solutions. The STM classification provided an interesting and detailed taxonomy, largely focused on the manuscript writing stage (e.g., correcting the text, reference management, translations), and thus lacking other important stages of the research cycle (e.g., planning and the actual execution of research). Another existing proposal 22 appears to be sufficiently granular (it proposes the following AI usage classes: content extraction, validation, generation, analysis, reformatting, discovery, translation). However, it lacks a clear alignment with the research cycle structure, making it less compatible with other reporting frameworks (e.g., CRediT). Crucially, AIdIT is not limited to journal articles. Its structure makes it particularly suitable for higher education contexts, including bachelor’s and master’s theses, doctoral dissertations, coursework, and project-based learning. By requiring students to explicitly document how AI tools contributed to idea generation, literature searches, data analysis, or writing, AIdIT promotes reflective and responsible use rather than concealment or over-reliance. For supervisors and examiners, such structured disclosure provides clarity and consistency, reducing ambiguity around acceptable AI use and helping to distinguish learning outcomes from automated assistance. More broadly, integrating AIdIT-like statements into academic training could support the development of AI literacy, making explicit which aspects of research require human judgment, creativity, and ethical responsibility. We attempted to make implementing the AIDiT framework feasible through an R Shiny app that automates disclosure statement generation. The app's source code is available on GitHub: https://github.com/szymekdr/AI_reporting . Concluding remarks Our systematic review of AI reporting practices in ecology and evolutionary biology found that they remain fragmented, inconsistent, and poorly aligned with other open research practices, despite AI’s rapid expansion in scientific methodology. Urgent action is required to normalise the transparency of AI applications in research and properly embed them within existing open science practices. A reporting framework we propose promises to bridge this glaring gap. By combining a taxonomy of uses, explicit tool listing and versioning, compliance and supervision statements, and optional additional information into a single, machine-readable declaration, AIdIT offers a practical path toward normalising AI transparency. Implemented consistently-whether in journals, grant applications, or educational settings-it can transform AI use from an opaque background activity into a documented and interpretable component of scholarly work. In a rapidly evolving methodological landscape, such normalisation is essential. Without it, AI risks becoming both ubiquitous and invisible, undermining trust, reproducibility, and training. With it, AI can be integrated into science and education in a way that is transparent, responsible, and aligned with the core values of open research. Declarations Extended supplementary results More detailed results for the text mining analysis can be found in the paper’s GitHub repository under the “Extended Supplementary Material” link here: https://szymekdr.github.io/AI_reporting/ . Open Research Open data and code The final database of journal-derived AI reporting recommendations has been deposited in the GitHub repository https://github.com/szymekdr/AI_reporting. The archive also contains a Shiny app that implements the AIDiT reporting framework and the Electronic Supplementary Materials, including a detailed version of the text-mining analysis. Author contributions SMD: Conceptualisation. Data curation, Formal analysis, Funding Acquisition, Investigation, Methodology, Software, Validation, Visualisation, Writing – original draft, Writing – review & editing; AA: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review & editing; JR: Conceptualisation, Formal analysis, Investigation, Methodology, Validation, Visualisation, Writing – original draft, Writing – review & editing; MCP: Conceptualisation, Investigation, Validation, Writing – review & editing; AG: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review & editing; KJ: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review & editing; PP: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review & editing; KS: Investigation, Validation, Writing – review & editing; MG: Investigation, Validation, Writing – review & editing; WO: Conceptualisation, Investigation, Validation, Writing – review & editing; FB: Investigation, Validation, Writing – review & editing; NB: Investigation, Validation, Writing – review & editing; MZN: Conceptualisation, Investigation, Methodology, Validation, Writing – review & editing; SN: Methodology, Writing – review & editing; ML: Methodology, Writing – review & editing. The order of Author’s was determined using the Dragon Kill Points method, as described in Martinig et al. (2025) 51 . A relevant spreadsheet is included with other open data & code. Generative AI methods reporting (AIdIT) We used the following AI engines in the present study: ChatGPT (v. 5.2); Claude Sonnet (v. 4.5). Area(s) of generative AI usage: ChatGPT - finding new data sources, data annotation, data cross-checking; Claude Sonnet - refining existing code, text mining of collected data. The authors declare that they have verified and approved all content generated or modified by the AI tools used. The use of AI in this paper was in compliance with the ethical regulations of all funders and host institutions. Representative prompts used in AI content generation are available upon request. Acknowledgments We thank Mariusz Cichoń for valuable comments at the initial stages of this project. SMD and KJ were supported by a CHIST-ERA ORD grant “FAIRBiRDS” (funding through the Polish National Science Centre (NCN), grant no. UMO-2022/04/Y/NZ8/00184). PP was funded by a National Science Centre OPUS project (grant no. UMO-2020/39/B/NZ8/01274), AA was supported by the Polish National Agency for Academic Exchange (NAWA) under the Bekker NAWA programme (grant number BPN/BEK/2024/1/00075/U/00001), AG was founded by a Polish National Science Centre SONATA project (grant no. UMO-2024/55/D/NZ8/00597). SN was supported by a Canada Excellence Research Chair program (CERC-2022-00074). References Nabout JC, Parreira MR, Teresa FB, Carneiro FM, da Cunha HF, de Souza Ondei L, Caramori SS, Soares TN. Publish (in a group) or perish (alone): the trend from single- to multi-authorship in biological papers. Scientometrics. 2015;102:357–64. https://doi.org/10.1007/s11192-014-1385-5 . Borer ET, MacDougall AS, Stevens CJ, Sullivan LL, Wilfahrt PA, Seabloom EW. Writing a massively multi-authored paper: overcoming barriers to meaningful authorship for all. Methods Ecol Evol. 2023;14:1432–42. https://doi.org/10.1111/2041-210X.14096 . van Dis EAM, Bollen J, Zuidema W, van Rooij R, Bockting CL. ChatGPT: five priorities for research. Nature. 2023;614:224–6. https://doi.org/10.1038/d41586-023-00288-7 . Blau W, Cerf VG, Enriquez J, Francisco JS, Gasser U, Gray ML, Greaves M, Grosz BJ, Jamieson KH, Haug GH et al. (2024). Protecting scientific integrity in an age of generative AI. Proc. Natl. Acad. Sci. 121 , e2407886121. https://doi.org/10.1073/pnas.2407886121 Ding L, Lawson C, Shapira P. Rise of Generative Artificial Intelligence in Science. Scientometrics. 2025;130:5093–114. https://doi.org/10.1007/s11192-025-05413-z . Van Noorden R, Perkel JM. AI and science: what 1,600 researchers think. Nature. 2023;621:672–5. https://doi.org/10.1038/d41586-023-02980-0 . Kwon D. Is it OK for AI to write science papers? Nature survey shows researchers are split. Nature. 2025;641:574–8. https://doi.org/10.1038/d41586-025-01463-8 . Kasneci E, Sessler K, Küchemann S, Bannert M, Dementieva D, Fischer F, Gasser U, Groh G, Günnemann S, Hüllermeier E, et al. ChatGPT for good? On opportunities and challenges of large language models for education. Learn Individ Differ. 2023;103:102274. https://doi.org/10.1016/j.lindif.2023.102274 . Khalifa M, Albadawy M. Using artificial intelligence in academic writing and research: An essential productivity tool. Comput Methods Programs Biomed Update. 2024;5:100145. https://doi.org/10.1016/j.cmpbup.2024.100145 . Spivakovsky OV, Omelchuk SA, Kobets VV, Valko NV, Malchykova DS. Institutional Policies on Artificial Intelligence in University Learning, Teaching and Research. Inf Technol Learn Tools. 2023;97:181–202. https://doi.org/10.33407/itlt.v97i5.5395 . Kaliisa R, Baker RS, Wasson B, Prinsloo P. The Coming but Uneven Storm: How AI Regulation Will Impact AI and Learning Analytics Research in Different Countries. J Learn Anal. 2025;12:140–57. https://doi.org/10.18608/jla.2025.8495 . Gonsalves C. Addressing student non-compliance in AI use declarations: implications for academic integrity and assessment in higher education. Assess Eval High Educ. 2025;50:592–606. https://doi.org/10.1080/02602938.2024.2415654 . Mahmud S. (2024). Fostering Academic Integrity in the Age of Artificial Intelligence. In Academic integrity in the age of artificial intelligence (IGI Global Scientific Publishing), pp. 1–20. https://doi.org/10.4018/979-8-3693-0240-8.ch001 Kim JJH, Um RS, Lee JWY, Ajilore O. Generative AI can fabricate advanced scientific visualizations: ethical implications and strategic mitigation framework. AI Ethics. 2025;5:4481–93. https://doi.org/10.1007/s43681-024-00439-0 . Bellini V, Semeraro F, Montomoli J, Cascella M, Bignami E. Between human and AI: assessing the reliability of AI text detection tools. Curr Med Res Opin. 2024;40:353–8. https://doi.org/10.1080/03007995.2024.2310086 . Hartung J, Reuter S, Kulow VA, Fähling M, Spreckelsen C, Mrowka R. Experts fail to reliably detect AI-generated histological data. Sci Rep. 2024;14:28677. https://doi.org/10.1038/s41598-024-73913-8 . Ganjavi C, Eppler MB, Pekcan A, Biedermann B, Abreu A, Collins GS, Gill IS, Cacciamani GE. (2024). Publishers’ and journals’ instructions to authors on use of generative artificial intelligence in academic and scientific publishing: bibliometric analysis. BMJ 384 , e077192. https://doi.org/10.1136/bmj-2023-077192 Editorial policies | Nature. Portfolio https://www.nature.com/nature-portfolio/editorial-policies Generative AI. policies for journals www.elsevier.com. https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals AI Policy Taylor &, Francis. https://taylorandfrancis.com/our-policies/ai-policy/ Alam S, Constantini R, Harp G, Hoekstra R, Konda S, Lisankie K, McNeice K, Olivero T, Rose I, van Rossum J, et al. Recommendations for a Classification of AI Use in Academic Manuscript Preparation. Association of Scientific, Technical and Medical; 2025. Hyde A, Chodacki J, Shannon P. An Initial Scholarly AI Taxonomy. Front Matter. 2023. https://doi.org/10.54900/6p6re-xyj61 . Allen L, O’Connell A, Kiermer V. How can we ensure visibility and diversity in research contributions? How the Contributor Role Taxonomy (CRediT) is helping the shift from authorship to contributorship. Learn Publish. 2019;32:71–4. https://doi.org/10.1002/leap.1210 . Nakagawa S, Ivimey-Cook ER, Grainger MJ, O’Dea RE, Burke S, Drobniak SM, Gould E, Macartney EL, Martinig AR, Morrison K, et al. Method reporting with initials for transparency (MeRIT) promotes more granularity and accountability for author contributions. Nat Commun. 2023;14:1788. https://doi.org/10.1038/s41467-023-37039-1 . Barnes AJ, Zhang Y, Valenzuela A. AI and culture: Culturally dependent responses to AI systems. Curr Opin Psychol. 2024;58:101838. https://doi.org/10.1016/j.copsyc.2024.101838 . Hosseini M, Resnik DB, Holmes K. The ethics of disclosing the use of artificial intelligence tools in writing scholarly manuscripts. Res Ethics. 2023;19:449–65. https://doi.org/10.1177/17470161231180449 . Gerlich M. Perceptions and Acceptance of Artificial Intelligence: A Multi-Dimensional Study. Soc Sci. 2023;12:502. https://doi.org/10.3390/socsci12090502 . O’Dea RE, Parker TH, Chee YE, Culina A, Drobniak SM, Duncan DH, Fidler F, Gould E, Ihle M, Kelly CD, et al. Towards open, reliable, and transparent ecology and evolutionary biology. BMC Biol. 2021;19:68. https://doi.org/10.1186/s12915-021-01006-3 . Pottier P, Lagisz M, Burke S, Drobniak SM, Downing PA, Macartney EL, Martinig AR, Mizuno A, Morrison K, Pollo P et al. (2024). Title, abstract and keywords: a practical guide to maximize the visibility and impact of academic papers. Proc. R. Soc. B: Biol. Sci. 291 , 20241222. https://doi.org/10.1098/rspb.2024.1222 Benoit K, Watanabe K, Wang H, Nulty P, Obeng A, Müller S, Matsuo A. quanteda: an R package for the quantitative analysis of textual data. J Open Source Softw. 2018;3:774. https://doi.org/10.21105/joss.00774 . Feinerer I, Hornik K, Meyer D. (2008). Text mining infrastructure in R . J. Stat. Softw. 25 . https://doi.org/10.18637/jss.v025.i05 Silge J, Robinson D. J Open Source Softw. 2016;1:37. https://doi.org/10.21105/joss.00037 . tidytext: Text Mining and Analysis Using Tidy Data Principles in R. Bouchet-Valat M. (2023). SnowballC. Version 0.7.1. Grün B, Hornik K. (2011). topicmodels : an R package for fitting topic models. J. Stat. Softw. 40 . https://doi.org/10.18637/jss.v040.i13 Jockers ML. (2015). Syuzhet: Extract Sentiment and Plot Arcs from Text. Chalmers A, Klingler-Vidra R, van den Malou O. From diffusion to diffuse-ability: a text-as-data approach to explaining the global diffusion of corporate sustainability policy. Int Stud Q. 2024;68:sqad106. https://doi.org/10.1093/isq/sqad106 . Schiff DS. Framing contestation and public influence on policymakers: evidence from US artificial intelligence policy discourse. Policy Soc. 2024;43:255–88. https://doi.org/10.1093/polsoc/puae007 . Stokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for science. Nature. 2023;614:214–6. https://doi.org/10.1038/d41586-023-00340-6 . Porsdam Mann S, Vazirani AA, Aboy M, Earp BD, Minssen T, Cohen IG, Savulescu J. Guidelines for ethical use and acknowledgement of large language models in academic writing. Nat Mach Intell. 2024;6:1272–4. https://doi.org/10.1038/s42256-024-00922-7 . Haibe-Kains B, Adam GA, Hosny A, Khodakarami F, Massive Analysis Quality Control (MAQC) Society Board of Directors, Shraddha, Kusko T, Sansone R, Tong S-A, Wolfinger W et al. R.D.,. (2020). Transparency and reproducibility in artificial intelligence. Nature 586 , E14–E16. https://doi.org/10.1038/s41586-020-2766-y Skulmowski A, Engel-Hermann P. The ethics of erroneous AI-generated scientific figures. Ethics Inf Technol. 2025;27:31. https://doi.org/10.1007/s10676-025-09835-4 . Emsley R. ChatGPT: these are not hallucinations – they’re fabrications and falsifications. Schizophrenia. 2023;9:52. https://doi.org/10.1038/s41537-023-00379-4 . Májovský M, Černý M, Kasal M, Komarc M, Netuka D. Artificial intelligence can generate fraudulent but authentic-looking scientific medical articles: pandora’s box has been opened. J Med Internet Res. 2023;25:e46924. https://doi.org/10.2196/46924 . Daneshjou R, Smith MP, Sun MD, Rotemberg V, Zou J. Lack of transparency and potential bias in artificial intelligence data sets and algorithms: a scoping review. JAMA Dermatol. 2021;157:1362–9. https://doi.org/10.1001/jamadermatol.2021.3129 . Siddiq ML, Islam-Gomes A, Sekerak N, Santos JCS. (2025). Large language models for software engineering: a reproducibility crisis. Preprint at arXiv. https://doi.org/10.48550/arXiv.2512.00651 https://doi.org/10.48550/arXiv.2512.00651 Sozon M, Parnther C. Scientific misconduct in the era of artificial intelligence: insights from the global south. High Educ Q. 2026;80:e70097. https://doi.org/10.1111/hequ.70097 . Nie X, Tian Y, Liu M, Wu D, Guo Y. The impact of generative artificial intelligence on students’ higher order thinking: Evidence from a three-level meta-analysis. Educ Inf Technol. 2025;30:25359–90. https://doi.org/10.1007/s10639-025-13735-x . Vaccaro M, Almaatouq A, Malone T. When combinations of humans and AI are useful: A systematic review and meta-analysis. Nat Hum Behav. 2024;8:2293–303. https://doi.org/10.1038/s41562-024-02024-1 . Wang S, Sun Z, Chen Y. Effects of higher education institutes’ artificial intelligence capability on students’ self-efficacy, creativity and learning performance. Educ Inf Technol. 2023;28:4919–39. https://doi.org/10.1007/s10639-022-11338-4 . Wahle JP, Ruas T, Mohammad SM, Meuschke N, Gipp B. (2023). AI usage cards: responsibly reporting AI-generated content. In 2023 ACM/IEEE Joint Conference on Digital Libraries (JCDL) (IEEE), pp. 282–284. https://doi.org/10.1109/JCDL57899.2023.00060 Martinig AR, Burk SLP, Drobniak SM, Perry I, Morrison K, Petersohn M, Pottier P, Nakagawa S, Pollo P, Ricolfi L et al. (2025). Dragon kill points: applying a transparent working template to relieve authorship stress (preprint). Preprint at EcoEvoRxiv. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 23 Apr, 2026 Reviews received at journal 15 Apr, 2026 Reviewers agreed at journal 11 Apr, 2026 Reviews received at journal 09 Apr, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers invited by journal 03 Apr, 2026 Editor assigned by journal 20 Mar, 2026 Submission checks completed at journal 20 Mar, 2026 First submitted to journal 18 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9160721","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":617697922,"identity":"5a26d55a-5e4c-4723-a77d-f90eaf8d1110","order_by":0,"name":"Szymon M. Drobniak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEUlEQVRIiWNgGAWjYNCCAwyMDQw8DAwJDAlyIB6IzcBHrBZjuBY2orQANSU2gCh8WuT7z5hu+HCGQXZ7e++xBw9q0tI3HDzA+OBtG0MeLi0GB86Y3Zxxg8F4zplz6QYJx3JyNxw4wGw4t42hGKcWxh6z2zwfGBJnSOSYSSSwVYC0sEnztjEktuFyWDOP2e0/cC3/KtINDhxg/41PC8MxoBaGG1AtiW05CUAtbMz4tBicYSu72XNGwngGz7k0icS+NMOZBw42S845J4HTL/L9h7fd+HHMRnYGe+8xyR/fkuX5bhw++OFNmU0ePy6HMXAYAAkJJAGJgw0gMgGnDgb2B2gC/A1gCo+WUTAKRsEoGGEAAP2vYsugKHOBAAAAAElFTkSuQmCC","orcid":"","institution":"Jagiellonian University","correspondingAuthor":true,"prefix":"","firstName":"Szymon","middleName":"M.","lastName":"Drobniak","suffix":""},{"id":617697923,"identity":"67ac43d2-9684-4968-92c0-0ba89b51fb5b","order_by":1,"name":"Martyna Cendrowska-Pek","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Martyna","middleName":"","lastName":"Cendrowska-Pek","suffix":""},{"id":617697924,"identity":"b3c84246-409e-430b-9592-d75340a7c172","order_by":2,"name":"Agnieszka Gudowska","email":"","orcid":"","institution":"Polish Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Agnieszka","middleName":"","lastName":"Gudowska","suffix":""},{"id":617697925,"identity":"475f1b53-d027-4af2-91a6-a1381597c97c","order_by":3,"name":"Katarzyna Janas","email":"","orcid":"","institution":"Polish Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Katarzyna","middleName":"","lastName":"Janas","suffix":""},{"id":617697926,"identity":"869c1d84-1a3d-4b25-8a99-6889a26e013a","order_by":4,"name":"Paweł Podkowa","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Paweł","middleName":"","lastName":"Podkowa","suffix":""},{"id":617697927,"identity":"b00cd3b9-ed94-4444-8ab2-a5ed6d41a0db","order_by":5,"name":"Karolina Skorb","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Karolina","middleName":"","lastName":"Skorb","suffix":""},{"id":617697928,"identity":"1b68248b-5376-4d8d-b856-f4c4eb010bf6","order_by":6,"name":"Monika Gronowska","email":"","orcid":"","institution":"University of Wrocław","correspondingAuthor":false,"prefix":"","firstName":"Monika","middleName":"","lastName":"Gronowska","suffix":""},{"id":617697929,"identity":"a9736f0f-05e8-46c2-bf32-ba68bb56f2ac","order_by":7,"name":"Wioleta Oleś","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Wioleta","middleName":"","lastName":"Oleś","suffix":""},{"id":617697930,"identity":"763de80c-c312-464c-9143-34882a3e4984","order_by":8,"name":"Faina Bikmurzina","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Faina","middleName":"","lastName":"Bikmurzina","suffix":""},{"id":617697931,"identity":"ba9cc052-7df5-43fd-aef3-e193accaed57","order_by":9,"name":"Natalia Boroń","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Natalia","middleName":"","lastName":"Boroń","suffix":""},{"id":617697932,"identity":"c975be34-0a99-4a36-956c-7c42f3e0a86b","order_by":10,"name":"Magdalena Zagalska-Neubauer","email":"","orcid":"","institution":"University of Wrocław","correspondingAuthor":false,"prefix":"","firstName":"Magdalena","middleName":"","lastName":"Zagalska-Neubauer","suffix":""},{"id":617697933,"identity":"bd0baf7a-2b91-4b78-92b3-4edec2df1935","order_by":11,"name":"Shinichi Nakagawa","email":"","orcid":"","institution":"University of New South Wales","correspondingAuthor":false,"prefix":"","firstName":"Shinichi","middleName":"","lastName":"Nakagawa","suffix":""},{"id":617697934,"identity":"964d1f90-8d2c-4c38-bc7b-ed396f0bb745","order_by":12,"name":"Małgorzata Lagisz","email":"","orcid":"","institution":"University of New South Wales","correspondingAuthor":false,"prefix":"","firstName":"Małgorzata","middleName":"","lastName":"Lagisz","suffix":""},{"id":617697935,"identity":"5d25d5c6-41af-416f-b961-efb8972b9243","order_by":13,"name":"Aneta Arct","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Aneta","middleName":"","lastName":"Arct","suffix":""},{"id":617697936,"identity":"2b88adf4-623c-4155-be0c-774d59730497","order_by":14,"name":"Joanna Rutkowska","email":"","orcid":"","institution":"Jagiellonian University","correspondingAuthor":false,"prefix":"","firstName":"Joanna","middleName":"","lastName":"Rutkowska","suffix":""}],"badges":[],"createdAt":"2026-03-18 14:53:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9160721/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9160721/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106725015,"identity":"eb8d8114-33b4-4abb-a1b0-0c06087e598b","added_by":"auto","created_at":"2026-04-12 18:31:01","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":274815,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAI guidelines in the journals in the field of ecology and evolutionary biology\u003c/strong\u003e. A.\u003cstrong\u003e Presence:\u003c/strong\u003e Of the 230 analysed journals, nearly half (45% “No”) provide no indication as to whether the use of AI in publications is permitted. B. \u003cstrong\u003eAuthor contribution guidelines:\u003c/strong\u003e Among 230 journals, there is a positive overlap between the presence of AI guidelines and the presence of guidelines for reporting author contributions. C. \u003cstrong\u003ePublishers’ recommendations:\u003c/strong\u003e Of the five publishers with the largest number of journals covered by our survey, only Elsevier consistently provided guidelines on their journals' AI policies. Note that only journals from the 5 top publishers (in terms of portfolio size) are visualised here to avoid excessive cluttering. Abbreviations: T\u0026amp;F – Taylor \u0026amp; Francis; OUP – Oxford University Press.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/e879edaa9641d39c4a76c098.png"},{"id":106515562,"identity":"6ffdaf25-5c9b-4fcd-bb2a-1af1c374a2c0","added_by":"auto","created_at":"2026-04-09 11:52:58","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28909,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eUpset plot of locations of AI guidelines in the surveyed journals. \u003c/strong\u003eThe plot illustrates subsets of journals based on the AI policy locations (3 categories, total counts in red). Singular locations and intersections of 2 or 3 locations (depicted with black dots + lines for multiple locations) are summarised as counts above blue bars.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/34bdf7779e33dba43f22f007.png"},{"id":106725639,"identity":"06c1cb46-9d7a-4ca6-9bba-bd39bd73aaa3","added_by":"auto","created_at":"2026-04-12 18:33:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":87970,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCompliance of published papers with the guidelines of the journals. \u003c/strong\u003eA. Author’s contribution statements. The journals' guidelines on whether and how authors should declare their contributions to the paper are followed most of the time. In 260 papers (12%), there was a contribution statement, even though this was not expected. B. Disclosure of using AI. AI usage declarations are significantly more likely to be present in papers from journals with relevant guidelines than in those from journals without such a policy (7 vs. 4%, χ\u003csup\u003e2\u003c/sup\u003e = 10.1, \u003cem\u003edf\u003c/em\u003e = 1, \u003cem\u003ep\u003c/em\u003e = 0.001).\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/e0609fd29f3bfbd0bad16434.png"},{"id":106724537,"identity":"485b0308-1bdd-4dd1-81dc-7e41f01e832d","added_by":"auto","created_at":"2026-04-12 18:28:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":239475,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe most frequently identified bigrams in the extracted AI guidelines texts. \u003c/strong\u003eThe set is dominated by bigrams containing the keyword “AI”.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/039e3fb5c7507ece1488f250.png"},{"id":106724733,"identity":"9800ef08-c778-43c1-a249-509bb02c1593","added_by":"auto","created_at":"2026-04-12 18:29:24","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":108405,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpecificity of AI guidelines, based on the KWIC and TF-IDF analyses (see main text for details). \u003c/strong\u003eMost journals cluster in a region of lower specificity/mixed specificity of the language used, according to the lexical analysis.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/9bc08543f7d310f78f2379ea.png"},{"id":106515565,"identity":"6cae54ed-19c3-44f6-bca7-55275fffc46f","added_by":"auto","created_at":"2026-04-09 11:52:58","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":212569,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRelationship between AI guidelines specificity and overall lexical sentiment score. \u003c/strong\u003eThe biggest publishers (according to the number of journals included) are labelled. The purple line shows an OLS best-fit relationship.\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/1d0e9b40626dc4489149316b.png"},{"id":106724596,"identity":"d2c770f0-2bdd-49b9-8d18-835be2fdee14","added_by":"auto","created_at":"2026-04-12 18:28:50","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":236326,"visible":true,"origin":"","legend":"\u003cp\u003eThe workflow of AIDiT. The block diagram presents the main elements of the reporting standard and how they respond to the actual structure of AI use in a manuscript or other relevant output. Table 1 presents the taxonomy of usage categories key for the “How was it used?” stage.\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/64edf5f09952e0b97f97df93.png"},{"id":106727097,"identity":"a565e96d-1b8a-437c-8768-e60b3b8cfabf","added_by":"auto","created_at":"2026-04-12 18:38:08","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1722137,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9160721/v1/1e035c64-a531-406a-a159-78a160730f77.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A systematic map of generative AI guidelines and reporting in ecology and evolutionary biology: towards the framework of AI disclosure for Improved Transparency (AIdIT)","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe explosive rise of generative artificial intelligence (AI) in scientific workflows has ignited debates about the transparency of its use among researchers across disciplines. Generative AI is increasingly used not only in neutral applications (e.g., text editing, spellchecking, style improvement) but also, e.g., to generate analytical code, reframe mathematical models or screen/curate large datasets. One field that has seen a rapid proliferation of AI applications is ecology and evolutionary biology (EEB).\u003c/p\u003e \u003cp\u003eEEB integrates field observations, experimental studies, large comparative datasets, and advanced statistical modelling, all of which converge within a highly collaborative team\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. This interdisciplinary nature makes the field particularly well placed to benefit from emerging AI tools, but also especially vulnerable to opacity in how these tools are used and reported. Since the public release of accessible large language models (LLMs) in late 2022, generative AI has rapidly become a routine assistant across these tasks, supporting information generation (e.g., literature scoping, text drafting, generating code snippets), content refinement (e.g., text polishing, bibliography formatting, annotating data), and content comparisons (e.g., comparative literature reviews, summarising multiple data sources)\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. A \u003cem\u003eNature\u003c/em\u003e survey revealed substantial uptake of such tools among researchers, many of whom already use them to brainstorm ideas, assist in research design and refine writing\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. According to the survey, over 66% of researchers appreciated faster data processing with AI, 58% noted increased effectiveness in previously infeasible computations, and 55% admitted that AI can help save time and resources. On the other hand, 69% of respondents noted that AI can generate results without proper insight or understanding, and 53% expressed concerns about the potential dangers AI poses to scientific reproducibility\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. This latter concern resonates well with a more recent poll\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e according to which, 70% of surveyed researchers accept using AI to edit papers (roughly half of this group only if AI use is properly disclosed), but only around 15% thinks the same about using AI in translations, summarising papers or generating initial drafts (additional 20–30% would allow for each of these uses conditional of proper disclosure of AI methods employed). Clearly, generative AI now touches multiple stages of the scientific process, from study conception through analysis and interpretation to writing and dissemination, as well as teaching and student assessment.\u003c/p\u003e \u003cp\u003eThere are multiple benefits to the rapid expansion of the use of generative AI tools in research. They can lower language barriers, especially for non-native speakers, accelerate repetitive or algorithmic tasks, and potentially free up time for higher-order reasoning, conceptual synthesis and creative study design\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. In principle, such efficiency gains could strengthen, rather than weaken, the quality and inclusiveness of research in EEB. A question arises: do these benefits outweigh the potential costs associated with the unethical or opaque use of AI tools? Being relatively new, widely available and fast-evolving, AI algorithms still evade many attempts to codify their use in informal norms or formal policies\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. At the same time, being widely available and democratised, AI tools’ use often falls through the cracks of existing policies \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Without transparent and accountable use, the risks are substantial: LLMs may fabricate facts and citations, encode and amplify social and scientific biases, and produce fluent yet incorrect prose. Such issues are difficult to detect, particularly for non-experts\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. This can spread misinformation, undermine critical thinking skills and blur the boundaries between human and machine contributions.\u003c/p\u003e \u003cp\u003eThese risks are particularly acute when AI tools are used unsupervised or without disclosure, for example, in educational settings where developing independent reasoning, methodological skills, and a clear sense of authorship is central. At the moment, only exposing cases of AI misuse can provide any sense of how much it may be affecting the current scientific record. Notably, beyond text generation, AI-based image manipulation and analysis have already contributed to retractions and corrections, exposing verification gaps in current quality-control procedures\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. At the same time, peer-review workflows face confidentiality and accountability challenges when manuscripts or reviews are shared with external AI systems, and multiple independent evaluations demonstrate that currently available AI-detection tools remain inconsistent and unreliable\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn response to the growing use of generative AI in research, many publishers have begun issuing policies on its use in manuscript preparation and peer review. Most allow limited use for language editing or copy-editing, if this is acknowledged, with strict bans usually associated with AI-assisted peer review or uploading manuscript drafts to AI agents (i.e., uses not directly associated with authoring a paper)\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. However, policy development has been uneven: major publishers, such as Springer-Nature\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, Elsevier\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e and Taylor \u0026amp; Francis\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e have issued portfolio-wide guidance on acceptable uses of generative AI, but audits report limited consensus on their scope and enforcement\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. Community-level resources have so far lagged behind the needs: to date, only two comprehensive attempts to define a taxonomy of AI uses have been proposed\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Both are preliminary and do not go beyond the classification of uses to actual reporting guidelines. As a result, in practice, authors, editors, and reviewers often operate in a landscape of partial, fragmented, and sometimes conflicting expectations.\u003c/p\u003e \u003cp\u003eIn EEB, open data, reproducible workflows, and explicit authorship contribution statements (e.g., via CRediT\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e and MeRIT\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e) are increasingly becoming the norm. Yet if the use of AI tools remains largely invisible in published articles, readers cannot assess which elements of a study reflect direct human intellectual input, which were assisted by AI and which tasks (e.g., language editing versus data analysis) were delegated to automated tools. At the same time, inconsistent or vague guidance may discourage honest disclosure and exacerbate inequalities between authors who – due to educational background or local research culture – feel comfortable disclosing the use of AI, and those who don’t\u003csup\u003e25–27\u003c/sup\u003e. However, in EEB, it is still unclear (i) what proportion of journals in the field currently have explicit polices on generative AI, (ii) where such policies are located, (iii) whether journals that have formalised authorship statements are also more likely to address AI usage, and (iv) to what extent these journals’ policies translate into article-level disclosure in practice.\u003c/p\u003e \u003cp\u003eTo address these gaps, we need a field-specific assessment of AI-related journal policies, their visibility and their alignment with open science practices related to transparent authorship crediting. We surveyed the webpages of 230 EEB journals (including multi-disciplinary journals that routinely publish EEB studies) and documented the presence, location, and within-publisher consistency of AI policies. AI reporting statements and recommendations were also extracted and text-mined to identify consistent patterns of semantic skew, uniformity, and language specificity. For the same journals, we also examined the co-occurrence of AI guidance with author-contribution practices (including their compliance with CRediT). Finally, we reviewed a sample of recently published articles to see whether they follow these journal guidelines. We examined whether journals that require contribution disclosures or AI-use statements have higher reporting rates, and we also noted cases where authors voluntarily disclosed AI use even when it was not required. Although the issues raised by the use of generative AI are universal, we use EEB as a model case. Being an increasingly collaborative, diverse and fast-moving subfield of biological sciences, EEB has recently embraced many practices from the open science movement\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e (transparent crediting, growing use of registered reports and pre-registrations). A discipline-wide approach to generative AI would therefore be a natural next step towards greater transparency, making EEB a well-defined subfield that can serve as a proxy to biological sciences in general.\u003c/p\u003e \u003cp\u003eBased on our overview, we identify current problems and inconsistencies in AI reporting and propose a single, standardised framework to make reporting clear, consistent, and complete. Although our work focuses on EEB studies, we designed the framework to be broadly applicable, with the goal of encouraging better open-research practices for the use of generative AI in research and education.\u003c/p\u003e "},{"header":"Materials \u0026 methods","content":"\u003cp\u003eJournal selection and assessment of formal policies on author contributions and AI\u003c/p\u003e\u003cp\u003eIn line with recent transparency recommendations, below we employ the MeRIT contribution reporting guidelines\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e to describe our methods and contributions.\u003c/p\u003e\u003cp\u003eA list of journals in the fields of \u003cem\u003eEcology\u003c/em\u003e and \u003cem\u003eEvolutionary Biology\u003c/em\u003e was compiled by SMD, following Pottier et al. (2024)\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. The list included all titles classified under these two subject categories in the Clarivate Journal Citation Reports. To ensure coverage of multidisciplinary venues that frequently publish relevant research, we supplemented this list with 13 additional high-profile journals, including Nature, Nature Communications, Nature Climate Change, Scientific Reports, Science, Science Advances, Communications Biology, Proceedings of the National Academy of Sciences, PLoS Biology, Biological Reviews, Current Biology, eLife, and Philosophical Transactions of the Royal Society B.\u003c/p\u003e\u003cp\u003eIn April 2025, we screened (SMD, JR, MCP, AG, KJ, PP, KS, MG, WO, FB, NB, AA) all 230 journals to identify formal policies regarding (i) reporting of author contributions and (ii) the use of AI. For each journal, we first searched the “Instructions for Authors” (or equivalent) on the journal’s website. If no mention of author contributions was found, we used keywords such as “contribution”, “CRediT”, and “statement”. If no mention of AI was found, we searched the site using keywords including “artificial”, “intelligence”, “AI”, “generative”, and “LLM”.\u003c/p\u003e\u003cp\u003eFor each journal, we used a standardised extraction template to record:\u003c/p\u003e\u003cp\u003e \u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eWebsite address (link to the information for authors);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePresence of guidelines on reporting author contributions (Yes, mandatory / Yes, suggestions / No);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePresence of guidelines on AI use (Yes / No); if Yes:\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eAre any types of AI use explicitly permitted or prohibited? (Yes / No);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAre there guidelines on where or how AI use should be declared? (Yes / No);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIs there a suggested format for declaring AI use? (Yes / No);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eVerbatim copy of any relevant guidelines.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003cp\u003e\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFollowing initial extraction, a second extractor (from the same pool of extractors) rechecked all records without guidelines to confirm that no relevant information had been overlooked. We also verified whether any AI-related policies were located in sections of the journal website other than the “Instructions for Authors” (e.g., Editorial Policies, Publishing Ethics) and recorded their locations and links when present.\u003c/p\u003e\u003cp\u003eArticle screening and data extraction\u003c/p\u003e\u003cp\u003eFor article-level screening, we searched (SMD, JR, MCP, AG, KJ, PP, KS, MG, WO, FB, NB, AA, MZN) the Web of Science (Core Collection) using the following standardised advanced search string:\u003c/p\u003e\u003cp\u003e(((SO=(JOURNAL_TITLE)) AND DT=(Article)) NOT DT=(Proceedings Paper OR Publication with Expression of Concern OR Biographical-Item OR Excerpt OR Book Chapter OR Record Review OR Note OR Book Review OR Correction OR Correction, Addition OR Database Review OR Data Paper OR Editorial Material OR Software Review OR Withdrawn Publication OR Retracted Publication))\u003c/p\u003e\u003cp\u003ewhere JOURNAL_TITLE was iteratively replaced with each journal name. Search results were sorted by date (newest first). From each journal, we exported metadata for the 20–30 most recent articles (retaining all bibliographic fields) and used these as a screening pool. From this pool, we selected the first 10 accessible, full-text research articles for detailed data extraction.\u003c/p\u003e\u003cp\u003eFor each article, we recorded the following information using a standardised extraction template:\u003c/p\u003e\u003cp\u003e \u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eContribution present (Yes / No): whether a statement detailing the contributions of all co-authors was included;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eContribution CRediT (Yes / No): whether contributions were reported using the exact version of the CRediT taxonomy;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAI statement present (Yes / No): whether any statement on the use of AI tools was included;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAI statement location: article section where the statement appeared (Title page / Methods / Acknowledgements / Ethical note / Separate section / Other);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAI model (Yes / No): whether the statement specified the AI model used (e.g. ChatGPT, Llama, Gemini, Copilot);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAI usage (Yes / No): whether the statement described how AI was used (e.g. writing assistance, language editing, data processing);\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eNegative statement (Yes / No): whether the article explicitly stated that no AI tools were used.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIf an AI statement was present, its text was copied verbatim. To ensure that declarations were not missed due to version differences, at least one randomly chosen article without an AI statement from each journal was cross-checked between its online and PDF versions. Articles that could not be accessed were excluded, and the reason for exclusion was recorded.\u003c/p\u003e\u003cp\u003eOur initial assessment was strictly limited to the author guidelines/instructions, i.e., a journal section that provides formatting, content, and related information for authors considering submitting a manuscript. After completing our survey, we realised that some journals provided additional/sole AI-related policies in locations other than the author guidelines. Thus, we repeated the search to identify other locations where AI usage guidelines might be available (e.g., on publisher websites, in separate ethics/editorial policy sections). These, if present, were recorded under “Other locations”.\u003c/p\u003e\u003cp\u003eText mining of journal AI guidelines\u003c/p\u003e\u003cp\u003eAll journal-level AI guidelines identified during policy screening were subjected to a structured text-mining analysis (done by SMD) to characterise their content, consistency, and semantic patterns. Analyses were conducted on the verbatim guideline texts extracted from journal websites and stored in a centralised database. Only records containing substantive guideline text (non-empty entries exceeding 20 characters) were retained, yielding 124 documents out of 230 journals.\u003c/p\u003e\u003cp\u003ePrior to analysis, publisher names were standardised to resolve inconsistencies across journal websites. Publisher identity was cross-validated using both declared publisher information and journal website domains, with discrepancies resolved using automated normalisation rules followed by manual verification where necessary. Basic document statistics (character count and word count) were recorded to assess heterogeneity in guideline length.\u003c/p\u003e\u003cp\u003eText preprocessing and analyses were conducted in R using the \u003cem\u003etidytext\u003c/em\u003e, \u003cem\u003etm\u003c/em\u003e, and \u003cem\u003equanteda\u003c/em\u003e packages\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Guideline texts were converted to lowercase, tokenised into unigrams and bigrams, and stripped of punctuation, numbers, and English stopwords. Tokens shorter than three characters were excluded, and stemming was applied to unigrams using the Snowball algorithm (\u003cem\u003eSnowballC\u003c/em\u003e package\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e). Token frequencies were calculated to identify the most common terms and phrases across guidelines.\u003c/p\u003e\u003cp\u003eTo identify vocabulary distinctive to individual journals or publishers, we computed term frequency–inverse document frequency (TF–IDF) scores at the document level\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Keywords-in-context (KWIC) analyses were performed for selected AI-related terms (e.g., \u003cem\u003eAI\u003c/em\u003e, \u003cem\u003eartificial intelligence\u003c/em\u003e, \u003cem\u003elanguage model\u003c/em\u003e, \u003cem\u003eChatGPT\u003c/em\u003e) to examine how these concepts were framed, particularly with respect to permissions, prohibitions, and disclosure requirements.\u003c/p\u003e\u003cp\u003eHigher-level semantic structure was explored using unsupervised topic modelling. Latent Dirichlet Allocation (LDA) models were fitted to document–term matrices using the \u003cem\u003etopicmodels\u003c/em\u003e package\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e, with the number of topics chosen based on interpretability. Word co-occurrence networks were constructed to visualise associations among frequently co-occurring concepts, and sentiment analysis was applied using a lexicon-based approach (\u003cem\u003esyuzhet\u003c/em\u003e\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e) to provide a coarse characterization of guideline framing.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eSystematic mapping\u003c/p\u003e \u003cp\u003eOf the 230 journals in the field of ecology and evolution, 104 (45%) had no guidelines, not even a mention, regarding the potential use of artificial intelligence in the articles they publish (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA). Journals that mentioned the use of AI usually included a statement on permitted use (119 out of 126) and on how it should be disclosed (123 out of 126). We also found that most journals with any mention of artificial intelligence policies had guidelines for author contribution statements (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB): AI disclosure requirements were present in 74% of journals with mandatory authorship statements and 68% of journals with suggested authorship statements. Interestingly, even journals belonging to the same publisher differed in this respect. For instance, some Wiley and Springer journals provide AI guidance, while others do not mention it at all (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC). Among the major publishers, Elsevier appeared to be the only one with a consistent AI policy across all its journals. Where information was provided, it was usually found on the journal's website, the publisher's website, or another external site (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAn overview of how the published papers comply with the journal's guidelines on author contribution statements shows that authors readily follow these guidelines and also include such statements even when they are not mandatory (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Overall, author statements were found in more than 67% of papers. Among them, 44% used the exact version of the CRediT standard.\u003c/p\u003e \u003cp\u003eThe situation is very different when it comes to declaring AI use. We found that approximately 94% of papers do not disclose the use of AI (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Unfortunately, assessing the accuracy of the non-disclosure cases (whether they represent genuine absences of AI involvement vs. false negatives) remains impossible under the current field-wide heterogeneity of AI-reporting recommendations.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eText mining\u003c/p\u003e \u003cp\u003eText mining was conducted on 124 journal-level AI guideline documents, which varied widely in length and detail (median 252 words; range 149\u0026ndash;11,766 words), indicating substantial heterogeneity in how journals address AI use (Extended Supplementary Results (ESM), Fig. S1). Most guidelines were short and embedded within broader publishing or ethics policies rather than presented as stand-alone AI documents.\u003c/p\u003e \u003cp\u003eAcross the full corpus, unigram and bigram frequency analyses revealed a highly standardised vocabulary dominated by generic policy language. The most frequent terms and stems were associated with authors, responsibility, content, manuscripts, tools, and use, with AI-specific terms (e.g. artificial intelligence, AI, language model) appearing far less frequently and often embedded within otherwise boilerplate text. Bigrams such as artificial intelligence, generative AI, and AI tools were common but typically appeared in formulaic statements rather than in detailed procedural guidance (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTerm Frequency-Inverse Document Frequency (TF\u0026ndash;IDF) analyses showed that most journals exhibited very low lexical distinctiveness, reflecting strong within-publisher reuse of near-identical guideline text. Distinctive terms were primarily associated with publisher-level policies rather than individual journals, indicating that AI guidance is largely standardised at the publisher level and rarely tailored to specific journal scopes or disciplinary practices (for details, please see the Text Mining section of the Electronic Supplementary Materials).\u003c/p\u003e \u003cp\u003eEven though the overall sentiment score indicated a positive lexical incline in the tone of most guideline texts (ESM, Fig. S4), most extracted policies sounded close to neutral. Still, the keywords-in-context (KWIC) analyses demonstrated that references to AI were most often framed in restrictive or cautionary contexts, emphasising the authors' responsibility, accountability, and prohibition of AI tools (see ESM). Consistently, mentions of permissible AI use (e.g. for language editing or stylistic improvement) were typically brief and weakly specified, while explicit guidance on acceptable analytical or data-processing uses was rare (see Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e for the overall distribution of recommendations\u0026rsquo; specificity). Statements explicitly requiring disclosure of AI use were uncommon and often vague about where or how such disclosure should be made.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTopic modelling identified a small number of dominant thematic clusters, primarily centred on (i) authorship and responsibility, (ii) ethical conduct and integrity, and (iii) permitted versus prohibited uses of automated tools (ESM, Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). Topics directly addressing transparency, reporting standards, or disclosure formats formed a minor component of the overall semantic structure, indicating that these issues are not central in most current guidelines.\u003c/p\u003e \u003cp\u003eWord co-occurrence networks further highlighted the close association between AI-related terms and concepts such as responsibility, accountability, and integrity, whereas links to methods, analysis, or data were weak or absent (ESM, Fig. S5). Sentiment analysis supported this pattern, with guidelines predominantly exhibiting neutral-to-cautionary framing rather than permissive or enabling language. In line with prior evidence from similar text-mining analyses\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e, the most specific recommendation texts were also the most negative and restrictive (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Notably, the largest players on the market (e.g., Springer Nature, Wiley, Elsevier) cluster in the region of average specificity and average sentiment score. There is one interesting outlier: MDPI shows the sentiment score close to zero (neutral wording) with an extremely high specificity index. This may be due to the exceedingly high density of AI-related words in MDPI guidelines (ESM, Fig. S17), which stems from the fact that the only included guidelines texts for MDPI were very short (\u0026lt;\u0026thinsp;50 words). Nonetheless, MDPI was represented by only 2 journals, largely precluding any generalisations for this publisher.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur systematic mapping of generative AI policies in ecology and evolutionary biology (EEB) highlights a growing structural risk: AI tools are rapidly becoming embedded across research workflows, while mechanisms for documenting their use remain sparse, fragmented, and weakly formalised as verifiable policies. Nearly half of the surveyed journals provide no guidance on AI use, and where guidance exists, it is typically generic, publisher-driven, and poorly translated into reporting practices. Moreover, negative AI usage statements (i.e., intentional indication of no generative AI use) are virtually non-existent (\u0026lt; 1% of the analysed sample of articles). Comparing the results of our survey with the reported incidence of AI use in recent survey\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e suggests considerable opacity in the disclosure of LLM use. As a result, AI use remains largely invisible in the published record, despite mounting evidence that it already affects how scientific knowledge is generated, analysed, and communicated\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe absence of clear AI disclosure standards has tangible consequences. Without knowing whether and how AI tools contributed to data processing, code generation, text mining, or interpretation, readers cannot fully assess the provenance of results\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. This opacity complicates research reproducibility, hinders methodological evaluation, and limits reviewers' and meta-researchers' ability to detect systematic biases or error propagation\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. Recent high-profile cases of retractions and corrections linked to AI-generated images\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e, fabricated citations\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e42\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, or unverifiable analytical steps\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e44\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e illustrate that these risks are no longer hypothetical but already materialising in the scientific literature.\u003c/p\u003e \u003cp\u003eImportantly, these problems are not caused by AI use alone, but also by the lack of a systemic AI disclosure practice that could help flag such problematic studies during peer review. When AI-assisted steps are not transparently reported, errors introduced upstream - whether through hallucinated references, inappropriate model assumptions, or subtle data transformations - become difficult or impossible to trace. Over time, this can erode trust in published findings, particularly in fields such as EEB that inform conservation policy, environmental management, and public decision-making. The situation can be seen as mirroring a similar erosion of trust that eventually became one of the driving forces behind the open data trend, now, for many publishers, hardwired into scientific publishing. It remains to be seen whether the current disconnected AI reporting policies will follow a similar trajectory and catalyse a field‑wide movement toward explicit, unified disclosure practices.\u003c/p\u003e \u003cp\u003ePrecautionary policies without operational guidance\u003c/p\u003e \u003cp\u003eOur text-mining analyses show that existing AI guidelines are relatively uniform. At the same time, they are dominated by precautionary language emphasising responsibility, integrity, and prohibitions, most notably the exclusion of AI systems from authorship, with limited specificity regarding acceptable uses, reporting requirements, or integration into existing open science and transparency practices.\u003c/p\u003e \u003cp\u003eWhile these policies serve an important symbolic function, they provide little operational guidance for authors navigating real research workflows. Mentions of AI use in methods development, data analysis, modelling, or visualisation are rare, despite the growing prevalence of AI-assisted tools in precisely these domains\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Such narrow framing reflects a broader pattern in which AI is treated primarily as an ethical risk rather than as a methodological variable\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Such an approach may help publishers manage liability, but it fails to support transparent science. Moreover, it risks becoming rapidly outdated as AI tools diversify and become increasingly integrated into standard analytical environments, often as implicit, seamless components of software suites used in research and writing.\u003c/p\u003e \u003cp\u003eThe contrast between widespread adoption of author contribution statements and the rarity of AI disclosures is instructive. Contribution statements are normalised and standardised and are often embedded in submission systems. In contrast, AI disclosures are optional, inconsistently worded, and rarely linked to specific sections of a manuscript. This difference suggests that, consistent with broader surveys\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, low disclosure rates are not primarily driven by authors’ unwillingness or bad faith. Instead, they reflect the absence of clear expectations, standardised formats, and workflow integration. As noted below, adding another vaguely specified declaration to an already complex submission process is unlikely to improve transparency and may even discourage disclosure. Social dynamics may further suppress reporting. Particularly for students and early-career researchers, admitting to AI use may carry a perceived stigma or fear of being judged as less competent or less original\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. In the absence of clear norms, silence can appear safer than transparency.\u003c/p\u003e \u003cp\u003eThe consequences of poor AI reporting extend beyond journals. In higher education, AI tools are increasingly used in theses, coursework, and student-led research, often without clear guidance on acceptable use or reporting standards\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. This creates ambiguity for students, supervisors, and examiners alike. Without structured disclosure, it becomes difficult to distinguish between legitimate assistance, inappropriate delegation, and outright misconduct.\u003c/p\u003e \u003cp\u003eComplicating these considerations, there is currently little empirical understanding of how sustained use of AI affects the development of core cognitive skills such as critical thinking, synthesis, and creative problem-solving. The existing evidence is, at best, fragmentary and heterogeneous\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e47\u003c/span\u003e–\u003cspan class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e. In this context, transparent reporting is not merely a matter of compliance, but a pedagogical tool that encourages reflection on how AI is used, where human judgment remains essential, and where over-reliance may be harmful.\u003c/p\u003e \u003cp\u003eStudy limitations\u003c/p\u003e \u003cp\u003eOur study has several limitations. It focuses on journals within ecology and evolutionary biology and may not capture governance patterns in other disciplines, some of which may be more advanced—or substantially less developed—in their approaches to AI disclosure. Article‑level reporting was assessed using a recent snapshot of publications, which provides a useful baseline but cannot reveal temporal trends, journal‑specific transitions, or the effects of newly introduced policies. Likewise, text‑mining approaches necessarily abstract away nuance in individual guidelines, editorial decisions, and informal practices that may shape AI reporting in ways not captured by our coding framework. Finally, the absence of AI disclosure does not imply actual AI use, and interpreting non‑disclosure remains inherently ambiguous: it may reflect genuine non‑use, a lack of awareness, explicit journal exemptions (e.g., allowing unreported use for language polishing), or insufficiently integrated submission workflows.\u003c/p\u003e \u003cp\u003eTaken together, these limitations highlight the need for longitudinal, cross‑disciplinary, and mixed‑methods research on AI governance - work that can track policy evolution, incorporate qualitative insight from editors and authors, and evaluate whether emerging standards genuinely improve transparency and research integrity.\u003c/p\u003e \u003cp\u003eThe AIdIT framework: standardizing AI disclosure across research and education\u003c/p\u003e \u003cp\u003eTo address the gaps identified by our map and text mining, we propose a new framework: \u003cem\u003eAIdIT\u003c/em\u003e (\u003cb\u003eAI d\u003c/b\u003eisclosure for \u003cb\u003eI\u003c/b\u003emproved \u003cb\u003eT\u003c/b\u003eransparency; see Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e and Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e for a schematic structure). It can serve as a standardised, non-punitive norm for reporting AI use across the research lifecycle. AIdIT is designed not to restrict innovation, but to make AI-assisted work interpretable, auditable, and comparable, much like author contribution statements or data availability declarations. The framework is grounded in a taxonomy of AI uses and research stages (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) mimicking the CRediT authorship roles classification. As such, it recognises that AI can contribute in qualitatively different ways across conceptualisation, literature review, data curation, formal analysis, methods development, visualisation, and writing. Rather than asking whether AI was used, the framework asks \u003cem\u003ehow\u003c/em\u003e it was used (e.g. generation, refinement, comparison) and \u003cem\u003ewhere\u003c/em\u003e it influenced the research process.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\"\u003e\u003c/div\u003e\u003ctable id=\"Tab1\" border=\"1\"\u003e \u003ccaption\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTaxonomy of AI uses, categorised into three areas of application. See also Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e for practical implementation of categories.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003c/colgroup\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" rowspan=\"2\"\u003e \u003cp\u003eResearch cycle stage\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\"\u003e \u003cp\u003eArea of AI use\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\"\u003e \u003cp\u003eContent generation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\"\u003e \u003cp\u003eContent refinement\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\"\u003e \u003cp\u003eContent comparisons\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eConceptualisation\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Idea generation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Idea evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Gap identification\u003c/p\u003e \u003cp\u003e- Literature review\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eData curation\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Finding new data sources\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Data cleaning\u003c/p\u003e \u003cp\u003e- Data annotation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Building relations between datasets\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eFormal analysis\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- New code generation\u003c/p\u003e \u003cp\u003e- Statistical model formulation\u003c/p\u003e \u003cp\u003e- Mathematical calculations/modelling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Refining existing code\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Comparing coding approaches\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eFunding acquisition\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Generating proposal sections\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Grant text editing and corrections\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Benchmarking against funding body guidelines\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eInvestigation\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- New data generation\u003c/p\u003e \u003cp\u003e- Assets generation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Augmentation of existing data\u003c/p\u003e \u003cp\u003e- Modification of existing assets\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Text mining of collected data\u003c/p\u003e \u003cp\u003e- Summarising multiple data sources\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eMethods\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Identifying appropriate methods\u003c/p\u003e \u003cp\u003e- Protocol generation/review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Methods/protocols refinement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Comparing multiple approaches\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eValidation\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Identifying validation approaches\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Data cross-checking\u003c/p\u003e \u003cp\u003e- Error management\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Comparing results of validations\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e\u003cem\u003eVisualisation\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Code generation for visuals\u003c/p\u003e \u003cp\u003e- Generative images\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Refinement of visual aesthetics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Evaluation against guidelines and practice standards\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\"\u003e \u003cp\u003eWriting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Identification of literature to cite\u003c/p\u003e \u003cp\u003e- Text generation based on prompts\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Bibliography management\u003c/p\u003e \u003cp\u003e- Text editing for style/grammar/spelling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\"\u003e \u003cp\u003e- Identifying contexts for own writing\u003c/p\u003e \u003cp\u003e- Evaluation against guidelines and policies\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003eA central feature of AIdIT is its emphasis on human oversight and verification. When disclosing the use of generative AI, authors are prompted to describe not only the role of AI, but also how outputs were checked, validated, or approved. This shifts the focus away from tool prohibition toward accountability and quality control, aligning AI disclosure with established scientific norms.\u003c/p\u003e \u003cp\u003eThe taxonomy of AI uses proposed above was inspired by the AI usage cards designed by the computer science community\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. Building on the initial ideas, we made the usage categories more general, user-friendly, and generalisable across a wider array of disciplines. Our taxonomy is more uniform in dividing the research process into clear, well-defined stages; it also includes areas of AI use that may be relevant in specific circumstances as standalone classes (e.g., Visualisation, Validation, and Funding acquisition). Our proposal can also be seen as an extension of other existing solutions. The STM classification provided an interesting and detailed taxonomy, largely focused on the manuscript writing stage (e.g., correcting the text, reference management, translations), and thus lacking other important stages of the research cycle (e.g., planning and the actual execution of research). Another existing proposal\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e appears to be sufficiently granular (it proposes the following AI usage classes: content extraction, validation, generation, analysis, reformatting, discovery, translation). However, it lacks a clear alignment with the research cycle structure, making it less compatible with other reporting frameworks (e.g., CRediT).\u003c/p\u003e \u003cp\u003eCrucially, AIdIT is not limited to journal articles. Its structure makes it particularly suitable for higher education contexts, including bachelor’s and master’s theses, doctoral dissertations, coursework, and project-based learning. By requiring students to explicitly document how AI tools contributed to idea generation, literature searches, data analysis, or writing, AIdIT promotes reflective and responsible use rather than concealment or over-reliance. For supervisors and examiners, such structured disclosure provides clarity and consistency, reducing ambiguity around acceptable AI use and helping to distinguish learning outcomes from automated assistance. More broadly, integrating AIdIT-like statements into academic training could support the development of AI literacy, making explicit which aspects of research require human judgment, creativity, and ethical responsibility. We attempted to make implementing the AIDiT framework feasible through an R Shiny app that automates disclosure statement generation. The app's source code is available on GitHub: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/szymekdr/AI_reporting\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e "},{"header":"Concluding remarks","content":"\u003cp\u003eOur systematic review of AI reporting practices in ecology and evolutionary biology found that they remain fragmented, inconsistent, and poorly aligned with other open research practices, despite AI’s rapid expansion in scientific methodology. Urgent action is required to normalise the transparency of AI applications in research and properly embed them within existing open science practices. A reporting framework we propose promises to bridge this glaring gap.\u003c/p\u003e\u003cp\u003eBy combining a taxonomy of uses, explicit tool listing and versioning, compliance and supervision statements, and optional additional information into a single, machine-readable declaration, AIdIT offers a practical path toward normalising AI transparency. Implemented consistently-whether in journals, grant applications, or educational settings-it can transform AI use from an opaque background activity into a documented and interpretable component of scholarly work. In a rapidly evolving methodological landscape, such normalisation is essential. Without it, AI risks becoming both ubiquitous and invisible, undermining trust, reproducibility, and training. With it, AI can be integrated into science and education in a way that is transparent, responsible, and aligned with the core values of open research.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eExtended supplementary results\u003c/h2\u003e\n\u003cp\u003eMore detailed results for the text mining analysis can be found in the paper’s GitHub repository under the “Extended Supplementary Material” link here: https://szymekdr.github.io/AI_reporting/ .\u003c/p\u003e\n\u003ch2\u003eOpen Research\u003c/h2\u003e\n\u003ch2\u003eOpen data and code\u003c/h2\u003e\n\u003cp\u003eThe final database of journal-derived AI reporting recommendations has been deposited in the GitHub repository https://github.com/szymekdr/AI_reporting. The archive also contains a Shiny app that implements the AIDiT reporting framework and the Electronic Supplementary Materials, including a detailed version of the text-mining analysis.\u003c/p\u003e\n\u003ch2\u003eAuthor contributions\u003c/h2\u003e\n\u003cp\u003eSMD: Conceptualisation. Data curation, Formal analysis, Funding Acquisition, Investigation, Methodology, Software, Validation, Visualisation, Writing – original draft, Writing – review \u0026amp; editing; AA: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review \u0026amp; editing; JR: Conceptualisation, Formal analysis, Investigation, Methodology, Validation, Visualisation, Writing – original draft, Writing – review \u0026amp; editing; MCP: Conceptualisation, Investigation, Validation, Writing – review \u0026amp; editing; AG: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review \u0026amp; editing; KJ: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review \u0026amp; editing; PP: Conceptualisation, Investigation, Methodology, Validation, Writing – original draft, Writing – review \u0026amp; editing; KS: Investigation, Validation, Writing – review \u0026amp; editing; MG: Investigation, Validation, Writing – review \u0026amp; editing; WO: Conceptualisation, Investigation, Validation, Writing – review \u0026amp; editing; FB: Investigation, Validation, Writing – review \u0026amp; editing; NB: Investigation, Validation, Writing – review \u0026amp; editing; MZN: Conceptualisation, Investigation, Methodology, Validation, Writing – review \u0026amp; editing; SN: Methodology, Writing – review \u0026amp; editing; ML: Methodology, Writing – review \u0026amp; editing. The order of Author’s was determined using the Dragon Kill Points method, as described in Martinig et al. (2025)\u003csup\u003e51\u003c/sup\u003e. A relevant spreadsheet is included with other open data \u0026amp; code.\u003c/p\u003e\n\u003ch2\u003eGenerative AI methods reporting (AIdIT)\u003c/h2\u003e\n\u003cp\u003eWe used the following AI engines in the present study: ChatGPT (v. 5.2); Claude Sonnet (v. 4.5). Area(s) of generative AI usage: ChatGPT - finding new data sources, data annotation, data cross-checking; Claude Sonnet - refining existing code, text mining of collected data. The authors declare that they have verified and approved all content generated or modified by the AI tools used. The use of AI in this paper was in compliance with the ethical regulations of all funders and host institutions. Representative prompts used in AI content generation are available upon request.\u003c/p\u003e\n\u003ch2\u003eAcknowledgments\u003c/h2\u003e\n\u003cp\u003eWe thank Mariusz Cichoń for valuable comments at the initial stages of this project. SMD and KJ were supported by a CHIST-ERA ORD grant “FAIRBiRDS” (funding through the Polish National Science Centre (NCN), grant no. UMO-2022/04/Y/NZ8/00184). PP was funded by a National Science Centre OPUS project (grant no. UMO-2020/39/B/NZ8/01274), AA was supported by the Polish National Agency for Academic Exchange (NAWA) under the Bekker NAWA programme (grant number BPN/BEK/2024/1/00075/U/00001), AG was founded by a Polish National Science Centre SONATA project (grant no. UMO-2024/55/D/NZ8/00597). SN was supported by a Canada Excellence Research Chair program (CERC-2022-00074).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eNabout JC, Parreira MR, Teresa FB, Carneiro FM, da Cunha HF, de Souza Ondei L, Caramori SS, Soares TN. Publish (in a group) or perish (alone): the trend from single- to multi-authorship in biological papers. Scientometrics. 2015;102:357\u0026ndash;64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11192-014-1385-5\u003c/span\u003e\u003cspan address=\"10.1007/s11192-014-1385-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBorer ET, MacDougall AS, Stevens CJ, Sullivan LL, Wilfahrt PA, Seabloom EW. Writing a massively multi-authored paper: overcoming barriers to meaningful authorship for all. Methods Ecol Evol. 2023;14:1432\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/2041-210X.14096\u003c/span\u003e\u003cspan address=\"10.1111/2041-210X.14096\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Dis EAM, Bollen J, Zuidema W, van Rooij R, Bockting CL. ChatGPT: five priorities for research. Nature. 2023;614:224\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-023-00288-7\u003c/span\u003e\u003cspan address=\"10.1038/d41586-023-00288-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlau W, Cerf VG, Enriquez J, Francisco JS, Gasser U, Gray ML, Greaves M, Grosz BJ, Jamieson KH, Haug GH et al. (2024). Protecting scientific integrity in an age of generative AI. Proc. Natl. Acad. Sci. \u003cem\u003e121\u003c/em\u003e, e2407886121. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1073/pnas.2407886121\u003c/span\u003e\u003cspan address=\"10.1073/pnas.2407886121\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDing L, Lawson C, Shapira P. Rise of Generative Artificial Intelligence in Science. Scientometrics. 2025;130:5093\u0026ndash;114. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11192-025-05413-z\u003c/span\u003e\u003cspan address=\"10.1007/s11192-025-05413-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Noorden R, Perkel JM. AI and science: what 1,600 researchers think. Nature. 2023;621:672\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-023-02980-0\u003c/span\u003e\u003cspan address=\"10.1038/d41586-023-02980-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwon D. Is it OK for AI to write science papers? Nature survey shows researchers are split. Nature. 2025;641:574\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-025-01463-8\u003c/span\u003e\u003cspan address=\"10.1038/d41586-025-01463-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKasneci E, Sessler K, K\u0026uuml;chemann S, Bannert M, Dementieva D, Fischer F, Gasser U, Groh G, G\u0026uuml;nnemann S, H\u0026uuml;llermeier E, et al. ChatGPT for good? On opportunities and challenges of large language models for education. Learn Individ Differ. 2023;103:102274. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.lindif.2023.102274\u003c/span\u003e\u003cspan address=\"10.1016/j.lindif.2023.102274\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhalifa M, Albadawy M. Using artificial intelligence in academic writing and research: An essential productivity tool. Comput Methods Programs Biomed Update. 2024;5:100145. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cmpbup.2024.100145\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpbup.2024.100145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpivakovsky OV, Omelchuk SA, Kobets VV, Valko NV, Malchykova DS. Institutional Policies on Artificial Intelligence in University Learning, Teaching and Research. Inf Technol Learn Tools. 2023;97:181\u0026ndash;202. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.33407/itlt.v97i5.5395\u003c/span\u003e\u003cspan address=\"10.33407/itlt.v97i5.5395\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaliisa R, Baker RS, Wasson B, Prinsloo P. The Coming but Uneven Storm: How AI Regulation Will Impact AI and Learning Analytics Research in Different Countries. J Learn Anal. 2025;12:140\u0026ndash;57. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18608/jla.2025.8495\u003c/span\u003e\u003cspan address=\"10.18608/jla.2025.8495\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGonsalves C. Addressing student non-compliance in AI use declarations: implications for academic integrity and assessment in higher education. Assess Eval High Educ. 2025;50:592\u0026ndash;606. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602938.2024.2415654\u003c/span\u003e\u003cspan address=\"10.1080/02602938.2024.2415654\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahmud S. (2024). Fostering Academic Integrity in the Age of Artificial Intelligence. In Academic integrity in the age of artificial intelligence (IGI Global Scientific Publishing), pp. 1\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4018/979-8-3693-0240-8.ch001\u003c/span\u003e\u003cspan address=\"10.4018/979-8-3693-0240-8.ch001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim JJH, Um RS, Lee JWY, Ajilore O. Generative AI can fabricate advanced scientific visualizations: ethical implications and strategic mitigation framework. AI Ethics. 2025;5:4481\u0026ndash;93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43681-024-00439-0\u003c/span\u003e\u003cspan address=\"10.1007/s43681-024-00439-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBellini V, Semeraro F, Montomoli J, Cascella M, Bignami E. Between human and AI: assessing the reliability of AI text detection tools. Curr Med Res Opin. 2024;40:353\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03007995.2024.2310086\u003c/span\u003e\u003cspan address=\"10.1080/03007995.2024.2310086\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHartung J, Reuter S, Kulow VA, F\u0026auml;hling M, Spreckelsen C, Mrowka R. Experts fail to reliably detect AI-generated histological data. Sci Rep. 2024;14:28677. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-73913-8\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-73913-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGanjavi C, Eppler MB, Pekcan A, Biedermann B, Abreu A, Collins GS, Gill IS, Cacciamani GE. (2024). Publishers\u0026rsquo; and journals\u0026rsquo; instructions to authors on use of generative artificial intelligence in academic and scientific publishing: bibliometric analysis. BMJ \u003cem\u003e384\u003c/em\u003e, e077192. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmj-2023-077192\u003c/span\u003e\u003cspan address=\"10.1136/bmj-2023-077192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEditorial policies | Nature. Portfolio \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nature.com/nature-portfolio/editorial-policies\u003c/span\u003e\u003cspan address=\"https://www.nature.com/nature-portfolio/editorial-policies\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGenerative AI. policies for journals www.elsevier.com. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals\u003c/span\u003e\u003cspan address=\"https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAI Policy Taylor \u0026amp;, Francis. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://taylorandfrancis.com/our-policies/ai-policy/\u003c/span\u003e\u003cspan address=\"https://taylorandfrancis.com/our-policies/ai-policy/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlam S, Constantini R, Harp G, Hoekstra R, Konda S, Lisankie K, McNeice K, Olivero T, Rose I, van Rossum J, et al. Recommendations for a Classification of AI Use in Academic Manuscript Preparation. Association of Scientific, Technical and Medical; 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHyde A, Chodacki J, Shannon P. An Initial Scholarly AI Taxonomy. Front Matter. 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.54900/6p6re-xyj61\u003c/span\u003e\u003cspan address=\"10.54900/6p6re-xyj61\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllen L, O\u0026rsquo;Connell A, Kiermer V. How can we ensure visibility and diversity in research contributions? How the Contributor Role Taxonomy (CRediT) is helping the shift from authorship to contributorship. Learn Publish. 2019;32:71\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/leap.1210\u003c/span\u003e\u003cspan address=\"10.1002/leap.1210\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakagawa S, Ivimey-Cook ER, Grainger MJ, O\u0026rsquo;Dea RE, Burke S, Drobniak SM, Gould E, Macartney EL, Martinig AR, Morrison K, et al. Method reporting with initials for transparency (MeRIT) promotes more granularity and accountability for author contributions. Nat Commun. 2023;14:1788. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41467-023-37039-1\u003c/span\u003e\u003cspan address=\"10.1038/s41467-023-37039-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarnes AJ, Zhang Y, Valenzuela A. AI and culture: Culturally dependent responses to AI systems. Curr Opin Psychol. 2024;58:101838. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.copsyc.2024.101838\u003c/span\u003e\u003cspan address=\"10.1016/j.copsyc.2024.101838\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosseini M, Resnik DB, Holmes K. The ethics of disclosing the use of artificial intelligence tools in writing scholarly manuscripts. Res Ethics. 2023;19:449\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/17470161231180449\u003c/span\u003e\u003cspan address=\"10.1177/17470161231180449\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerlich M. Perceptions and Acceptance of Artificial Intelligence: A Multi-Dimensional Study. Soc Sci. 2023;12:502. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/socsci12090502\u003c/span\u003e\u003cspan address=\"10.3390/socsci12090502\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eO\u0026rsquo;Dea RE, Parker TH, Chee YE, Culina A, Drobniak SM, Duncan DH, Fidler F, Gould E, Ihle M, Kelly CD, et al. Towards open, reliable, and transparent ecology and evolutionary biology. BMC Biol. 2021;19:68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12915-021-01006-3\u003c/span\u003e\u003cspan address=\"10.1186/s12915-021-01006-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePottier P, Lagisz M, Burke S, Drobniak SM, Downing PA, Macartney EL, Martinig AR, Mizuno A, Morrison K, Pollo P et al. (2024). Title, abstract and keywords: a practical guide to maximize the visibility and impact of academic papers. Proc. R. Soc. B: Biol. Sci. \u003cem\u003e291\u003c/em\u003e, 20241222. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1098/rspb.2024.1222\u003c/span\u003e\u003cspan address=\"10.1098/rspb.2024.1222\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenoit K, Watanabe K, Wang H, Nulty P, Obeng A, M\u0026uuml;ller S, Matsuo A. quanteda: an R package for the quantitative analysis of textual data. J Open Source Softw. 2018;3:774. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21105/joss.00774\u003c/span\u003e\u003cspan address=\"10.21105/joss.00774\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeinerer I, Hornik K, Meyer D. (2008). Text mining infrastructure in \u003cem\u003eR\u003c/em\u003e. J. Stat. Softw. \u003cem\u003e25\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v025.i05\u003c/span\u003e\u003cspan address=\"10.18637/jss.v025.i05\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilge J, Robinson D. J Open Source Softw. 2016;1:37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21105/joss.00037\u003c/span\u003e\u003cspan address=\"10.21105/joss.00037\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. tidytext: Text Mining and Analysis Using Tidy Data Principles in R.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBouchet-Valat M. (2023). SnowballC. Version 0.7.1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGr\u0026uuml;n B, Hornik K. (2011). topicmodels : an \u003cem\u003eR\u003c/em\u003e package for fitting topic models. J. Stat. Softw. \u003cem\u003e40\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v040.i13\u003c/span\u003e\u003cspan address=\"10.18637/jss.v040.i13\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJockers ML. (2015). Syuzhet: Extract Sentiment and Plot Arcs from Text.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChalmers A, Klingler-Vidra R, van den Malou O. From diffusion to diffuse-ability: a text-as-data approach to explaining the global diffusion of corporate sustainability policy. Int Stud Q. 2024;68:sqad106. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/isq/sqad106\u003c/span\u003e\u003cspan address=\"10.1093/isq/sqad106\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchiff DS. Framing contestation and public influence on policymakers: evidence from US artificial intelligence policy discourse. Policy Soc. 2024;43:255\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/polsoc/puae007\u003c/span\u003e\u003cspan address=\"10.1093/polsoc/puae007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for science. Nature. 2023;614:214\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-023-00340-6\u003c/span\u003e\u003cspan address=\"10.1038/d41586-023-00340-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePorsdam Mann S, Vazirani AA, Aboy M, Earp BD, Minssen T, Cohen IG, Savulescu J. Guidelines for ethical use and acknowledgement of large language models in academic writing. Nat Mach Intell. 2024;6:1272\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s42256-024-00922-7\u003c/span\u003e\u003cspan address=\"10.1038/s42256-024-00922-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaibe-Kains B, Adam GA, Hosny A, Khodakarami F, Massive Analysis Quality Control (MAQC) Society Board of Directors, Shraddha, Kusko T, Sansone R, Tong S-A, Wolfinger W et al. R.D.,. (2020). Transparency and reproducibility in artificial intelligence. Nature \u003cem\u003e586\u003c/em\u003e, E14\u0026ndash;E16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-020-2766-y\u003c/span\u003e\u003cspan address=\"10.1038/s41586-020-2766-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkulmowski A, Engel-Hermann P. The ethics of erroneous AI-generated scientific figures. Ethics Inf Technol. 2025;27:31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10676-025-09835-4\u003c/span\u003e\u003cspan address=\"10.1007/s10676-025-09835-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEmsley R. ChatGPT: these are not hallucinations \u0026ndash; they\u0026rsquo;re fabrications and falsifications. Schizophrenia. 2023;9:52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41537-023-00379-4\u003c/span\u003e\u003cspan address=\"10.1038/s41537-023-00379-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM\u0026aacute;jovsk\u0026yacute; M, Čern\u0026yacute; M, Kasal M, Komarc M, Netuka D. Artificial intelligence can generate fraudulent but authentic-looking scientific medical articles: pandora\u0026rsquo;s box has been opened. J Med Internet Res. 2023;25:e46924. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/46924\u003c/span\u003e\u003cspan address=\"10.2196/46924\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDaneshjou R, Smith MP, Sun MD, Rotemberg V, Zou J. Lack of transparency and potential bias in artificial intelligence data sets and algorithms: a scoping review. JAMA Dermatol. 2021;157:1362\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamadermatol.2021.3129\u003c/span\u003e\u003cspan address=\"10.1001/jamadermatol.2021.3129\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiddiq ML, Islam-Gomes A, Sekerak N, Santos JCS. (2025). Large language models for software engineering: a reproducibility crisis. Preprint at arXiv. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2512.00651 https://doi.org/10.48550/arXiv.2512.00651\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2512.00651 10.48550/arXiv.2512.00651\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSozon M, Parnther C. Scientific misconduct in the era of artificial intelligence: insights from the global south. High Educ Q. 2026;80:e70097. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/hequ.70097\u003c/span\u003e\u003cspan address=\"10.1111/hequ.70097\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNie X, Tian Y, Liu M, Wu D, Guo Y. The impact of generative artificial intelligence on students\u0026rsquo; higher order thinking: Evidence from a three-level meta-analysis. Educ Inf Technol. 2025;30:25359\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-025-13735-x\u003c/span\u003e\u003cspan address=\"10.1007/s10639-025-13735-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVaccaro M, Almaatouq A, Malone T. When combinations of humans and AI are useful: A systematic review and meta-analysis. Nat Hum Behav. 2024;8:2293\u0026ndash;303. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41562-024-02024-1\u003c/span\u003e\u003cspan address=\"10.1038/s41562-024-02024-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Sun Z, Chen Y. Effects of higher education institutes\u0026rsquo; artificial intelligence capability on students\u0026rsquo; self-efficacy, creativity and learning performance. Educ Inf Technol. 2023;28:4919\u0026ndash;39. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-022-11338-4\u003c/span\u003e\u003cspan address=\"10.1007/s10639-022-11338-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWahle JP, Ruas T, Mohammad SM, Meuschke N, Gipp B. (2023). AI usage cards: responsibly reporting AI-generated content. In 2023 ACM/IEEE Joint Conference on Digital Libraries (JCDL) (IEEE), pp. 282\u0026ndash;284. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JCDL57899.2023.00060\u003c/span\u003e\u003cspan address=\"10.1109/JCDL57899.2023.00060\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartinig AR, Burk SLP, Drobniak SM, Perry I, Morrison K, Petersohn M, Pottier P, Nakagawa S, Pollo P, Ricolfi L et al. (2025). Dragon kill points: applying a transparent working template to relieve authorship stress (preprint). Preprint at EcoEvoRxiv.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"research-integrity-and-peer-review","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ripr","sideBox":"Learn more about [Research Integrity and Peer Review](http://researchintegrityjournal.biomedcentral.com)","snPcode":"41073","submissionUrl":"https://submission.nature.com/new-submission/41073/3","title":"Research Integrity and Peer Review","twitterHandle":"@RIPRJournal","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, generative AI, LLM, large language models, transparent science, open research practises, reporting guidelines","lastPublishedDoi":"10.21203/rs.3.rs-9160721/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9160721/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGenerative artificial intelligence (AI) is rapidly becoming embedded across scientific workflows, yet mechanisms for transparently documenting its use remain fragmented and weakly enforced. Focusing on ecology and evolutionary biology as a model discipline, we systematically mapped AI-related journal policies across 230 journals and assessed article-level compliance using a large sample of recent publications. To provide a reporting background, we also synthesised author contribution guidelines. Nearly half of journals provided no guidance on AI use, and where policies existed, they were largely generic, publisher-driven, and poorly translated into reporting practice. While author contribution statements were widely adopted, explicit AI disclosures appeared in fewer than 6% of papers, even in journals with formal AI policies. Text-mining of 124 guideline documents revealed highly standardised, precautionary language emphasising responsibility and prohibitions, with minimal operational guidance on acceptable uses or disclosure formats.\u003c/p\u003e \u003cp\u003eTo address this gap, we introduce AIdIT (AI disclosure for Improved Transparency), a standardised, taxonomy-based framework for reporting AI use across all stages of the research lifecycle. AIdIT integrates structured categories of AI use, human oversight statements, and machine-readable outputs to support reproducibility, accountability, and comparability. Together, our systematic evidence synthesis and proposed framework highlight an urgent need to normalise AI transparency as a core component of open research practice.\u003c/p\u003e","manuscriptTitle":"A systematic map of generative AI guidelines and reporting in ecology and evolutionary biology: towards the framework of AI disclosure for Improved Transparency (AIdIT)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-09 11:52:53","doi":"10.21203/rs.3.rs-9160721/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-23T15:25:49+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-15T14:32:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"261541390713680586394396615139281979938","date":"2026-04-11T07:46:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-09T13:17:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"114741939962665169782945881791007358901","date":"2026-04-09T13:03:39+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-03T15:56:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-20T04:38:43+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-20T04:38:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"Research Integrity and Peer Review","date":"2026-03-18T14:38:41+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"research-integrity-and-peer-review","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ripr","sideBox":"Learn more about [Research Integrity and Peer Review](http://researchintegrityjournal.biomedcentral.com)","snPcode":"41073","submissionUrl":"https://submission.nature.com/new-submission/41073/3","title":"Research Integrity and Peer Review","twitterHandle":"@RIPRJournal","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"08197cce-59cf-4841-8ace-698a4eb367ab","owner":[],"postedDate":"April 9th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-04-23T15:40:35+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-09 11:52:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9160721","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9160721","identity":"rs-9160721","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}