How Do Biology Undergraduates Use AI-enabled Feedback to Revise Scientific Writing?

How Do Biology Undergraduates Use AI-enabled Feedback to Revise Scientific Writing? | 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 Research Article How Do Biology Undergraduates Use AI-enabled Feedback to Revise Scientific Writing? Michele Weston, Tammy Long This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9271072/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Instructors and education researchers are increasingly leveraging genAI to support feedback practices, yet we still know relatively little about how students use AI-enabled feedback. This gap is consequential because students must actively interpret and use feedback to benefit from it. This study examined how two teams of undergraduate biology students used AI-enabled feedback to revise their research proposals. The teams revised their proposal drafts after receiving AI-enabled feedback developed by a GPT and subsequently reviewed and edited by the instructor. Our findings show that AI-enabled feedback prompted actionable revision across both teams. Each team addressed approximately half of the feedback comments, indicating comparable levels of uptake despite their differing revision approaches. This pattern indicates that both teams were inconsistent in taking up the additional information and reasoning highlighted in the feedback, reflecting challenges similar to those commonly observed with traditional feedback. These findings provide behavioral evidence of how students used AI-enabled feedback to revise their scientific writing and demonstrates ways that prior research on feedback use and barriers can extend to AI-enabled feedback. Artificial Intelligence and Machine Learning artifical intelligence generative AI scientific writing undergraduate biology education Introduction Feedback on learning tasks is understood to be one of the most influential factors contributing to student achievement (Black & Wiliam, 1998 ; Hattie, 1999 ). Modern conceptualizations of feedback consider it an active, learner-centered process in which students use an input (e.g. written evaluative comments) to inform changes to an output (e.g. subsequent work; Henderson et al., 2019 ; Sadler, 2010 ). This view places students’ ability to understand, interpret, and act on feedback at the center of effective learning. Accordingly, recent research has emphasized students’ perspectives to better understand how they perceive and make use of feedback (e.g. Dawson et al., 2019 ; Jonsson, 2013 ). Students’ engagement with feedback on writing varies across contexts: in some cases they do not read or use it (Hyland, 1998 ; MacDonald, 1991 ), whereas in others, they incorporate it into their revisions (Higgins et al., 2002 ; Winstone, Nash, et al., 2017 ). Yet not all revisions lead to meaningful improvement; students often focus on changes to spelling, grammar, and unnecessary details rather than substantive refinements (Ashwell, 2000 ; Zhu et al., 2020 ). Work examining student perceptions offers insight into why it can be challenging to use feedback. Winstone et al. ( 2017 ) found that students often struggle to translate feedback comments into next steps. Research in this area has identified several barriers that contribute to these difficulties: students may be unable to decode what the feedback means, lack strategies for implementing it, feel they do not have the agency to revise, or lack the volition to invest effort in improving their work (Jonsson & Panadero, 2018 ; Nicol & Macfarlane-Dick, 2006 ; Sutton, 2012 ; Winstone et al., 2017 ). Unsurprisingly, students’ perceptions of what makes feedback useful aligns closely with overcoming these barriers. They prefer feedback that is personalized, detailed, and actionable (Dawson et al., 2019 ), as personalization and detail support comprehension while actionable guidance scaffolds next steps—promoting both agency and willingness to revise (Dawson et al., 2019 ; Higgins et al., 2002 ). Instructor feedback plays a crucial role in supporting students’ scientific writing, such as lab reports and research proposals, particularly when there are opportunities to reflect and revise (Jerde & Taper, 2004 ; Yeh, 2010 ). For instance, Weaver et al. ( 2014 ) examined an undergraduate organic chemistry lab course in which students engaged in iterative cycles of revision to lab reports informed by peer and instructor feedback. Students who participated in this redesigned curriculum earned significantly higher lab report grades than students in previous semesters and attested to substantial improvements in their writing skills. Despite these benefits, many instructors face time and workload constraints that limit their ability to provide personalized feedback (Heim & Holt, 2019 ). As a result, students may receive minimal feedback—or none at all. AI-enabled Feedback Generative AI (genAI) is emerging as a promising tool for assisting instructors in developing formative feedback on students’ scientific writing (Pellegrino, 2024 ). GenAI refers to artificial intelligence (AI) systems capable of producing original content, such as written text, in response to a user prompting (Stryker & Scapicchio, 2024 ). Compared to traditional machine learning (ML) approaches, genAI systems offer advantages for feedback generation: their natural-language prompting makes them easy to interact with, and their pretrained large language models allow them to perform tasks without requiring the large, scored datasets typically needed for supervised ML systems. These affordances create a low barrier to entry and position genAI to be adopted more quickly and widely by instructors. Promoting Quality AI-enabled Feedback As genAI becomes more widely applied in educational settings, it is important to foster ethical use by intentionally aligning AI implementations with instructional goals and values. Human-centered AI offers one such approach by emphasizing design practices that augment rather than replace human capabilities and that maintain human oversight of AI systems (Schmager et al., 2025 ). We define AI-enabled feedback as that which is developed within a human-centered AI process. Keeping instructors actively involved in the feedback process helps ensure higher-quality output and preserves accountability for assessment. In practice, human involvement may include configuring and testing an assignment-specific AI system, as well as reviewing and refining AI output before sharing it with students. A related strategy for promoting quality AI-enabled feedback involves customizing AI systems to perform the specific task reflecting the course context, learning goals, and assessment criteria. Such customization enhances the accuracy of the feedback and reduces hallucinations, which are instances in which the model produces inaccurate yet plausible text (Mills et al., 2025 ; OpenAI et al., 2024). Understanding GenAI’s Impact on Learning GenAI tools are increasingly used by instructors to automate, augment, and expand their feedback practices, yet we currently lack evidence about how students use AI-enabled feedback (Cooper, 2026 ). This gap is consequential because students must actively interpret and use feedback to benefit from it. It is unclear whether existing understandings about feedback processes similarly apply to AI-enabled feedback (Kasneci et al., 2023 ). One emerging application of genAI is the development of feedback on students’ scientific writing. In this study, we provide empirical evidence of how undergraduate biology students used AI-enabled feedback to revise their research proposals. Methods We used a qualitative, multiple case study design with a comparative document analysis to investigate students’ engagement with feedback during the revision process. Multiple case studies examine a phenomenon within bounded contexts, or cases, and take a holistic approach to understanding the complex and situated experiences of participants (Merriam, 1998 ; Miles & Huberman, 1994 ). This approach is well suited for developing and presenting insights that inform our understanding of student use of AI-enabled feedback. The study includes two cases, each consisting of a team of three students. Study Context This study was conducted within a STEM-focused residential college at a large research university located in the Midwest. The research took place in the laboratory component of a combined lecture and laboratory course on introductory organismal biology. The laboratory curriculum was designed to explore systemic inequities and human impacts on ecology and health. The first part of the course introduced students to different ecological systems, methods, and types of data analysis. Following this, student teams developed independent research projects investigating the underlying cause of a human health and ecological disparity of their choice. Although research projects often utilize final papers resembling journal articles, in this course each team was required to incorporate their results into a written research proposal as preliminary findings that will inform a new project. Although the original intention was for students to complete full research proposals, time constraints required the assignment to be shortened to the introduction section only. Therefore, this report will refer to the piece of writing as the “proposal introduction”. Students were instructed to follow a rubric that specified the content of four required paragraphs, each focusing on a different scale of human impact. Descriptions of the paragraphs are shown in Table 1 . Participants All students enrolled in the laboratory course were invited to participate in the research study. There were two student teams where all members consented to participate, and they were each considered a case. Table 2 provides demographic and academic information for these participants. Table 1 Content of Proposal Introduction Paragraphs Paragraph Name Description of Content Macro This is the first paragraph of the proposal introduction. Describe the problem at the macro (historical and societal) scale. Explain how historical practices (e.g. redlining) result in current racial disparities in health outcomes (e.g., asthma rates) and local ecosystems (e.g., local canopy cover and heat islands). Meso Part 1 This is the second paragraph of the proposal introduction. Describe the problem at the meso (community) scale. Explain the impacts of specific policies or practices (e.g. urban renewal, highway construction) on local resources, community health, and environmental conditions, focusing on neighborhood structures and resources. Meso Part 2 This is the third paragraph of the proposal introduction. Describe one or more community responses to inequity that are relevant to the experimental question and variables. Micro This is the fourth paragraph of the proposal introduction. Describe the problem at the micro (individual) scale. Explain how systemic inequities affect personal health outcomes and ecological interactions, highlighting specific cases and personal experiences (e.g., personal stories of residents in affected areas). Table 2 Student Participant Demographics Participant a Team Number Gender Race(s) and Ethnicity/ies Final Course Cumulative GPA c Student 1 1 Woman White 4.0 3.83 Student 2 1 Man White 3.0 2.75 Student 3 1 Woman South Asian 4.0 3.79 Student 4 2 Man Black or African American 4.0 3.59 Student 5 2 Woman Black or African American 4.0 2.44 Student 6 2 Woman White 4.0 4.0 a Student participants are listed without pseudonyms to prevent the association of demographic data with the comments of individuals. b Final course grade is reported on a 4.0 scale. c Cumulative GPA is reported on a 4.0 scale and at the start of the semester in which the study took place. Procedures Student teams began the assignment by drafting the proposal introduction and submitting it to their instructor. The instructor then used a genAI tool (described below) to develop customized feedback aligned with the assignment rubric, reviewed the feedback for accuracy and quality, and made edits as necessary before providing it to students. The instructor had the option of using follow-up prompts to refine the GPTs’ output but opted not to do so. Students then submitted the revised proposal introduction as the final draft. Generative AI Tool This study utilized a GPT, which is a specialized version of ChatGPT developed to carry out a specific task ( Introducing GPTs , 2024). Although it appeared similar to the standard ChatGPT interface where the instructor interacted with a chatbot through a chat window, this GPT (model 4o; OpenAI, 2025 ) utilized detailed instructions and a curated knowledge base, allowing the AI to query course-relevant documents or data and integrate these materials into its responses, which reduces hallucination by grounding the model’s predictions ( Introducing GPTs , 2024; What Is RAG? , n.d.). Importantly, GPTs retain their protocols and reference materials across sessions, allowing the user to repeat the specialized task without re-entering instructions. Additional details about the GPT are provided in the Supplemental Materials. Having outlined the tool’s general features, we now describe how this GPT was intentionally grounded in learning theory and course-specific supports. The lab coordinator of the course and the first author collaborated to design the research proposal assignment and its rubric, ensuring alignment with the coordinator’s pedagogical priorities. The rubric incorporated the Claim, Evidence, and Reasoning (CER) framework, which had been used throughout the course to support students’ scientific explanations (McNeill & Krajcik, 2008 ). To further guide students, they developed a scaffolding tool that outlined the structure of the proposal introduction and specified the content expected in each paragraph. Building on this foundation, the first author configured and tested the GPT to provide feedback tailored to the rubric, CER framework, and scaffolding tool. The GPT was instructed to give an output consisting of 1) the proposal’s strengths, 2) suggestions for improvement, and 3) examples to scaffold the suggestions for improvement for each rubric section as described in Table 3 . Student data was protected by collecting written consent for the use of genAI in analyzing student work, anonymizing proposals prior to use of the GPT, and by ensuring that these proposals were excluded from the GPT training dataset. Table 3 Rubric and Feedback Sections Feedback Section Scope Description Macro Paragraph-specific Feedback pertaining to the content of the Macro paragraph. Meso Paragraph-specific Feedback pertaining to the content of both the Meso Part 1 and Meso Part 2 paragraphs. Micro Paragraph-specific Feedback pertaining to the content of the Micro paragraph. Flow of Writing Whole-proposal introduction Feedback on how effectively each sentence and paragraph transitions to the next, with a clear flow of ideas. Claims are Supported Whole-proposal introduction Feedback on how well claims are supported by scientific reasoning and evidence from credible cited sources. Note. Feedback in each section followed the standardized format of strengths, suggestions for improvement, and scaffolding tips. Data Analysis We conducted a two-part revision analysis: first, we compared the first and final drafts to identify and categorize changes; then, we cross-referenced those changes with the AI-enabled feedback to examine how students revised in response to it. We applied an adapted revision operations coding scheme (Faigley & Witte, 1981 ) to the drafts to capture the meaning-based changes as additions, deletions, moves, and substitutions. A change was considered meaning-based if it involved conceptual ideas or information. Additions and deletions of citations were coded separately to distinguish them from meaning-based changes. Surface-level revisions were excluded from analyses and included changes that corrected spelling or grammar, or edits that influenced the implicit or explicit nature of an idea without changing the idea itself. When a revision was longer than a single sentence, we coded each sentence separately. Table 4 in the Supplemental Materials presents illustrative examples of the revision operations coding. One exception occurred with Team 1, whose first draft was missing one of the required paragraphs. Because there was no original text to revise, the corresponding paragraph in the final draft was not coded for revisions. This ensured that the revision analysis focused only on changes to existing content. While the initial coding captured the types of changes students made, it did not indicate whether they addressed the feedback provided. To assess this, we identified draft segments corresponding to each feedback comment and categorized the comments as fully, partially, or not addressed, based on the extent to which suggested changes appeared in the revised draft. The analysis focused only on feedback that offered a suggestion for improvement or a scaffolding tip (e.g., examples of how to improve). Feedback highlighting strengths were excluded because they did not require revision. We conducted all coding and analysis in MaxQDA (VERBI Software, 2024 ). Table 5 in the Supplemental Materials illustrates examples of feedback comments, their categorization, and associated excerpts from student drafts. Analysis This section begins with the results of the revision analysis including students’ use of feedback comments. That is followed by within-case analyses for each team to illustrate and contextualize the results of the revision analysis, and a cross-case analysis synthesizing the findings. The section ends with theoretical interpretation of the findings. Revisions Analysis We conducted a two-part revision analysis to examine how students used AI-enabled feedback to revise their writing. First, we compared each team’s first and final drafts to identify and categorize fine-grained differences. We then cross-referenced the results with the AI-enabled feedback to determine the extent to which comments were addressed in the revised drafts. Below, we first summarize the revision operations used by each team and then describe the extent to which revisions addressed feedback comments. Across both teams, all four types of meaning-based revisions were observed. Deletions were the most frequent (15 of 31 changes), followed by additions (10 of 31 changes). Team 1 made more revisions overall, despite their first draft missing a required paragraph. The teams also differed in their distribution of meaning-based changes: Team 1 primarily deleted content, whereas Team 2 made more additions than deletions. Notably, both teams removed citations from their first drafts and did not introduce any new citations in their final drafts. Table 6 in the Supplemental Materials presents a detailed breakdown of revision operations used by each team. To examine how student teams incorporated AI-enabled feedback into their revisions, we classified each feedback comment as fully, partially, or not addressed, based on the extent to which suggested changes were reflected in the revised draft. Because Team 1’s initial draft was missing a required paragraph, they had fewer feedback comments included in the analysis. Across both teams, 10 of the 22 comments were fully or partially addressed, leaving more than half unaddressed. The distribution of comments across the categories fully, partially, and not addressed was similar for both teams and consistent across comment type (suggestions for improvement vs. scaffolding tips). Table 7 in the Supplemental Materials presents the distribution of comments across categories of feedback use. Team 1 Within-Case Analysis Team 1 investigated the question, “In previously redlined communities, what is the relationship between tree density and median income, asthma frequency, and UV exposure?” In addition to examining the effects of redlining, their proposal introduction incorporated the construction of a nearby highway as a historically linked event with ongoing environmental and health consequences. They also described a local garden program as a community response initiative that could mitigate the harms perpetuated by redlining and the highway project. As noted earlier, Team 1 did not include a Micro paragraph in their first draft, and because the Micro paragraph in their final draft was newly written, it is excluded from this within-case analysis. Overall, as the following analysis shows, Team 1 responded to AI-enabled feedback by improving conciseness and organization more than enhancing substantive content. Although feedback comments consistently requested additional information, explanation, reasoning, and evidence, Team 1 made only one addition to their final draft. This was a transition sentence that fully addressed the feedback comment suggesting they “…add a transition sentence between the Meso and Micro paragraphs to set up the shift from community impacts to individual outcomes.” In contrast, feedback comments that required adding new explanations or information largely went unaddressed. For instance, Team 1 did not address comments that asked them to 1) explain ways the local pattern of redlining compared to other cities, 2) explain how the community response initiative might relate to the study’s variables, or 3) better support claims with scientific reasoning or evidence, particularly those around the benefits of the community response initiative. Instead, the team most often responded to AI-enabled feedback through structural rather than conceptual revisions. Collectively, three comments on the Macro paragraph had encouraged them to trace a timeline of consequences that would “help clarify how specific national practices led to systemic racial inequalities in housing and health that persist today”, beginning with the policies that institutionalized redlining. In response, Team 1 moved information about the highway project and its enduring impacts from the Meso Part 1 paragraph into the Macro paragraph. This shift partially addressed the feedback by extending the list of consequences. However, it did not conceptually connect the highway project to redlining nor elaborate on the policies that institutionalized redlining. In other cases, Team 1 responded to feedback by removing text. For example, feedback advised that “Some claims—particularly around individual health outcomes and the benefits of the Garden Program—could be better supported with scientific reasoning or specific evidence”. The students responded to this comment indirectly by deleting an unsupported claim about the garden program in the Meso Part 2 paragraph. While this removal can be understood as an attempt to resolve the feedback by avoiding a weak claim, it did not involve adding reasoning or evidence. As such, the feedback was categorized as unaddressed. Team 1’s revision pattern also reflected a selective uptake of AI-enabled feedback, with many deletions aimed at tightening the writing rather than expanding the content as comments had encouraged. In fact, Team 1 made so many deletions that the word count shrunk from 465 words in the first draft to 292 words in the final draft. As an example, the Macro paragraph originally contained two redundant explanations of redlining. The team removed the first explanation by deleting three text segments, leaving behind a sufficient introduction to redlining. This pattern extended across the proposal, as other deletions cut out unnecessary or redundant details, improving clarity and conciseness. In contrast, two of the team’s deletions of text from the Meso Part 1 paragraph removed details that were important for meeting rubric requirements. The rubric asked students to describe events that led to the perpetuation and exacerbation of inequities in ecosystems and human health. However, the team removed an explanation that the area's historically lower tree density had resulted from redlining and disinvestment, and that economic barriers compounded to make it difficult for residents to mitigate harmful environmental effects. Similarly, the team deleted 5 citations that were essential for supporting claims. These examples show how the Team’s selective uptake of feedback is complex and involved competing goals for the revisions. Team 2 Within-Case Analysis Team 2 investigated the question, “How has redlining affected community garden quality and prevalence in the greater [Midwestern city] area and how is it tied to obesity?” Their proposal focuses on redlining and positioned obesity as a long-term health outcome tied to it. To contextualize the relationship, they examined how poor soil quality and limited access to grocery stores in historically redlined areas contribute to elevated obesity rates. The team also incorporated community gardens that donate produce to a local food bank as a response initiative aimed at reducing environmental and health disparities. The following analysis demonstrates that Team 2 responded to AI-enabled feedback by selectively expanding the proposal’s substantive content. One way Team 2 responded to AI-enabled feedback was by adding the explanations and details that comments requested. For example, feedback on the Meso paragraphs suggested they provide more description of the local food bank to clarify its role as a community response initiative, as the first draft had referenced it without supporting details. The team fully addressed the comments by adding a description that the food bank “aims to provide food pantries in the area with fresh produce. The [local food bank] is a non-profit organization that works with the community to provide food to those in need.” Team 2 also incorporated a small addition in their Macro paragraph in response to feedback asking them to “ explain how these policies caused long-term economic and health disparities. The connection between historical redlining and present-day diabetes is mentioned, but the causal chain could be more fully explained.” (Team 2 originally wrote about diabetes as another outcome variable before narrowing the scope to only obesity). They added the qualifier “limited” to describe healthcare access in the sentence, “This segregation has had lasting effects, contributing to health inequities like higher obesity and diabetes rates due to systemic barriers such as limited healthcare access and economic hardship.” Although their argument remained similar across drafts, the revision clarified the causal mechanism by specifying that healthcare access was less available in redlined areas. In contrast, Team 2 also made an addition that did not meaningfully address the feedback comment or help meet rubric criteria. In the Macro paragraph, they added the sentence, “Red and yellow areas, home to low-income Black residents, faced mortgage discrimination, while blue and green zones housed wealthier white populations.” While this elaboration identifies the color labels used in redlining maps, it does not explain how redlining’s short- and long-term consequences connect to present-day inequities, nor or as another feedback comment requested, does it provide the context around the unique history of redlining in the specific Midwestern city compared to other cities. It is notable that the team incorporated this addition immediately after applying a scaffolding tip that provided a suggested sentence starter (“In the 1930s, redlining was codified through federal housing policy…”). Their revision reads, “In the 1930s, the Home Owners’ Loan Corporation (HOLC) and the Federal Housing Administration (FHA) institutionalized segregation through redlining, classifying neighborhoods into four color-coded zones. Red and yellow areas, …” A prior feedback comment also mentioned HOLC and FHA as historical policies, and it appears that the students found prescriptive or highly structured feedback easier to follow, whereas more open-ended feedback made it difficult for them to identify what substantive information to include next. Another way Team 2 selectively responded to AI-enabled feedback was by focusing on paragraph-specific comments over broader suggestions aimed at improving the proposal’s flow and strength of arguments. Team 2 fully or partially addressed 6 of the 9 paragraph-specific comments, but none of the 4 whole-proposal comments. For example, a suggestion in the “Flow of writing” section noted that, “some paragraphs jump between topics (e.g., soil quality to obesity to garden location) without clear transitions. Strengthen the flow by using topic sentences that guide the reader and connect ideas more smoothly.” None of the team’s revisions improved transitions, so this feedback was categorized as unaddressed. Similarly, the “Claims are supported” section advised strengthening several claims with citations or scientific explanation and identified the link between worms, soil quality, and hunger as an example. Because none of the final draft’s added content provided evidence or reasoning—and because four citations from the first draft were deleted—these comments also remained unaddressed. Cross-Case Analysis A comparison of the two within-case analyses shows similarities and differences in how students used AI-enabled feedback to revise their research proposals. The teams’ responses indicate that AI-enabled feedback can prompt different types of revision work: Team 1 primarily used it to refine structure through text deletions and moves, while Team 2 used it to elaborate content via additions. Despite these differences in revision strategies, the distribution of feedback comments across the categories fully, partially, and not addressed was similar for both teams, suggesting comparable levels of uptake. This similarity in uptake reflects neither team consistently incorporating the conceptual information or reasoning that the feedback most often requested. Although Team 1’s structural revisions occasionally aligned with feedback—such as by moving information to a location that needed elaboration or removing unsupported claims—these changes did not expand the substantive content the feedback sought. Similarly, while Team 2 prioritized additions, they left many open-ended or conceptually demanding suggestions unaddressed. The remainder of the cross-case analysis describes the types of feedback that had limited use by both teams and offers possible explanations for the patterns observed. A key source of limited uptake for both teams involved open-ended feedback that required identifying new information or conducting additional research. Across cases, comments that encouraged students to add location-specific historical details or to provide conceptual connections, scientific reasoning, or evidence were typically only partially addressed or left unaddressed. This pattern was especially apparent for Team 1, whose only addition was a transition sentence responding to a relatively prescriptive suggestion. Team 2 added new details—such as a description of the local food bank and the qualifier “limited” in their explanation of healthcare access— but these additions involved general knowledge that may not have required searching for external sources. For both teams, revisions that partially addressed feedback tended to meet the surface-level or easily actionable portion of a comment while omitting the more demanding element that required new evidence or conceptual elaboration. For example, Team 1 moved their description of the highway project into the Macro paragraph, which extended the list of consequences but did not include the causal explanation requested in the feedback (e.g., linking the highway project to redlining or describing the policies that institutionalized it). Incorporating new information into the proposal would have required multiple steps such as finding appropriate sources, assessing their relevance and quality, and integrating and citing them, and neither team demonstrated this type of engagement with the feedback. We interpret these missing or incomplete revisions as evidence that students preferred working with existing material over conducting additional research, suggesting either difficulty researching requested information or a reluctance to do so. Both teams also made limited use of feedback that required self‑evaluation, whether judging the coherence of the proposal, evaluating claims, or recognizing gaps in reasoning. Except for Team 1’s new transition sentence, neither team addressed the feedback in the ‘Flow of Writing’ or ‘Claims Are Supported’ sections, which applied to the whole proposal. This suggests difficulty responding to whole-proposal comments, although the assignment’s design may have contributed; each student was responsible for writing one or two paragraphs, and they may have focused only on the feedback corresponding to paragraphs they authored. However, each whole-proposal suggestion included examples tied to specific paragraphs, and students still did not revise accordingly. The scaffolding tips in the ‘Claims are Supported’ section further illustrate the self-evaluative work expected of students. These tips directed them to examine their claims across the proposal and ask questions such as, “What’s the evidence?” and “Why does the evidence support the claim?” Neither team added support for claims, suggesting difficulty evaluating the adequacy of their reasoning in addition to the reluctance to research new information described previously. This difficulty identifying weaknesses in their explanations extended beyond the ‘Claims Are Supported’ section. As described in the previous paragraph, both teams often addressed only the easily actionable portions of feedback while neglecting the more conceptually demanding aspects, such as explaining the causal chain linking redlining to present-day inequities in the environment and human health. We interpret this as an indication that students struggled not only to seek out new information but also to recognize where conceptual gaps existed in their arguments, and as a result were unsuccessful in addressing the AI-enabled feedback. Theoretical Interpretation This study adopted a learner‑centered perspective on feedback, emphasizing that students must actively use feedback—not merely receive it—to benefit from it (Sadler, 2010 ). From this standpoint, we examined the teams’ observable revision behaviors to understand how they used AI-enabled feedback. Both teams engaged in meaning-based revisions, and each addressed roughly half of the feedback comments. This suggests that many of the AI-enabled comments prompted actionable revision. Given how often students struggle to take up feedback in general (Winstone, Nash, et al., 2017 ), the degree of engagement observed here is reassuring. These findings offer insight into the behavioral side of students’ experiences with AI-enabled feedback by demonstrating how it shaped their revisions in practice. Even so, the partial uptake across teams indicates that students did not consistently incorporate the additional information or reasoning emphasized in the feedback. Even Team 2, who made more additions overall, left many open-ended or conceptually demanding suggestions unaddressed. Both teams made limited use of feedback that 1) requested new information, reasoning, or evidence, or 2) required self-evaluation. The difficulties that these teams faced align with Winstone et al.’s ( 2017 ) framework of barriers to feedback recipience, including difficulty decoding discipline-specific language, limited strategies for implementing feedback, and challenges translating comments into action. They may also experience apathy or lack the volition to take action. The challenges both teams faced here mirror well-established difficulties with using traditional feedback, suggesting that the barriers shaping feedback uptake may extend to AI-enabled feedback as well. In addition to the cognitive and affective processes described above, characteristics of the AI-enabled feedback itself may have shaped how students used it. Prior work shows that students find feedback useful when it is personalized, detailed, and actionable (Dawson et al., 2019 ). Studies of AI-enabled or -generated feedback on scientific writing similarly indicate that students perceive such feedback as useful, detailed, and actionable (Dai et al., 2023 ; Mills et al., 2025 ). However, this is not always the case. For example, Lim and Go ( 2025 ) reported that students found AI-generated feedback on essays to be overly broad, generic, and vague. Since Lim and Go relied on a standard ChatGPT prompt rather than a customized system with additional instructions and a domain-specific knowledge base, the feedback they obtained largely consisted of generalized comments resembling grading criteria. Differences in students’ perceptions across studies may reflect variation in AI tools and prompting approaches. Our findings contribute to this discussion by illustrating that a customized AI tool, grounded in learning theory and integrated with course‑level supports, can produce feedback that students are able to act on when revising their scientific writing. Conclusion Instructors and education researchers are increasingly developing feedback with the assistance of genAI, yet we still know relatively little about how students use this feedback (Cooper, 2026 ). This makes it unclear whether AI-enabled feedback supports learning or whether existing understandings about feedback processes extend to this new context. To address this, we examined how two teams of undergraduate biology students used AI-enabled feedback to revise their research proposal introductions. Our findings show that AI-enabled feedback prompted actionable revision across both teams. Each team addressed approximately half of the feedback comments, indicating comparable levels of uptake despite their differing revision approaches. While feedback comments consistently requested additional information, reasoning, and evidence, both teams commonly left open-ended and conceptually demanding suggestions unaddressed. This pattern indicates that both teams were inconsistent in incorporating the additional information and reasoning highlighted in the feedback, reflecting challenges similar to those commonly observed with traditional feedback. These findings provide behavioral evidence of how students used AI-enabled feedback to revise their scientific writing and demonstrates ways that prior research on feedback use and barriers can extend to AI-enabled feedback. Declarations Author Note Michele Weston https://orcid.org/0009-0007-7810-938X Tammy Long https://orcid.org/0000-0003-0849-9056 We have no conflict of interest to disclose. This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant 2235783, the SimBio Foundation grant program, and the Plant Biology Department of Michigan State University. Correspondence concerning this article should be addressed to Tammy Long, Plant Biology Laboratories, 612 Wilson Rd, East Lansing, MI 48824. Email: [email protected] References Ashwell T (2000) Patterns of Teacher Response to Student Writing in a Multiple-Draft Composition Classroom: Is Content Feedback Followed by Form Feedback the Best Method? J Second Lang Writ 9(3):227–257. https://doi.org/10.1016/S1060-3743(00)00027-8 Black P, Wiliam D (1998) Assessment and Classroom Learning. Assess Education: Principles Policy Pract 5(1):7–74. https://doi.org/10.1080/0969595980050102 Cooper MM (2026) A Preliminary Set of Principles to Support Learning in the Context of Generative AI. J Chem Educ 103(1):36–42. https://doi.org/10.1021/acs.jchemed.5c01413 Dai W, Lin J, Jin H, Li T, Tsai Y-S, Gašević D, Chen G (2023) Can Large Language Models Provide Feedback to Students? A Case Study on ChatGPT. 2023 IEEE International Conference on Advanced Learning Technologies (ICALT) , 323–325. https://doi.org/10.1109/ICALT58122.2023.00100 Dawson P, Henderson,Michael M, Paige, Phillips, Michael, Ryan, Tracii, Boud, David,and, Molloy E (2019) What makes for effective feedback: Staff and student perspectives. Assessment & Evaluation in Higher Education , 44 (1), 25–36. https://doi.org/10.1080/02602938.2018.1467877 Faigley L, Witte S (1981) Analyzing Revision. Coll Composition Communication 32(4):400–414. https://doi.org/10.2307/356602 Hattie J (1999), August 2 Influences on student learning [Inaugural lecture] Heim AB, Holt EA (2019) Benefits and Challenges of Instructing Introductory Biology Course-Based Undergraduate Research Experiences (CUREs) as Perceived by Graduate Teaching Assistants. CBE—Life Sci Educ 18(3):ar43. https://doi.org/10.1187/cbe.18-09-0193 Henderson M, Ajjawi R, Boud D, Molloy E (2019) Identifying Feedback That Has Impact. In M. Henderson, R. Ajjawi, D. Boud, & E. Molloy (Eds.), The Impact of Feedback in Higher Education: Improving Assessment Outcomes for Learners (pp. 15–34). Springer International Publishing. https://doi.org/10.1007/978-3-030-25112-3_2 Higgins R, Hartley P, Skelton A (2002) The Conscientious Consumer: Reconsidering the role of assessment feedback in student learning. Stud High Educ 27(1):53–64. https://doi.org/10.1080/03075070120099368 Hyland F (1998) The impact of teacher written feedback on individual writers. J Second Lang Writ 7(3):255–286. https://doi.org/10.1016/S1060-3743(98)90017-0 Introducing GPTs . (2024), March 13 https://openai.com/index/introducing-gpts/ Jerde CL, Taper ML (2004) Preparing Undergraduates for Professional Writing: Evidence Supporting the Benefits of Scientific Writing within the Biology Curriculum. J Coll Sci Teach 33(7):34–37 Jonsson A (2013) Facilitating productive use of feedback in higher education. Act Learn High Educ 14(1):63–76. https://doi.org/10.1177/1469787412467125 Jonsson A, Panadero E (2018) Facilitating Students’ Active Engagement with Feedback. In A. A. Lipnevich & J. K. Smith (Eds.), The Cambridge handbook of instructional feedback (pp. 531–553). Cambridge University Press. https://doi.org/10.1017/9781316832134 Kasneci E, Sessler K, Küchemann S, Bannert M, Dementieva D, Fischer F, Gasser U, Groh G, Günnemann S, Hüllermeier E, Krusche S, Kutyniok G, Michaeli T, Nerdel C, Pfeffer J, Poquet O, Sailer M, Schmidt A, Seidel T, Kasneci G (2023) ChatGPT for good? On opportunities and challenges of large language models for education. Learn Individual Differences 103:102274. https://doi.org/10.1016/j.lindif.2023.102274 Lim JM, Go C (2025) Integrating ChatGPT into Teacher Feedback: Practical Insights for L2 Writing Instruction. TESL-EJ, 29 (3). https://eric.ed.gov/?id=EJ1489053 MacDonald RB (1991) Developmental Students’ Processing of Teacher Feedback in Composition Instruction. In Review of Research in Developmental Education (Vol. 8, Issue 5). Managing Editor, RRIDE, National Center for Developmental Education, Reich College of Education, Appalachian State University, Boone, NC 28608 ( $ 9. https://eric.ed.gov/?id=ED354965 McNeill KL, Krajcik J (2008) Inquiry and Scientific Explanations: Helping Students Use Evidence and Reasoning. Science as inquiry in the secondary setting. NSTA, pp 121–134 Merriam SB (1998) Qualitative research and case study applications in education (Rev. and expanded). Jossey-Bass Publishers Miles MB, Huberman AM (1994) Qualitative data analysis: An expanded sourcebook, 2 edn. Sage. [Nachdr.]) Mills E, Mizouri A, Peach A (2025) Prompting Better Feedback: A Study of Custom GPT for Formative Assessment in Undergraduate Physics. Educ Sci 15(8):1058. https://doi.org/10.3390/educsci15081058 Nicol DJ, Macfarlane-Dick D (2006) Formative assessment and self‐regulated learning: A model and seven principles of good feedback practice. Stud High Educ 31(2):199–218. https://doi.org/10.1080/03075070600572090 OpenAI (2025) ChatGPT (Version GPT-4o) [Large Language Model] OpenAI, Achiam J, Adler S, Agarwal S, Ahmad L, Akkaya I, Aleman FL, Almeida D, Altenschmidt J, Altman S, Anadkat S, Avila R, Babuschkin I, Balaji S, Balcom V, Baltescu P, Bao H, Bavarian M, Belgum J, Zoph B (2024) GPT-4 Technical Report (arXiv:2303.08774). arXiv. https://doi.org/10.48550/arXiv.2303.08774 Pellegrino J (2024) A New Era for STEM Assessment: Considerations of Assessment, Technology, and Artificial Intelligence. Uses of Artificial Intelligence in STEM Education. Oxford University Press, pp 17–37 Sadler DR (2010) Beyond feedback: Developing student capability in complex appraisal. Assess Evaluation High Educ 35(5):535–550. https://doi.org/10.1080/02602930903541015 Schmager S, Pappas IO, Vassilakopoulou P (2025) Understanding Human-Centred AI: A review of its defining elements and a research agenda. Behav Inform Technol 44(15):3771–3810. https://doi.org/10.1080/0144929X.2024.2448719 Stryker C, Scapicchio M (2024), March 22 What is Generative AI? IBM. https://www.ibm.com/think/topics/generative-ai Sutton P (2012) Conceptualizing feedback literacy: Knowing, being, and acting. Innovations Educ Teach Int 49(1):31–40. https://doi.org/10.1080/14703297.2012.647781 VERBI Software (2024) MaxQDA 24.7 (Version 24.7) [Computer software]. Available from maxqda.com Weaver CL, Duran EC, Nikles JA (2014) An Integrated Approach for Development of Scientific Writing Skills in Undergraduate Organic Lab. In Addressing the Millennial Student in Undergraduate Chemistry (Vol. 1180, pp. 105–123). American Chemical Society. https://doi.org/10.1021/bk-2014-1180.ch008 What is RAG? - Retrieval-Augmented Generation AI Explained - AWS . (n.d.). Amazon Web Services, Inc. Retrieved March 8, (2026) from https://aws.amazon.com/what-is/retrieval-augmented-generation/ Winstone NE, Nash RA, Parker M, Rowntree J (2017) Supporting Learners’ Agentic Engagement With Feedback: A Systematic Review and a Taxonomy of Recipience Processes. Educational Psychol 52(1):17–37. https://doi.org/10.1080/00461520.2016.1207538 Winstone NE, Nash,Robert A, Rowntree, James,and, Parker M (2017) ‘It’d be useful, but I wouldn’t use it’: Barriers to university students’ feedback seeking and recipience. Studies in Higher Education , 42 (11), 2026–2041. https://doi.org/10.1080/03075079.2015.1130032 Yeh SS (2010) Understanding and addressing the achievement gap through individualized instruction and formative assessment. Assess Education: Principles Policy Pract 17(2):169–182. https://doi.org/10.1080/09695941003694466 Zhu M, Liu OL, Lee H-S (2020) The effect of automated feedback on revision behavior and learning gains in formative assessment of scientific argument writing. Comput Educ 143:103668. https://doi.org/10.1016/j.compedu.2019.103668 Additional Declarations The authors declare no competing interests. Supplementary Files SupplementalMaterials.docx Cite Share Download PDF Status: Posted Version 1 posted 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-9271072","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":617191576,"identity":"95d375df-214b-42b9-9d4b-c553d79e8652","order_by":0,"name":"Michele Weston","email":"","orcid":"https://orcid.org/0009-0007-7810-938X","institution":"Michigan State University","correspondingAuthor":false,"prefix":"","firstName":"Michele","middleName":"","lastName":"Weston","suffix":""},{"id":617192219,"identity":"47cab349-f5c7-4fb5-8452-7d9ff76b79e4","order_by":1,"name":"Tammy Long","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAArElEQVRIiWNgGAWjYBACPghlA6F4iNHCBiIOMKSRruUwKVrYew+//lBzPlp+RgLjg7dtxGjhOZdmceDY7dwNNxKYDecSpUUix8zgABtQi0QCmzQv8Vr+ncudPyOB/TexWowfHGw7kNtwI4GNmTgtPGfMGM72JeduOPOwWXLOOSK08LP3GH+o+GaXO789+eCHN2VEaAG7DUIzNhCnHgiYPxCtdBSMglEwCkYmAAByHjb6XTy0VwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-0849-9056","institution":"Michigan State University","correspondingAuthor":true,"prefix":"","firstName":"Tammy","middleName":"","lastName":"Long","suffix":""}],"badges":[],"createdAt":"2026-03-30 18:28:26","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9271072/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9271072/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106403994,"identity":"3e83f3e3-024f-4596-9e0f-9ee59751898a","added_by":"auto","created_at":"2026-04-08 09:15:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":620820,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9271072/v1/9224688b-bcd7-4b4c-8640-39d8e694db6e.pdf"},{"id":106201791,"identity":"1f0b65b1-379d-4544-8c50-d78dce4227c3","added_by":"auto","created_at":"2026-04-06 03:30:57","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":30245,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-9271072/v1/64851917f953d7ac871a50e8.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eHow Do Biology Undergraduates Use AI-enabled Feedback to Revise Scientific Writing?\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eFeedback on learning tasks is understood to be one of the most influential factors contributing to student achievement (Black \u0026amp; Wiliam, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Hattie, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e1999\u003c/span\u003e). Modern conceptualizations of feedback consider it an active, learner-centered process in which students use an input (e.g. written evaluative comments) to inform changes to an output (e.g. subsequent work; Henderson et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Sadler, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). This view places students\u0026rsquo; ability to understand, interpret, and act on feedback at the center of effective learning. Accordingly, recent research has emphasized students\u0026rsquo; perspectives to better understand how they perceive and make use of feedback (e.g. Dawson et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Jonsson, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2013\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eStudents\u0026rsquo; engagement with feedback on writing varies across contexts: in some cases they do not read or use it (Hyland, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; MacDonald, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1991\u003c/span\u003e), whereas in others, they incorporate it into their revisions (Higgins et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Winstone, Nash, et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Yet not all revisions lead to meaningful improvement; students often focus on changes to spelling, grammar, and unnecessary details rather than substantive refinements (Ashwell, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Work examining student perceptions offers insight into why it can be challenging to use feedback. Winstone et al. (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) found that students often struggle to translate feedback comments into next steps. Research in this area has identified several barriers that contribute to these difficulties: students may be unable to decode what the feedback means, lack strategies for implementing it, feel they do not have the agency to revise, or lack the volition to invest effort in improving their work (Jonsson \u0026amp; Panadero, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Nicol \u0026amp; Macfarlane-Dick, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Sutton, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Winstone et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Unsurprisingly, students\u0026rsquo; perceptions of what makes feedback useful aligns closely with overcoming these barriers. They prefer feedback that is personalized, detailed, and actionable (Dawson et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), as personalization and detail support comprehension while actionable guidance scaffolds next steps\u0026mdash;promoting both agency and willingness to revise (Dawson et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Higgins et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2002\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eInstructor feedback plays a crucial role in supporting students\u0026rsquo; scientific writing, such as lab reports and research proposals, particularly when there are opportunities to reflect and revise (Jerde \u0026amp; Taper, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Yeh, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). For instance, Weaver et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) examined an undergraduate organic chemistry lab course in which students engaged in iterative cycles of revision to lab reports informed by peer and instructor feedback. Students who participated in this redesigned curriculum earned significantly higher lab report grades than students in previous semesters and attested to substantial improvements in their writing skills. Despite these benefits, many instructors face time and workload constraints that limit their ability to provide personalized feedback (Heim \u0026amp; Holt, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). As a result, students may receive minimal feedback\u0026mdash;or none at all.\u003c/p\u003e\n\u003ch3\u003eAI-enabled Feedback\u003c/h3\u003e\n\u003cp\u003eGenerative AI (genAI) is emerging as a promising tool for assisting instructors in developing formative feedback on students\u0026rsquo; scientific writing (Pellegrino, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). GenAI refers to artificial intelligence (AI) systems capable of producing original content, such as written text, in response to a user prompting (Stryker \u0026amp; Scapicchio, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Compared to traditional machine learning (ML) approaches, genAI systems offer advantages for feedback generation: their natural-language prompting makes them easy to interact with, and their pretrained large language models allow them to perform tasks without requiring the large, scored datasets typically needed for supervised ML systems. These affordances create a low barrier to entry and position genAI to be adopted more quickly and widely by instructors.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePromoting Quality AI-enabled Feedback\u003c/h2\u003e \u003cp\u003eAs genAI becomes more widely applied in educational settings, it is important to foster ethical use by intentionally aligning AI implementations with instructional goals and values. Human-centered AI offers one such approach by emphasizing design practices that augment rather than replace human capabilities and that maintain human oversight of AI systems (Schmager et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). We define AI-enabled feedback as that which is developed within a human-centered AI process. Keeping instructors actively involved in the feedback process helps ensure higher-quality output and preserves accountability for assessment. In practice, human involvement may include configuring and testing an assignment-specific AI system, as well as reviewing and refining AI output before sharing it with students. A related strategy for promoting quality AI-enabled feedback involves customizing AI systems to perform the specific task reflecting the course context, learning goals, and assessment criteria. Such customization enhances the accuracy of the feedback and reduces hallucinations, which are instances in which the model produces inaccurate yet plausible text (Mills et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; OpenAI et al., 2024).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eUnderstanding GenAI’s Impact on Learning\u003c/h3\u003e\n\u003cp\u003eGenAI tools are increasingly used by instructors to automate, augment, and expand their feedback practices, yet we currently lack evidence about how students use AI-enabled feedback (Cooper, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). This gap is consequential because students must actively interpret and use feedback to benefit from it. It is unclear whether existing understandings about feedback processes similarly apply to AI-enabled feedback (Kasneci et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). One emerging application of genAI is the development of feedback on students\u0026rsquo; scientific writing. In this study, we provide empirical evidence of how undergraduate biology students used AI-enabled feedback to revise their research proposals.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eWe used a qualitative, multiple case study design with a comparative document analysis to investigate students\u0026rsquo; engagement with feedback during the revision process. Multiple case studies examine a phenomenon within bounded contexts, or cases, and take a holistic approach to understanding the complex and situated experiences of participants (Merriam, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Miles \u0026amp; Huberman, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e1994\u003c/span\u003e). This approach is well suited for developing and presenting insights that inform our understanding of student use of AI-enabled feedback. The study includes two cases, each consisting of a team of three students.\u003c/p\u003e\n\u003ch3\u003eStudy Context\u003c/h3\u003e\n\u003cp\u003eThis study was conducted within a STEM-focused residential college at a large research university located in the Midwest. The research took place in the laboratory component of a combined lecture and laboratory course on introductory organismal biology. The laboratory curriculum was designed to explore systemic inequities and human impacts on ecology and health. The first part of the course introduced students to different ecological systems, methods, and types of data analysis. Following this, student teams developed independent research projects investigating the underlying cause of a human health and ecological disparity of their choice. Although research projects often utilize final papers resembling journal articles, in this course each team was required to incorporate their results into a written research proposal as preliminary findings that will inform a new project. Although the original intention was for students to complete full research proposals, time constraints required the assignment to be shortened to the introduction section only. Therefore, this report will refer to the piece of writing as the \u0026ldquo;proposal introduction\u0026rdquo;. Students were instructed to follow a rubric that specified the content of four required paragraphs, each focusing on a different scale of human impact. Descriptions of the paragraphs are shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eParticipants\u003c/h3\u003e\n\u003cp\u003eAll students enrolled in the laboratory course were invited to participate in the research study. There were two student teams where all members consented to participate, and they were each considered a case. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e provides demographic and academic information for these participants.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e\u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eContent of Proposal Introduction Paragraphs\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParagraph Name\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription of Content\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMacro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThis is the first paragraph of the proposal introduction. Describe the problem at the macro (historical and societal) scale. Explain how historical practices (e.g. redlining) result in current racial disparities in health outcomes (e.g., asthma rates) and local ecosystems (e.g., local canopy cover and heat islands).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeso Part 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThis is the second paragraph of the proposal introduction. Describe the problem at the meso (community) scale. Explain the impacts of specific policies or practices (e.g. urban renewal, highway construction) on local resources, community health, and\u003c/p\u003e \u003cp\u003eenvironmental conditions, focusing on neighborhood structures and resources.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeso Part 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThis is the third paragraph of the proposal introduction. Describe one or more community responses to inequity that are relevant to the experimental question and variables.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMicro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThis is the fourth paragraph of the proposal introduction. Describe the problem at the micro (individual) scale. Explain how systemic inequities affect personal health outcomes and ecological interactions, highlighting specific cases and personal experiences (e.g., personal stories of residents in affected areas).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e\u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eStudent Participant Demographics\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParticipant \u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTeam Number\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRace(s) and Ethnicity/ies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFinal Course\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCumulative GPA \u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWoman\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWhite\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWhite\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWoman\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSouth Asian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBlack or African American\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.59\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWoman\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBlack or African American\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.44\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudent 6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWoman\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWhite\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003csup\u003ea\u003c/sup\u003e Student participants are listed without pseudonyms to prevent the association of demographic data with the comments of individuals. \u003csup\u003eb\u003c/sup\u003e Final course grade is reported on a 4.0 scale. \u003csup\u003ec\u003c/sup\u003e Cumulative GPA is reported on a 4.0 scale and at the start of the semester in which the study took place.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eProcedures\u003c/h2\u003e \u003cp\u003eStudent teams began the assignment by drafting the proposal introduction and submitting it to their instructor. The instructor then used a genAI tool (described below) to develop customized feedback aligned with the assignment rubric, reviewed the feedback for accuracy and quality, and made edits as necessary before providing it to students. The instructor had the option of using follow-up prompts to refine the GPTs\u0026rsquo; output but opted not to do so. Students then submitted the revised proposal introduction as the final draft.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eGenerative AI Tool\u003c/h3\u003e\n\u003cp\u003eThis study utilized a GPT, which is a specialized version of ChatGPT developed to carry out a specific task (\u003cem\u003eIntroducing GPTs\u003c/em\u003e, 2024). Although it appeared similar to the standard ChatGPT interface where the instructor interacted with a chatbot through a chat window, this GPT (model 4o; OpenAI, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) utilized detailed instructions and a curated knowledge base, allowing the AI to query course-relevant documents or data and integrate these materials into its responses, which reduces hallucination by grounding the model\u0026rsquo;s predictions (\u003cem\u003eIntroducing GPTs\u003c/em\u003e, 2024; \u003cem\u003eWhat Is RAG?\u003c/em\u003e, n.d.). Importantly, GPTs retain their protocols and reference materials across sessions, allowing the user to repeat the specialized task without re-entering instructions. Additional details about the GPT are provided in the Supplemental Materials.\u003c/p\u003e \u003cp\u003eHaving outlined the tool\u0026rsquo;s general features, we now describe how this GPT was intentionally grounded in learning theory and course-specific supports. The lab coordinator of the course and the first author collaborated to design the research proposal assignment and its rubric, ensuring alignment with the coordinator\u0026rsquo;s pedagogical priorities. The rubric incorporated the Claim, Evidence, and Reasoning (CER) framework, which had been used throughout the course to support students\u0026rsquo; scientific explanations (McNeill \u0026amp; Krajcik, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). To further guide students, they developed a scaffolding tool that outlined the structure of the proposal introduction and specified the content expected in each paragraph. Building on this foundation, the first author configured and tested the GPT to provide feedback tailored to the rubric, CER framework, and scaffolding tool. The GPT was instructed to give an output consisting of 1) the proposal\u0026rsquo;s strengths, 2) suggestions for improvement, and 3) examples to scaffold the suggestions for improvement for each rubric section as described in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Student data was protected by collecting written consent for the use of genAI in analyzing student work, anonymizing proposals prior to use of the GPT, and by ensuring that these proposals were excluded from the GPT training dataset.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e\u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRubric and Feedback Sections\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeedback Section\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eScope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMacro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eParagraph-specific\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeedback pertaining to the content of the Macro paragraph.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eParagraph-specific\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeedback pertaining to the content of both the Meso Part 1 and Meso Part 2 paragraphs.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMicro\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eParagraph-specific\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeedback pertaining to the content of the Micro paragraph.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlow of Writing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWhole-proposal introduction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeedback on how effectively each sentence and paragraph transitions to the next, with a clear flow of ideas.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClaims are Supported\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWhole-proposal introduction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeedback on how well claims are supported by scientific reasoning and evidence from credible cited sources.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003e\u003cem\u003eNote.\u003c/em\u003e Feedback in each section followed the standardized format of strengths, suggestions for improvement, and scaffolding tips.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eData Analysis\u003c/h2\u003e \u003cp\u003eWe conducted a two-part revision analysis: first, we compared the first and final drafts to identify and categorize changes; then, we cross-referenced those changes with the AI-enabled feedback to examine how students revised in response to it. We applied an adapted revision operations coding scheme (Faigley \u0026amp; Witte, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1981\u003c/span\u003e) to the drafts to capture the meaning-based changes as additions, deletions, moves, and substitutions. A change was considered meaning-based if it involved conceptual ideas or information. Additions and deletions of citations were coded separately to distinguish them from meaning-based changes. Surface-level revisions were excluded from analyses and included changes that corrected spelling or grammar, or edits that influenced the implicit or explicit nature of an idea without changing the idea itself. When a revision was longer than a single sentence, we coded each sentence separately. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e in the Supplemental Materials presents illustrative examples of the revision operations coding. One exception occurred with Team 1, whose first draft was missing one of the required paragraphs. Because there was no original text to revise, the corresponding paragraph in the final draft was not coded for revisions. This ensured that the revision analysis focused only on changes to existing content.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003eWhile the initial coding captured the types of changes students made, it did not indicate whether they addressed the feedback provided. To assess this, we identified draft segments corresponding to each feedback comment and categorized the comments as fully, partially, or not addressed, based on the extent to which suggested changes appeared in the revised draft. The analysis focused only on feedback that offered a suggestion for improvement or a scaffolding tip (e.g., examples of how to improve). Feedback highlighting strengths were excluded because they did not require revision. We conducted all coding and analysis in MaxQDA (VERBI Software, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Table \u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e in the Supplemental Materials illustrates examples of feedback comments, their categorization, and associated excerpts from student drafts.\u003c/div\u003e\n\u003c/div\u003e"},{"header":"Analysis","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003cp\u003eThis section begins with the results of the revision analysis including students\u0026rsquo; use of feedback comments. That is followed by within-case analyses for each team to illustrate and contextualize the results of the revision analysis, and a cross-case analysis synthesizing the findings. The section ends with theoretical interpretation of the findings.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eRevisions Analysis\u003c/h2\u003e\n \u003cp\u003eWe conducted a two-part revision analysis to examine how students used AI-enabled feedback to revise their writing. First, we compared each team\u0026rsquo;s first and final drafts to identify and categorize fine-grained differences. We then cross-referenced the results with the AI-enabled feedback to determine the extent to which comments were addressed in the revised drafts. Below, we first summarize the revision operations used by each team and then describe the extent to which revisions addressed feedback comments.\u003c/p\u003e\n \u003cp\u003eAcross both teams, all four types of meaning-based revisions were observed. Deletions were the most frequent (15 of 31 changes), followed by additions (10 of 31 changes). Team 1 made more revisions overall, despite their first draft missing a required paragraph. The teams also differed in their distribution of meaning-based changes: Team 1 primarily deleted content, whereas Team 2 made more additions than deletions. Notably, both teams removed citations from their first drafts and did not introduce any new citations in their final drafts. Table \u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e in the Supplemental Materials presents a detailed breakdown of revision operations used by each team.\u003c/p\u003e\n \u003cp\u003eTo examine how student teams incorporated AI-enabled feedback into their revisions, we classified each feedback comment as fully, partially, or not addressed, based on the extent to which suggested changes were reflected in the revised draft. Because Team 1\u0026rsquo;s initial draft was missing a required paragraph, they had fewer feedback comments included in the analysis. Across both teams, 10 of the 22 comments were fully or partially addressed, leaving more than half unaddressed. The distribution of comments across the categories fully, partially, and not addressed was similar for both teams and consistent across comment type (suggestions for improvement vs. scaffolding tips). Table \u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e in the Supplemental Materials presents the distribution of comments across categories of feedback use.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eTeam 1 Within-Case Analysis\u003c/h2\u003e\n \u003cp\u003eTeam 1 investigated the question, \u0026ldquo;In previously redlined communities, what is the relationship between tree density and median income, asthma frequency, and UV exposure?\u0026rdquo; In addition to examining the effects of redlining, their proposal introduction incorporated the construction of a nearby highway as a historically linked event with ongoing environmental and health consequences. They also described a local garden program as a community response initiative that could mitigate the harms perpetuated by redlining and the highway project. As noted earlier, Team 1 did not include a Micro paragraph in their first draft, and because the Micro paragraph in their final draft was newly written, it is excluded from this within-case analysis.\u003c/p\u003e\n \u003cp\u003eOverall, as the following analysis shows, Team 1 responded to AI-enabled feedback by improving conciseness and organization more than enhancing substantive content. Although feedback comments consistently requested additional information, explanation, reasoning, and evidence, Team 1 made only one addition to their final draft. This was a transition sentence that fully addressed the feedback comment suggesting they \u0026ldquo;\u0026hellip;add a transition sentence between the Meso and Micro paragraphs to set up the shift from community impacts to individual outcomes.\u0026rdquo; In contrast, feedback comments that required adding new explanations or information largely went unaddressed. For instance, Team 1 did not address comments that asked them to 1) explain ways the local pattern of redlining compared to other cities, 2) explain how the community response initiative might relate to the study\u0026rsquo;s variables, or 3) better support claims with scientific reasoning or evidence, particularly those around the benefits of the community response initiative.\u003c/p\u003e\n \u003cp\u003eInstead, the team most often responded to AI-enabled feedback through structural rather than conceptual revisions. Collectively, three comments on the Macro paragraph had encouraged them to trace a timeline of consequences that would \u0026ldquo;help clarify \u003cem\u003ehow\u003c/em\u003e specific national practices led to systemic racial inequalities in housing and health that persist today\u0026rdquo;, beginning with the policies that institutionalized redlining. In response, Team 1 moved information about the highway project and its enduring impacts from the Meso Part 1 paragraph into the Macro paragraph. This shift partially addressed the feedback by extending the list of consequences. However, it did not conceptually connect the highway project to redlining nor elaborate on the policies that institutionalized redlining. In other cases, Team 1 responded to feedback by removing text. For example, feedback advised that \u0026ldquo;Some claims\u0026mdash;particularly around individual health outcomes and the benefits of the Garden Program\u0026mdash;could be better supported with scientific reasoning or specific evidence\u0026rdquo;. The students responded to this comment indirectly by deleting an unsupported claim about the garden program in the Meso Part 2 paragraph. While this removal can be understood as an attempt to resolve the feedback by avoiding a weak claim, it did not involve adding reasoning or evidence. As such, the feedback was categorized as unaddressed.\u003c/p\u003e\n \u003cp\u003eTeam 1\u0026rsquo;s revision pattern also reflected a selective uptake of AI-enabled feedback, with many deletions aimed at tightening the writing rather than expanding the content as comments had encouraged. In fact, Team 1 made so many deletions that the word count shrunk from 465 words in the first draft to 292 words in the final draft. As an example, the Macro paragraph originally contained two redundant explanations of redlining. The team removed the first explanation by deleting three text segments, leaving behind a sufficient introduction to redlining. This pattern extended across the proposal, as other deletions cut out unnecessary or redundant details, improving clarity and conciseness. In contrast, two of the team\u0026rsquo;s deletions of text from the Meso Part 1 paragraph removed details that were important for meeting rubric requirements. The rubric asked students to describe events that led to the perpetuation and exacerbation of inequities in ecosystems and human health. However, the team removed an explanation that the area\u0026apos;s historically lower tree density had resulted from redlining and disinvestment, and that economic barriers compounded to make it difficult for residents to mitigate harmful environmental effects. Similarly, the team deleted 5 citations that were essential for supporting claims. These examples show how the Team\u0026rsquo;s selective uptake of feedback is complex and involved competing goals for the revisions.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003eTeam 2 Within-Case Analysis\u003c/h2\u003e\n \u003cp\u003eTeam 2 investigated the question, \u0026ldquo;How has redlining affected community garden quality and prevalence in the greater [Midwestern city] area and how is it tied to obesity?\u0026rdquo; Their proposal focuses on redlining and positioned obesity as a long-term health outcome tied to it. To contextualize the relationship, they examined how poor soil quality and limited access to grocery stores in historically redlined areas contribute to elevated obesity rates. The team also incorporated community gardens that donate produce to a local food bank as a response initiative aimed at reducing environmental and health disparities.\u003c/p\u003e\n \u003cp\u003eThe following analysis demonstrates that Team 2 responded to AI-enabled feedback by selectively expanding the proposal\u0026rsquo;s substantive content. One way Team 2 responded to AI-enabled feedback was by adding the explanations and details that comments requested. For example, feedback on the Meso paragraphs suggested they provide more description of the local food bank to clarify its role as a community response initiative, as the first draft had referenced it without supporting details. The team fully addressed the comments by adding a description that the food bank \u0026ldquo;aims to provide food pantries in the area with fresh produce. The [local food bank] is a non-profit organization that works with the community to provide food to those in need.\u0026rdquo; Team 2 also incorporated a small addition in their Macro paragraph in response to feedback asking them to \u0026ldquo;\u003cstrong\u003eexplain how these policies caused long-term economic and health disparities.\u003c/strong\u003e The connection between historical redlining and present-day diabetes is mentioned, but \u003cstrong\u003ethe causal chain could be more fully explained.\u0026rdquo;\u003c/strong\u003e (Team 2 originally wrote about diabetes as another outcome variable before narrowing the scope to only obesity). They added the qualifier \u0026ldquo;limited\u0026rdquo; to describe healthcare access in the sentence, \u0026ldquo;This segregation has had lasting effects, contributing to health inequities like higher obesity and diabetes rates due to systemic barriers such as limited healthcare access and economic hardship.\u0026rdquo; Although their argument remained similar across drafts, the revision clarified the causal mechanism by specifying that healthcare access was less available in redlined areas.\u003c/p\u003e\n \u003cp\u003eIn contrast, Team 2 also made an addition that did not meaningfully address the feedback comment or help meet rubric criteria. In the Macro paragraph, they added the sentence, \u0026ldquo;Red and yellow areas, home to low-income Black residents, faced mortgage discrimination, while blue and green zones housed wealthier white populations.\u0026rdquo; While this elaboration identifies the color labels used in redlining maps, it does not explain how redlining\u0026rsquo;s short- and long-term consequences connect to present-day inequities, nor or as another feedback comment requested, does it provide the context around the unique history of redlining in the specific Midwestern city compared to other cities. It is notable that the team incorporated this addition immediately after applying a scaffolding tip that provided a suggested sentence starter (\u0026ldquo;In the 1930s, redlining was codified through federal housing policy\u0026hellip;\u0026rdquo;). Their revision reads, \u0026ldquo;In the 1930s, the Home Owners\u0026rsquo; Loan Corporation (HOLC) and the Federal Housing Administration (FHA) institutionalized segregation through redlining, classifying neighborhoods into four color-coded zones. Red and yellow areas, \u0026hellip;\u0026rdquo; A prior feedback comment also mentioned HOLC and FHA as historical policies, and it appears that the students found prescriptive or highly structured feedback easier to follow, whereas more open-ended feedback made it difficult for them to identify what substantive information to include next.\u003c/p\u003e\n \u003cp\u003eAnother way Team 2 selectively responded to AI-enabled feedback was by focusing on paragraph-specific comments over broader suggestions aimed at improving the proposal\u0026rsquo;s flow and strength of arguments. Team 2 fully or partially addressed 6 of the 9 paragraph-specific comments, but none of the 4 whole-proposal comments. For example, a suggestion in the \u0026ldquo;Flow of writing\u0026rdquo; section noted that, \u0026ldquo;some paragraphs jump between topics (e.g., soil quality to obesity to garden location) without clear transitions. Strengthen the flow by using topic sentences that guide the reader and connect ideas more smoothly.\u0026rdquo; None of the team\u0026rsquo;s revisions improved transitions, so this feedback was categorized as unaddressed. Similarly, the \u0026ldquo;Claims are supported\u0026rdquo; section advised strengthening several claims with citations or scientific explanation and identified the link between worms, soil quality, and hunger as an example. Because none of the final draft\u0026rsquo;s added content provided evidence or reasoning\u0026mdash;and because four citations from the first draft were deleted\u0026mdash;these comments also remained unaddressed.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003eCross-Case Analysis\u003c/h2\u003e\n \u003cp\u003eA comparison of the two within-case analyses shows similarities and differences in how students used AI-enabled feedback to revise their research proposals. The teams\u0026rsquo; responses indicate that AI-enabled feedback can prompt different types of revision work: Team 1 primarily used it to refine structure through text deletions and moves, while Team 2 used it to elaborate content via additions. Despite these differences in revision strategies, the distribution of feedback comments across the categories fully, partially, and not addressed was similar for both teams, suggesting comparable levels of uptake. This similarity in uptake reflects neither team consistently incorporating the conceptual information or reasoning that the feedback most often requested. Although Team 1\u0026rsquo;s structural revisions occasionally aligned with feedback\u0026mdash;such as by moving information to a location that needed elaboration or removing unsupported claims\u0026mdash;these changes did not expand the substantive content the feedback sought. Similarly, while Team 2 prioritized additions, they left many open-ended or conceptually demanding suggestions unaddressed. The remainder of the cross-case analysis describes the types of feedback that had limited use by both teams and offers possible explanations for the patterns observed.\u003c/p\u003e\n \u003cp\u003eA key source of limited uptake for both teams involved open-ended feedback that required identifying new information or conducting additional research. Across cases, comments that encouraged students to add location-specific historical details or to provide conceptual connections, scientific reasoning, or evidence were typically only partially addressed or left unaddressed. This pattern was especially apparent for Team 1, whose only addition was a transition sentence responding to a relatively prescriptive suggestion. Team 2 added new details\u0026mdash;such as a description of the local food bank and the qualifier \u0026ldquo;limited\u0026rdquo; in their explanation of healthcare access\u0026mdash; but these additions involved general knowledge that may not have required searching for external sources. For both teams, revisions that partially addressed feedback tended to meet the surface-level or easily actionable portion of a comment while omitting the more demanding element that required new evidence or conceptual elaboration. For example, Team 1 moved their description of the highway project into the Macro paragraph, which extended the list of consequences but did not include the causal explanation requested in the feedback (e.g., linking the highway project to redlining or describing the policies that institutionalized it). Incorporating new information into the proposal would have required multiple steps such as finding appropriate sources, assessing their relevance and quality, and integrating and citing them, and neither team demonstrated this type of engagement with the feedback. We interpret these missing or incomplete revisions as evidence that students preferred working with existing material over conducting additional research, suggesting either difficulty researching requested information or a reluctance to do so.\u003c/p\u003e\n \u003cp\u003eBoth teams also made limited use of feedback that required self‑evaluation, whether judging the coherence of the proposal, evaluating claims, or recognizing gaps in reasoning. Except for Team 1\u0026rsquo;s new transition sentence, neither team addressed the feedback in the \u0026lsquo;Flow of Writing\u0026rsquo; or \u0026lsquo;Claims Are Supported\u0026rsquo; sections, which applied to the whole proposal. This suggests difficulty responding to whole-proposal comments, although the assignment\u0026rsquo;s design may have contributed; each student was responsible for writing one or two paragraphs, and they may have focused only on the feedback corresponding to paragraphs they authored. However, each whole-proposal suggestion included examples tied to specific paragraphs, and students still did not revise accordingly. The scaffolding tips in the \u0026lsquo;Claims are Supported\u0026rsquo; section further illustrate the self-evaluative work expected of students. These tips directed them to examine their claims across the proposal and ask questions such as, \u0026ldquo;What\u0026rsquo;s the evidence?\u0026rdquo; and \u0026ldquo;Why does the evidence support the claim?\u0026rdquo; Neither team added support for claims, suggesting difficulty evaluating the adequacy of their reasoning in addition to the reluctance to research new information described previously. This difficulty identifying weaknesses in their explanations extended beyond the \u0026lsquo;Claims Are Supported\u0026rsquo; section. As described in the previous paragraph, both teams often addressed only the easily actionable portions of feedback while neglecting the more conceptually demanding aspects, such as explaining the causal chain linking redlining to present-day inequities in the environment and human health. We interpret this as an indication that students struggled not only to seek out new information but also to recognize where conceptual gaps existed in their arguments, and as a result were unsuccessful in addressing the AI-enabled feedback.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003eTheoretical Interpretation\u003c/h2\u003e\n \u003cp\u003eThis study adopted a learner‑centered perspective on feedback, emphasizing that students must actively use feedback\u0026mdash;not merely receive it\u0026mdash;to benefit from it (Sadler, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). From this standpoint, we examined the teams\u0026rsquo; observable revision behaviors to understand how they used AI-enabled feedback. Both teams engaged in meaning-based revisions, and each addressed roughly half of the feedback comments. This suggests that many of the AI-enabled comments prompted actionable revision. Given how often students struggle to take up feedback in general (Winstone, Nash, et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), the degree of engagement observed here is reassuring. These findings offer insight into the behavioral side of students\u0026rsquo; experiences with AI-enabled feedback by demonstrating how it shaped their revisions in practice.\u003c/p\u003e\n \u003cp\u003eEven so, the partial uptake across teams indicates that students did not consistently incorporate the additional information or reasoning emphasized in the feedback. Even Team 2, who made more additions overall, left many open-ended or conceptually demanding suggestions unaddressed. Both teams made limited use of feedback that 1) requested new information, reasoning, or evidence, or 2) required self-evaluation. The difficulties that these teams faced align with Winstone et al.\u0026rsquo;s (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) framework of barriers to feedback recipience, including difficulty decoding discipline-specific language, limited strategies for implementing feedback, and challenges translating comments into action. They may also experience apathy or lack the volition to take action. The challenges both teams faced here mirror well-established difficulties with using traditional feedback, suggesting that the barriers shaping feedback uptake may extend to AI-enabled feedback as well.\u003c/p\u003e\n \u003cp\u003eIn addition to the cognitive and affective processes described above, characteristics of the AI-enabled feedback itself may have shaped how students used it. Prior work shows that students find feedback useful when it is personalized, detailed, and actionable (Dawson et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Studies of AI-enabled or -generated feedback on scientific writing similarly indicate that students perceive such feedback as useful, detailed, and actionable (Dai et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Mills et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, this is not always the case. For example, Lim and Go (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) reported that students found AI-generated feedback on essays to be overly broad, generic, and vague. Since Lim and Go relied on a standard ChatGPT prompt rather than a customized system with additional instructions and a domain-specific knowledge base, the feedback they obtained largely consisted of generalized comments resembling grading criteria. Differences in students\u0026rsquo; perceptions across studies may reflect variation in AI tools and prompting approaches. Our findings contribute to this discussion by illustrating that a customized AI tool, grounded in learning theory and integrated with course‑level supports, can produce feedback that students are able to act on when revising their scientific writing.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eInstructors and education researchers are increasingly developing feedback with the assistance of genAI, yet we still know relatively little about how students use this feedback (Cooper, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). This makes it unclear whether AI-enabled feedback supports learning or whether existing understandings about feedback processes extend to this new context. To address this, we examined how two teams of undergraduate biology students used AI-enabled feedback to revise their research proposal introductions. Our findings show that AI-enabled feedback prompted actionable revision across both teams. Each team addressed approximately half of the feedback comments, indicating comparable levels of uptake despite their differing revision approaches. While feedback comments consistently requested additional information, reasoning, and evidence, both teams commonly left open-ended and conceptually demanding suggestions unaddressed. This pattern indicates that both teams were inconsistent in incorporating the additional information and reasoning highlighted in the feedback, reflecting challenges similar to those commonly observed with traditional feedback. These findings provide behavioral evidence of how students used AI-enabled feedback to revise their scientific writing and demonstrates ways that prior research on feedback use and barriers can extend to AI-enabled feedback.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor Note\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMichele Weston \u003ca href=\"https://orcid.org/0009-0007-7810-938X\" target=\"orcid.widget\"\u003e\u003cimg width=\"16\" height=\"16\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAMAAAAoLQ9TAAAALVBMVEUAAACmzjmmzjmmzjmmzjmmzjn////0+ebj8MHe7bXT55y82mus0UWmzjmSwCmNJ4LqAAAABnRSTlMAIGC/z+8mlFLTAAAAYElEQVR42m2PWw7AIAgEEdA+l/sft4Km1tj5IUwIsOSwoCJMjaToaIo+o5yGILtRDAElYgBmLm43TIKYKJW9KiF0Ybi2A8BH4CqT8Po7IfMOIZ6vcDw20Pb6S05ruCX+A9I+CQ+hA8ZPAAAAAElFTkSuQmCC\" v:shapes=\"Picture_x0020_2\" alt=\"image\"\u003e\u003c/a\u003ehttps://orcid.org/0009-0007-7810-938X\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTammy Long \u003cimg width=\"16\" height=\"16\" src=\"data:image/jpeg;base64,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\" v:shapes=\"Picture_x0020_11\" alt=\"image\"\u003ehttps://orcid.org/0000-0003-0849-9056\u003c/p\u003e\n\u003cp\u003eWe have no conflict of interest to disclose.\u003c/p\u003e\n\u003cp\u003eThis material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant 2235783, the SimBio Foundation grant program, and the Plant Biology Department of Michigan State University.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCorrespondence concerning this article should be addressed to Tammy Long, Plant Biology Laboratories, 612 Wilson Rd, East Lansing, MI 48824. Email: [email protected] \u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAshwell T (2000) Patterns of Teacher Response to Student Writing in a Multiple-Draft Composition Classroom: Is Content Feedback Followed by Form Feedback the Best Method? J Second Lang Writ 9(3):227\u0026ndash;257. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1060-3743(00)00027-8\u003c/span\u003e\u003cspan address=\"10.1016/S1060-3743(00)00027-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlack P, Wiliam D (1998) Assessment and Classroom Learning. Assess Education: Principles Policy Pract 5(1):7\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/0969595980050102\u003c/span\u003e\u003cspan address=\"10.1080/0969595980050102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCooper MM (2026) A Preliminary Set of Principles to Support Learning in the Context of Generative AI. J Chem Educ 103(1):36\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/acs.jchemed.5c01413\u003c/span\u003e\u003cspan address=\"10.1021/acs.jchemed.5c01413\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDai W, Lin J, Jin H, Li T, Tsai Y-S, Gašević D, Chen G (2023) Can Large Language Models Provide Feedback to Students? A Case Study on ChatGPT. \u003cem\u003e2023 IEEE International Conference on Advanced Learning Technologies (ICALT)\u003c/em\u003e, 323\u0026ndash;325. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ICALT58122.2023.00100\u003c/span\u003e\u003cspan address=\"10.1109/ICALT58122.2023.00100\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDawson P, Henderson,Michael M, Paige, Phillips, Michael, Ryan, Tracii, Boud, David,and, Molloy E (2019) What makes for effective feedback: Staff and student perspectives. \u003cem\u003eAssessment \u0026amp; Evaluation in Higher Education\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e(1), 25\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602938.2018.1467877\u003c/span\u003e\u003cspan address=\"10.1080/02602938.2018.1467877\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaigley L, Witte S (1981) Analyzing Revision. Coll Composition Communication 32(4):400\u0026ndash;414. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/356602\u003c/span\u003e\u003cspan address=\"10.2307/356602\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHattie J (1999), August 2 \u003cem\u003eInfluences on student learning\u003c/em\u003e [Inaugural lecture]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeim AB, Holt EA (2019) Benefits and Challenges of Instructing Introductory Biology Course-Based Undergraduate Research Experiences (CUREs) as Perceived by Graduate Teaching Assistants. CBE\u0026mdash;Life Sci Educ 18(3):ar43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1187/cbe.18-09-0193\u003c/span\u003e\u003cspan address=\"10.1187/cbe.18-09-0193\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHenderson M, Ajjawi R, Boud D, Molloy E (2019) Identifying Feedback That Has Impact. In M. Henderson, R. Ajjawi, D. Boud, \u0026amp; E. Molloy (Eds.), \u003cem\u003eThe Impact of Feedback in Higher Education: Improving Assessment Outcomes for Learners\u003c/em\u003e (pp. 15\u0026ndash;34). Springer International Publishing. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-030-25112-3_2\u003c/span\u003e\u003cspan address=\"10.1007/978-3-030-25112-3_2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiggins R, Hartley P, Skelton A (2002) The Conscientious Consumer: Reconsidering the role of assessment feedback in student learning. Stud High Educ 27(1):53\u0026ndash;64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03075070120099368\u003c/span\u003e\u003cspan address=\"10.1080/03075070120099368\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHyland F (1998) The impact of teacher written feedback on individual writers. J Second Lang Writ 7(3):255\u0026ndash;286. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1060-3743(98)90017-0\u003c/span\u003e\u003cspan address=\"10.1016/S1060-3743(98)90017-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u003cem\u003eIntroducing GPTs\u003c/em\u003e. (2024), March 13 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://openai.com/index/introducing-gpts/\u003c/span\u003e\u003cspan address=\"https://openai.com/index/introducing-gpts/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJerde CL, Taper ML (2004) Preparing Undergraduates for Professional Writing: Evidence Supporting the Benefits of Scientific Writing within the Biology Curriculum. J Coll Sci Teach 33(7):34\u0026ndash;37\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJonsson A (2013) Facilitating productive use of feedback in higher education. Act Learn High Educ 14(1):63\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/1469787412467125\u003c/span\u003e\u003cspan address=\"10.1177/1469787412467125\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJonsson A, Panadero E (2018) Facilitating Students\u0026rsquo; Active Engagement with Feedback. In A. A. Lipnevich \u0026amp; J. K. Smith (Eds.), \u003cem\u003eThe Cambridge handbook of instructional feedback\u003c/em\u003e (pp. 531\u0026ndash;553). Cambridge University Press. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/9781316832134\u003c/span\u003e\u003cspan address=\"10.1017/9781316832134\" 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, Krusche S, Kutyniok G, Michaeli T, Nerdel C, Pfeffer J, Poquet O, Sailer M, Schmidt A, Seidel T, Kasneci G (2023) ChatGPT for good? On opportunities and challenges of large language models for education. Learn Individual Differences 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\u003eLim JM, Go C (2025) Integrating ChatGPT into Teacher Feedback: Practical Insights for L2 Writing Instruction. TESL-EJ, \u003cem\u003e29\u003c/em\u003e(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://eric.ed.gov/?id=EJ1489053\u003c/span\u003e\u003cspan address=\"https://eric.ed.gov/?id=EJ1489053\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMacDonald RB (1991) Developmental Students\u0026rsquo; Processing of Teacher Feedback in Composition Instruction. In \u003cem\u003eReview of Research in Developmental Education\u003c/em\u003e (Vol. 8, Issue 5). Managing Editor, RRIDE, National Center for Developmental Education, Reich College of Education, Appalachian State University, Boone, NC 28608 (\u003cspan\u003e$\u003c/span\u003e9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://eric.ed.gov/?id=ED354965\u003c/span\u003e\u003cspan address=\"https://eric.ed.gov/?id=ED354965\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcNeill KL, Krajcik J (2008) Inquiry and Scientific Explanations: Helping Students Use Evidence and Reasoning. Science as inquiry in the secondary setting. NSTA, pp 121\u0026ndash;134\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMerriam SB (1998) \u003cem\u003eQualitative research and case study applications in education\u003c/em\u003e (Rev. and expanded). Jossey-Bass Publishers\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiles MB, Huberman AM (1994) Qualitative data analysis: An expanded sourcebook, 2 edn. Sage. [Nachdr.])\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMills E, Mizouri A, Peach A (2025) Prompting Better Feedback: A Study of Custom GPT for Formative Assessment in Undergraduate Physics. Educ Sci 15(8):1058. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/educsci15081058\u003c/span\u003e\u003cspan address=\"10.3390/educsci15081058\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNicol DJ, Macfarlane-Dick D (2006) Formative assessment and self‐regulated learning: A model and seven principles of good feedback practice. Stud High Educ 31(2):199\u0026ndash;218. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03075070600572090\u003c/span\u003e\u003cspan address=\"10.1080/03075070600572090\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOpenAI (2025) \u003cem\u003eChatGPT\u003c/em\u003e (Version GPT-4o) [Large Language Model]\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOpenAI, Achiam J, Adler S, Agarwal S, Ahmad L, Akkaya I, Aleman FL, Almeida D, Altenschmidt J, Altman S, Anadkat S, Avila R, Babuschkin I, Balaji S, Balcom V, Baltescu P, Bao H, Bavarian M, Belgum J, Zoph B (2024) \u003cem\u003eGPT-4 Technical Report\u003c/em\u003e (arXiv:2303.08774). arXiv. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2303.08774\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2303.08774\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePellegrino J (2024) A New Era for STEM Assessment: Considerations of Assessment, Technology, and Artificial Intelligence. Uses of Artificial Intelligence in STEM Education. Oxford University Press, pp 17\u0026ndash;37\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSadler DR (2010) Beyond feedback: Developing student capability in complex appraisal. Assess Evaluation High Educ 35(5):535\u0026ndash;550. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602930903541015\u003c/span\u003e\u003cspan address=\"10.1080/02602930903541015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmager S, Pappas IO, Vassilakopoulou P (2025) Understanding Human-Centred AI: A review of its defining elements and a research agenda. Behav Inform Technol 44(15):3771\u0026ndash;3810. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/0144929X.2024.2448719\u003c/span\u003e\u003cspan address=\"10.1080/0144929X.2024.2448719\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStryker C, Scapicchio M (2024), March 22 \u003cem\u003eWhat is Generative AI?\u003c/em\u003e IBM. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ibm.com/think/topics/generative-ai\u003c/span\u003e\u003cspan address=\"https://www.ibm.com/think/topics/generative-ai\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSutton P (2012) Conceptualizing feedback literacy: Knowing, being, and acting. Innovations Educ Teach Int 49(1):31\u0026ndash;40. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/14703297.2012.647781\u003c/span\u003e\u003cspan address=\"10.1080/14703297.2012.647781\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVERBI Software (2024) \u003cem\u003eMaxQDA 24.7\u003c/em\u003e (Version 24.7) [Computer software]. Available from maxqda.com\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeaver CL, Duran EC, Nikles JA (2014) An Integrated Approach for Development of Scientific Writing Skills in Undergraduate Organic Lab. In \u003cem\u003eAddressing the Millennial Student in Undergraduate Chemistry\u003c/em\u003e (Vol. 1180, pp. 105\u0026ndash;123). American Chemical Society. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/bk-2014-1180.ch008\u003c/span\u003e\u003cspan address=\"10.1021/bk-2014-1180.ch008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u003cem\u003eWhat is RAG? - Retrieval-Augmented Generation AI Explained - AWS\u003c/em\u003e. (n.d.). Amazon Web Services, Inc. Retrieved March 8, (2026) from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://aws.amazon.com/what-is/retrieval-augmented-generation/\u003c/span\u003e\u003cspan address=\"https://aws.amazon.com/what-is/retrieval-augmented-generation/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWinstone NE, Nash RA, Parker M, Rowntree J (2017) Supporting Learners\u0026rsquo; Agentic Engagement With Feedback: A Systematic Review and a Taxonomy of Recipience Processes. Educational Psychol 52(1):17\u0026ndash;37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00461520.2016.1207538\u003c/span\u003e\u003cspan address=\"10.1080/00461520.2016.1207538\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWinstone NE, Nash,Robert A, Rowntree, James,and, Parker M (2017) \u0026lsquo;It\u0026rsquo;d be useful, but I wouldn\u0026rsquo;t use it\u0026rsquo;: Barriers to university students\u0026rsquo; feedback seeking and recipience. \u003cem\u003eStudies in Higher Education\u003c/em\u003e, \u003cem\u003e42\u003c/em\u003e(11), 2026\u0026ndash;2041. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/03075079.2015.1130032\u003c/span\u003e\u003cspan address=\"10.1080/03075079.2015.1130032\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYeh SS (2010) Understanding and addressing the achievement gap through individualized instruction and formative assessment. Assess Education: Principles Policy Pract 17(2):169\u0026ndash;182. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09695941003694466\u003c/span\u003e\u003cspan address=\"10.1080/09695941003694466\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu M, Liu OL, Lee H-S (2020) The effect of automated feedback on revision behavior and learning gains in formative assessment of scientific argument writing. Comput Educ 143:103668. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compedu.2019.103668\u003c/span\u003e\u003cspan address=\"10.1016/j.compedu.2019.103668\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[{"identity":"5904f836-f57e-4b33-9702-e4b9e5d7e2d9","identifier":"10.13039/100000001","name":"National Science Foundation","awardNumber":"2235783","order_by":0}],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Michigan State University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"artifical intelligence, generative AI, scientific writing, undergraduate biology education","lastPublishedDoi":"10.21203/rs.3.rs-9271072/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9271072/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eInstructors and education researchers are increasingly leveraging genAI to support feedback practices, yet we still know relatively little about how students use AI-enabled feedback. This gap is consequential because students must actively interpret and use feedback to benefit from it. This study examined how two teams of undergraduate biology students used AI-enabled feedback to revise their research proposals. The teams revised their proposal drafts after receiving AI-enabled feedback developed by a GPT and subsequently reviewed and edited by the instructor. Our findings show that AI-enabled feedback prompted actionable revision across both teams. Each team addressed approximately half of the feedback comments, indicating comparable levels of uptake despite their differing revision approaches. This pattern indicates that both teams were inconsistent in taking up the additional information and reasoning highlighted in the feedback, reflecting challenges similar to those commonly observed with traditional feedback. These findings provide behavioral evidence of how students used AI-enabled feedback to revise their scientific writing and demonstrates ways that prior research on feedback use and barriers can extend to AI-enabled feedback.\u003c/p\u003e","manuscriptTitle":"How Do Biology Undergraduates Use AI-enabled Feedback to Revise Scientific Writing?","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-06 03:30:49","doi":"10.21203/rs.3.rs-9271072/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c9cc355e-1064-4a2e-9814-389967ff7549","owner":[],"postedDate":"April 6th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":65683639,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2026-04-06T03:30:49+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-06 03:30:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9271072","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9271072","identity":"rs-9271072","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}