AI as Animal Welfare Safety Net: Large Language Models Favor Veterinary Referral Over Online Community Advice for Suspected Reptile Infections

AI as Animal Welfare Safety Net: Large Language Models Favor Veterinary Referral Over Online Community Advice for Suspected Reptile Infections | 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 AI as Animal Welfare Safety Net: Large Language Models Favor Veterinary Referral Over Online Community Advice for Suspected Reptile Infections Richard Digirolamo This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8735785/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 The global trade in pet reptiles is substantial, with Japan ranking as the second-highest importer worldwide. However, access to specialized veterinary care remains limited, encouraging owners to seek advice on unregulated online forums. We analyzed 692 Yahoo! Chiebukuro Q&A threads (2010–2025) describing suspected infections in three pet lizards—central bearded dragon ( Pogona vitticeps ; n = 168), leopard gecko ( Eublepharis macularius ; n = 431), and endemic Japanese grass lizard ( Takydromus tachydromoides ; n = 93)—and compared community-selected human “Best Answers” with outputs from three large language models (ChatGPT-5.2, Claude Sonnet 4.5, Gemini 3.0 Pro). Advice was classified into five hierarchical categories, with the primary safety endpoint being immediate veterinary referral (Vet-direct) versus all non-referral categories. Human referral rates differed by species (23.66–41.07%), with the endemic Japanese grass lizard receiving the fewest Vet-direct recommendations and the highest rate of home-treatment advice. In contrast, all LLMs produced substantially higher Vet-direct rates across species (61.25–95.70%); paired thread-level McNemar tests showed LLMs were significantly more likely than humans to recommend Vet-direct for every species–model comparison (all p ≤ 9.58×10⁻⁶). However, biosafety guidance related to zoonotic risk and hygiene was rare in both human and AI responses. These findings suggest that online communities may reinforce a “valuation hierarchy” that disadvantages low-economic-value species, while LLMs can function as a welfare-oriented safety net; nevertheless, proactive One Health messaging remains an important gap. Animal Science One Health Infodemic leopard gecko bearded dragon veterinary referral artificial intelligence Figures Figure 1 1. Introduction The landscape of companion animal ownership has diversified significantly, with non-traditional pets—particularly reptiles—experiencing a surge in global popularity. Japan ranks as the second-largest reptile importer globally, accounting for 24% of the global market share in 2023 (Trendeconomy, 2023 ). This trend is exemplified by species such as the leopard gecko ( Eublepharis macularius ) and the central bearded dragon ( Pogona vitticeps ) (hereafter bearded dragon), which have become staples in the international pet trade and are top two popular non-native pet reptile species in Japan (Digirolamo, 2025a ). Concurrently, endemic reptile species are also popular pets. The Japanese grass lizard ( Takydromus tachydromoides ) is the most popular endemic pet lizard species in Japan (Digirolamo, 2025a ), frequently kept due to its widespread availability and ease of capture or low purchase cost. However, this diversification in herpetoculture has outpaced the availability and awareness of specialized veterinary care. Owners of these species face significant hurdles, including the high cost of exotic animal medicine and a scarcity of qualified practitioners. According to hospital search databases, only 3–5% of animal hospitals in the Tokyo area treat reptiles, with even fewer options available in non-urban regions (Anicom, 2025 ; CalooPet, 2025 ; EPARK Pet Life, 2025 ). Compounding this infrastructure deficit is a cultural context where exotic animal welfare is often compromised. Japan has faced scrutiny for policies and practices that commodify wildlife, most notably the proliferation of "exotic animal cafes". These establishments house species ranging from otters (Ushine et al., 2024 ), owls (Leupen et al., 2024 ) to various reptiles (Sigaud et al., 2023 ; Tanaka et al., 2025 ) in conditions often characterized by poor welfare, restricted mobility, and forced human interaction (Tanaka et al., 2025 ). This environment potentially normalizes the perception of exotics as "living toys" rather than patients requiring medical care, further widening the gap between ownership and responsible veterinary utilization. This care gap, combined with the lack of reliable institutional guidance, has necessitated a large-scale behavioral shift. In the absence of accessible professional care, owners are increasingly turning to online communities and social media as their primary source of information (Digirolamo, 2025a ; Lai et al., 2021 ; Pienaar & Sturgeon, 2024 ; Springer et al., 2024 ). While these platforms democratize information by offering rapid access to anecdotal experience, they remain unregulated and un-vetted. This environment is conducive to the propagation of an "infodemic"—an overabundance of information, both accurate and inaccurate, that impedes the ability of individuals to locate trustworthy guidance during critical times. In human and animal health, infodemics are a recognized threat (Abuhaloob et al., 2024 ; Gallotti et al., 2020 ; Richartz et al., 2024 ; Shi et al., 2024 ; Wenzel et al., 2023 ). In the context of pet reptile welfare, they represent an unquantified but significant multidimensional risk. This risk is most acute when owners seek advice for symptoms indicative of infectious disease/infections. While chronic nutritional or metabolic disorders (e.g., metabolic bone disease) constitute a significant portion of reptile morbidity, this study explicitly limits its scope to suspected infectious etiologies. This focus was selected due to three distinct factors that amplify the potential for welfare compromise. First, infectious states in small ectotherms are often acute and rapidly fatal. Unlike chronic conditions, the "time-to-intervention" is critical; delays caused by misinformation carry a disproportionately high welfare cost. Reptiles are stoic and adept at concealing signs of illness, often until they are so severely physiologically compromised that they can no longer maintain the appearance of health. Consequently, observable clinical signs often reflect advanced disease, making timely intervention critical (Broughton & Webb, 2022 ; Paré & Lentini, 2010 ; Schilliger et al., 2020 ). This biological trait creates a diagnostic lag, where visible symptoms typically indicate the late stages of a disease rather than its onset. Second, suspected infections are more likely than husbandry issues to prompt owners to attempt unregulated pharmaceutical interventions (e.g., human ointments), creating a risk of iatrogenic harm. Finally, this issue must be viewed through the lens of "One Health"—a collaborative framework recognizing that the health of people is closely connected to the health of animals and their shared environment (Garcia et al., 2025 ; World Organization for Animal Health, 2024 ). In herpetoculture, animal welfare and public health are intrinsically linked; for example, the mishandling of waste like diarrhea could increase the risk of zoonotic diseases, most notably reptile-associated salmonellosis (RAS) (Nakadai et al., 2005 ; Pees et al., 2023 ). Inaccurate online advice that lacks proper guidance on quarantine and hygiene constitutes not only a "herd health" failure for the animals but a direct biosafety risk to the human handler and the wider household. The welfare impact is, therefore, multidimensional. It begins with the physical dimension of the pathogen but is immediately compounded by an informational dimension. The advice an owner receives online becomes a direct effector of animal welfare. Medically sound advice (e.g., "See a veterinarian immediately") could lead to positive outcomes. Conversely, harmful advice (e.g., "Scrub the mouth with iodine" or "Just wait and see") prolongs suffering, delays in professional treatment, and can lead to fatal outcomes. Despite anecdotal acknowledgment of this issue, scientific literature lacks a quantitative assessment of this infodemic for pet reptiles. We currently lack data on the prevalence of harmful advice versus sound veterinary referrals on these platforms. Furthermore, it is unknown if the quality of information differs between captive-bred species (like leopard gecko and bearded dragons) and endemic, often wild-caught species (like the Japanese grass lizard), where husbandry data may be sparser. To address this gap, this study utilizes data from Yahoo! Chiebukuro, Japan's largest question-answering website (Informatics Research Data Repository, 2024 ). The platform offers a unique longitudinal dataset of owner concerns across the country. Crucially, its "Best Answer" feature—where the questioner selects the most helpful response—provides a proxy for the "accepted" community consensus, allowing us to evaluate the specific advice that owners are most likely to follow. Additionally, the rise of Large Language Models (LLMs) and AI-integrated search introduces a new variable to this ecosystem. As Q&A sites begin to offer AI-generated answers (D’Angelo, 2023 ; LY Corporation, 2024 ), it is crucial to understand whether these automated systems mitigate or exacerbate the welfare risk compared to human community advice. Previous research indicates that LLMs (ChatGPT and Gemini) provided desirable answers for clean, best-case scenario health questions regarding leopard gecko (Digirolamo, 2025b ). This study expands that investigation by using raw, natural-language questions posted to Yahoo! Chiebukuro to evaluate three specific LLMs: Claude Sonnet 4.5 (hereafter Claude), ChatGPT-5.2 (hereafter ChatGPT), and Gemini 3 Pro (hereafter Gemini). These models were selected to represent different capabilities of the AI landscape: Claude is currently integrated directly into Yahoo! Chiebukuro as a user-selected option; ChatGPT represents the most globally popular LLM (StatCounter Global Stats, 2025 ) and represents one of the highest benchmarks including scientific reasoning (GPQA Diamond: 92.4%) and multilingual Q&A (MMMLU: 89.6%); and Gemini represents a state-of-the-art multimodal model, also consistently achieving top-tier benchmarks including scientific reasoning (GPQA Diamond: 91.9%) and multilingual Q&A (MMMLU: 91.8%) (Google, 2025 ). This study aims to fill the knowledge gap regarding the safety of online veterinary advice by systematically analyzing a longitudinal dataset of owner questions and community-provided answers from a major Japanese Q&A website. We quantify the prevalence and nature of treatment advice given for symptoms of suspected infectious disease across three popular reptile species. Our primary objectives are: To classify the quality and safety of "accepted" community advice for infectious symptoms. To generate advice using three prominent LLMs and compare their safety classification against human advice. To compare the distribution of advice quality across the three different species. We hypothesize that a significant proportion of community-sourced advice for suspected reptile infections is ethically and medically inappropriate and poses a direct, quantifiable risk to animal welfare compared to the safety baselines established by modern LLMs. 2. Methods 2.1. Data Source and Acquisition Data were obtained from Yahoo! Chiebukuro (Yahoo! JAPAN), the largest publicly accessible Japanese question-and-answer (Q&A) platform. All threads were retrieved from the “Pet” category using commercial web-scraping software (Silk Script Inc., 2025). Species-specific subcorpora were constructed using commonly used Japanese terms and abbreviations: bearded dragon (“フトアゴ”), leopard gecko (“ヒョウモントカゲモドキ” and “レオパ”), and Japanese grass lizard (“カナヘビ”). The collection window spanned 01 January 2010 to 30 September 2025, corresponding to the earliest retrievable posts available through the scraping software. For each thread, the question text, posting metadata, and the community-selected “Best Answer” were extracted. Only publicly available text was collected; no attempt was made to identify users, and all data were analyzed in aggregate. 2.2. Filtering and eligibility criteria A multi-stage screening procedure was applied to derive the analytical sample. First, each species subcorpus was filtered using a curated dictionary of Japanese keywords suggestive of potential infectious disease, infected wounds, or localized infections (Table 1 ), developed from husbandry and clinical references for the focal species (Koieyama, 2019 ; Miyazaki, 2022 ; Takenaka, 2023 ). Second, duplicate threads and threads without an original-poster-selected “Best Answer” were removed. Third, the remaining threads were manually reviewed to confirm that the original post described current health signs and was actively seeking management advice. Threads focused primarily on breeding, baseline husbandry without illness signs, or retrospective discussion of death causes were excluded. Threads in which an infection had already been diagnosed, but the user requested additional guidance on care or treatment, were retained, as these posts represent decision-making regarding infectious disease management. Table 1 Keyword Dictionary for Infectious Disease Filtering Category Key terms in Japanese English translation General Signs 腫れ Swelling 爛れ Sore / Erosion 死* Death ぐったり Limp / Lethargic 痩せ Thin 病気 Disease/Illness/Sickness Pathogens 菌* Bacteria/germs カビ Fungus ダニ Mites 寄生虫 Parasite ウイルス Virus 感染 Infection Excretions 膿 Pus 下痢 Diarrhea 吐* Vomit 血便 Blood in stool 鼻水 Runny nose よだれ Drooling/Saliva Conditions コクシ* Coccidia クリプト* Cryptosporidiosis マウスロット Mouth rot 炎* Inflammation Location 口を|口が Mouth (sentence start) 呼吸 Breathing くしゃみ Sneezing *The term “死” covers death/dying *The term “菌” covers general term of germs including bacteria *The term "クリプト" covers "クリプトスポリジウム"(Cryptosporidiosis) *The term "吐" covers all common forms of vomit/vomiting/nausea *The term "コクシ" covers "コクシジウム"(Coccidia) *The term "炎" covers various conditions, including general inflammation, pneumonia, and eye infections 2.3. Advice classification framework The primary unit of analysis was the “Best Answer” selected by the original poster. Each answer was coded into one of five mutually exclusive categories representing the primary recommended action: (1) Vet-direct: explicit recommendation to seek professional veterinary assessment or treatment as the primary action. (2) Husbandry-check: advice to evaluate or correct environmental or routine husbandry parameters (e.g., temperature, humidity, lighting, feeding schedule) within standard care. (3) Home-treat: recommendation of active non-veterinary interventions (e.g., topical human medications or home remedies). (4) Wait-and-see: recommendation to observe without intervention or minimization of the condition (e.g., “it will resolve on its own”). (5) No action: non-actionable or non-substantive guidance (e.g., sympathy-only responses or irrelevant content). When multiple recommendations were present, a hierarchical decision rule was applied to assign the category representing the most immediate action likely to influence welfare outcomes. Answers were coded as Vet-direct only when veterinary consultation was advised as the first-line action. If veterinary consultation was explicitly contingent on prior non-veterinary steps (e.g., “try home treatment first and see a veterinarian if it fails”), the response was coded according to the initial action (Home-treat, Husbandry-check, or Wait-and-see). This approach distinguishes between immediate referral and delayed referral from an animal welfare risk perspective. 2.4. AI response generation For each eligible thread, the question text (excluding human answers) was submitted to three large language models (LLMs): ChatGPT (GPT-5.2 Thinking; OpenAI), Claude (Claude Sonnet 4.5; Anthropic), and Gemini (Gemini 3.0 Pro; Google). All models were accessed via their standard web interfaces using free-tier accounts to approximate public accessibility. To reduce carry-over effects, new isolated sessions were used for each batch under Temporary Chat function for ChatGPT and Gemini; Incognito Chat for Claude. Questions were processed in batches (approximately 80 questions) to remain within context window constraints, with total token counts checked using the OpenAI Tokenizer (OpenAI, 2025 ). A standardized Japanese prompt instructed the model to provide the most important advice and to return outputs in a two-column Markdown table (ID, Answer). The full prompt text (Japanese original with English translation) is provided in supplementary material 1. AI outputs were subsequently classified using the same coding framework applied to human Best Answers. The dates of model access were from 15–26 December 2025. 2.5. Statistical Analysis Analyses focused on the safety-relevant outcome of veterinary referral. The primary endpoint was binary: Vet-direct (Category 1) versus non-referral (Categories 2–5). For each species, descriptive statistics were used to summarize the distribution of advice categories for human Best Answers and for each LLM (ChatGPT, Claude, and Gemini). Because each thread yielded matched observations (one human Best Answer and one response from each LLM), human–AI comparisons of Vet-direct recommendations were conducted at the thread level. For each species and for each model separately, paired differences in Vet-direct proportions between human Best Answers and LLM responses were evaluated using two-sided McNemar’s tests with continuity correction, using discordant pairs only. Statistical significance was defined as α = 0.05. Among human Best Answers, Vet-direct proportions were compared across species using a chi-square test of independence, and effect size was summarized using Cramér’s V. All analyses were performed in Microsoft Excel. 3. Results Following the scraping, filtering, and cleaning process, the final dataset comprised 168 threads for bearded dragons, 431 for leopard geckos, and 93 for Japanese grass lizards. Across all species, veterinary referral (Category 1: Vet-direct) was recommended more frequently by large language models (LLMs) than by human Best Answers. Model-specific veterinary referral rates are shown in Table 3 ; paired thread-level comparisons are summarized in Table 4 . 3.1. Bearded dragon For bearded dragons (N = 168), human Best Answers recommended veterinary consultation in 41.07% of cases (69/168). The most common alternative human recommendation was husbandry correction (Category 2: 34.52%, 58/168), followed by home treatment (Category 3: 11.31%, 19/168). In contrast, LLM responses showed higher vet-direct rates across all models (Claude: 61.90%, 104/168; ChatGPT: 79.76%, 134/168; Gemini: 90.48%, 152/168). 3.2. Leopard gecko For leopard geckos (N = 431), human Best Answers recommended veterinary referral in 34.34% of threads (148/431). Non-referral advice was distributed across husbandry checks (27.15%, 117/431), home treatment (14.62%, 63/431), and watchful waiting (13.92%, 60/431). LLM outputs recommended vet-direct action more frequently than humans (Claude: 61.25%, 264/431; ChatGPT: 75.41%, 325/431; Gemini: 77.96%, 336/431). 3.3. Japanese grass lizard For Japanese grass lizards (N = 93), veterinary referral was recommended in 23.66% of human Best Answers (22/93), with comparatively high use of home treatment (31.18%, 29/93) and no actionable guidance (Category 5: 17.20%, 16/93). LLM responses were more risk-averse, recommending vet-direct action in 66.67%–95.70% of cases (Claude: 62/93; ChatGPT: 82/93; Gemini: 89/93). 3.4. Paired human–AI comparisons Because each thread generated matched observations (one human Best Answer and one response per LLM), human–AI differences in veterinary referral were evaluated using McNemar’s tests at the thread level (Table 4 ). Across all three species and all three models, LLM outputs were significantly more likely than human Best Answers to recommend immediate veterinary consultation (Vet-direct) (all p ≤ 9.58×10⁻⁶). Discordant-pair counts consistently showed more threads where the LLM recommended Vet-direct when the human Best Answer did not (c) than the reverse (b), indicating a consistent shift toward veterinary referral in LLM-generated guidance compared with community-selected human responses. 3.5. Comparison of human advice across species Among human Best Answers, the probability of receiving Vet-direct advice differed by species (χ²(2, N = 692) = 8.05, p = 0.018; Cramér’s V = 0.108). Bearded dragons and leopard geckos received veterinary referral advice more often (41.07% and 34.34%, respectively) than Japanese grass lizards (23.66%), which also showed the highest proportion of home-treatment recommendations (31.18%). Table 2 Comparative distribution of advice categories (%) between LLMs (mean) and Human respondents. Species & Source N Cat. 1 (Vet) Cat. 2 (Husbandry) Cat. 3 (Home-treat) Cat. 4 (Wait&See) Cat. 5 (No action) Bearded Dragon LLMs Mean 168 77.38 14.48 2.58 4.96 0.6 Human (Yahoo!) 168 41.07 34.52 11.31 7.14 5.95 Leopard gecko LLMs Mean 431 71.54 19.33 2.4 6.03 0.7 Human (Yahoo!) 431 34.34 27.15 14.62 13.92 9.98 Japanese grass lizard LLMs Mean 93 83.51 6.81 2.15 3.94 3.58 Human (Yahoo!) 93 23.66 26.88 31.18 1.08 17.2 Table 3 Comparison of veterinary referral rates (Category 1) across three LLMs. Species Claude ChatGPT Gemini AI Mean Bearded Dragon (N = 168) 61.90% 79.76% 90.48% 77.38% Leopard gecko (N = 431) 61.25% 75.41% 77.96% 71.54% Japanese grass lizard (N = 93) 66.67% 88.17% 95.70% 83.51% Table 4 Thread-level paired comparison of Vet-direct recommendations between human Best Answers and LLM outputs (McNemar’s test). Species N Comparison Human Vet-direct n (%) LLM Vet-direct n (%) Human Yes / LLM No (b) Human No / LLM Yes (c) McNemar χ² p-value Bearded dragon 168 Human vs Claude 69 (41.07) 104 (61.90) 12 47 19.59 9.58E-06 Bearded dragon 168 Human vs ChatGPT 69 (41.07) 134 (79.76) 6 71 53.19 3.02E-13 Bearded dragon 168 Human vs Gemini 69 (41.07) 152 (90.48) 2 85 77.29 1.48E-18 Leopard gecko 431 Human vs Claude 148 (34.34) 264 (61.25) 23 139 81.64 1.64E-19 Leopard gecko 431 Human vs ChatGPT 148 (34.34) 325 (75.41) 15 192 149.64 2.08E-34 Leopard gecko 431 Human vs Gemini 148 (34.34) 336 (77.96) 11 199 166.52 4.26E-38 Japanese grass lizard 93 Human vs Claude 22 (23.66) 62 (66.67) 6 46 29.25 6.36E-08 Japanese grass lizard 93 Human vs ChatGPT 22 (23.66) 82 (88.17) 5 65 49.73 1.77E-12 Japanese grass lizard 93 Human vs Gemini 22 (23.66) 89 (95.70) 0 67 65.02 7.43E-16 Notes: b = discordant threads where the human Best Answer recommended Vet-direct but the LLM did not. c = discordant threads where the human Best Answer did not recommend Vet-direct but the LLM did. Continuity-corrected McNemar statistic: χ² = (|b − c| − 1)² / (b + c) (df = 1). P-values are computed using the chi-square distribution with 1 degree of freedom. 3.6. Model-to-model variation Within LLM outputs, Gemini and ChatGPT generally produced higher Vet-direct rates than Claude across all species (Table 3 ). However, even the least risk-averse model (Claude) recommended veterinary referral more frequently than human Best Answers for every species’ veterinary care. Claude produced the lowest Vet-direct proportions across species (61.25%–66.67%), but still exceeded human Best Answers in every species. 3.7. Distribution of advice categories Figure 1 Comparative distribution of medical advice categories provided by LLMs versus human community respondents across three reptile species. The "AI" columns represent the calculated mean frequency of advice generated by three LLMs (ChatGPT, Claude, Gemini). The "Human" columns represent the "Best Answer" selected by users on the Yahoo! Chiebukuro platform. Advice was classified into five hierarchical categories: Category 1 (Vet Referral), Category 2 (Husbandry Check), Category 3 (Home Treatment), Category 4 (Wait & See), and Category 5 (No Actionable Advice). 3.8. Evaluation of Zoonotic Risk and Biosafety Guidance A qualitative review was conducted on all human "Best Answers" and AI-generated responses to evaluate the prevalence of zoonotic safety and hygiene guidance. Guidance regarding human biosafety (handwashing/disinfection, waste handling) was found to be rare across all platforms and models. Among the 692 human-sourced "Best Answers," only 0.43% (n = 3) mentioned handwashing or hygiene for the purpose of safeguarding human health, and only 0.14% (n = 1) explicitly identified the risk of salmonella. Similarly, across the 2,076 AI-generated responses (n = 692 questions x 3 models), the mention rate for human-centric hygiene protocols was 0.14% (n = 3). Notably, the cases for the AI were limited to a single specific query that appeared in both the human and AI datasets regarding whether a diagnosed pinworm infection in a bearded dragon could be transmitted to humans. In this isolated instance, both the human community and all three LLMs correctly recommended handwashing. None of the LLMs proactively cautioned against the risk of salmonella. 4. Discussion 4.1. The "Care Gap" and the Informational Infodemic To our knowledge, this is the first quantitative assessment of the "information ecosystem" available to reptile owners in Japan during medical crises. The results confirm that community-sourced advice frequently fails to recommend veterinary intervention for acute infectious symptoms, posing a significant welfare risk. This failure is heavily stratified by species. While LLMs maintained a consistent "safety baseline" (recommending veterinary care in 71.54%–83.51% of cases regardless of species), human advice quality fluctuated markedly. The analysis reveals a valuation hierarchy: owners of the endemic Japanese grass lizard were significantly less likely to receive veterinary referrals (23.66%) compared to owners of Leopard geckos (34.34%) and Bearded dragons (41.07%). 4.2. Economic barriers and species-based differences in referral advice The stark disparity in advice quality suggests a deep-seated economic bias. Even for the bearded dragons—the "best performing" species in the human dataset—fewer than half of the owners were advised to seek professional medical assistance for potentially fatal symptoms. This low baseline is likely driven by an "economic inversion" of value, where the cost of veterinary intervention often exceeds the replacement cost of the animal. In the Japanese market, a juvenile leopard gecko or bearded dragon of a standard morph can frequently be purchased for 5,000 to 15,000 JPY (approx. $ 35– $ 100 USD). In contrast, a single veterinary consultation, combined with diagnostics (e.g., bloodwork, X-rays) and treatment, can easily exceed, or be comparable to, these prices. For the Japanese grass lizard, often sold for ~ 1,000 JPY or caught freely in the wild, this economic disparity is more extreme. When a pet’s market value is lower than the price of its medical treatment, it risks being treated as readily replaceable, leading owners (and advice-givers) to prefer low-cost home remedies (Irvine, 2003 ). This is evidenced by the high prevalence of Category 3 (Home Treatment) advice for Japanese grass lizards (31.18%), which exceeded the rate of veterinary referrals. 4.3. AI as a Welfare "Safety Net" A critical finding of this study is the inverse relationship between human and AI performance regarding the endemic species. While human advice quality degraded for the Japanese grass lizard, LLMs effectively compensated for this bias, recommending veterinary care in 83.51% (AI Mean) of cases. This suggests that LLMs, trained on vast global datasets, do not share the localized economic biases that devalue native wildlife. Consequently, the integration of AI-generated answers into Q&A platforms could serve as a vital "welfare safety net," guiding novice owners of undervalued species toward professional care when human communities fail to do so. 4.4. The One Health Blind Spot: Zoonotic Risks and Proactive Triage A critical finding of this study is the near-total absence of proactive biosafety guidance. While the LLMs and human respondents correctly identified handwashing as a necessity when explicitly queried about pinworm transmission, they failed to offer similar warnings for symptoms that may increase the risk of Salmonella transmission or other enteric pathogens. In the context of "One Health," this represents a dangerous failure in triage logic. Within herpetoculture, animal welfare and public health are inextricably linked; the mishandling of infectious waste during a medical crisis is a primary driver of Reptile-Associated Salmonellosis (RAS) (Nakadai et al., 2005 ; Pees et al., 2023 ). It could be argued that the lack of biosafety warnings in the AI-generated responses was a byproduct of the specific prompt used in this study, which instructed the models to "emphasize the most important advice" regarding the reptile's health. Under a narrow clinical interpretation, the AI prioritized the survival of the animal (the patient) over the safety of the handler. However, we contend that the open-ended nature of this prompt was appropriate and necessary for two reasons. First, it accurately reflects the "natural language" behavior of novice owners, who typically seek solutions for the animal’s visible distress rather than inquiring about their own potential exposure. Second, a truly robust medical AI should operate under a "safety-first" paradigm that recognizes the human-animal interface as a singular clinical unit. If an LLM recognizes an infection risk to the animal but fails to trigger a corresponding warning for the handler, its utility as a public health tool remains limited. The "0.14% success rate" observed here suggests that current LLMs are reactive rather than proactive regarding zoonosis; they provide safety advice only when the user possesses enough prior knowledge to ask a specific question. This creates a "knowledge trap" where the most vulnerable owners—those unaware of RAS risks—are the least likely to receive the hygiene guidance. 4.5. The Paradox of Choice and Confirmation Bias Integrating AI answers alongside human responses introduces the risk of conflicting advice. If an LLM suggests a veterinary visit while a human user suggests a convenient "home remedy," owners may be psychologically predisposed to follow the human advice due to confirmation bias—the tendency to favor information that aligns with one's desire to avoid costs/effort. Therefore, simply displaying AI answers is insufficient. Q&A platforms should implement Context-Aware Warning Systems. If a thread contains keywords indicative of illness, an automatic banner should explicitly warn users that "home remedies for these symptoms carry high risks," prioritizing clinical guidelines over anecdotal suggestions. 4.6. Limitations Several limitations must be acknowledged. First, because this was an observational study using publicly posted threads, we could not control unmeasured confounding (e.g., owner resources, severity of clinical signs, or prior veterinary access). Second, the data is derived exclusively from Yahoo! Chiebukuro . While the anonymity of the platform likely yields more honest data regarding "home remedy" attempts than face-to-face interviews (mitigating social desirability bias—answering in a way they believe is "correct" or "ethical" (Krumpal, 2011 )), the population may not fully represent all reptile owners. Third, our classification focused on safety (process) rather than diagnostic accuracy (outcome). We did not attempt to verify if the "home remedy" suggested might have incidentally worked, nor did we diagnose the animals ourselves. However, this distinction is critical: treating a suspected infectious disease requires professional diagnostics. Even if a layperson's guess ("wait and see") happens to be correct in a specific instance, the process of relying on unverified internet advice is inherently dangerous. The welfare risk lies in the lack of professional triage, not solely in the outcome of a specific case. Finally, regarding LLM consistency, while all models outperformed humans, Gemini and ChatGPT were generally stricter than Claude. Thus, usage of models not tested or paid tiers or future iterations may show different outcomes. However, even the "least strict" model consistently outperformed the human average, confirming that the "AI Safety Net" effect holds true despite model variation. Additionally, all advice categorization was conducted by a single coder; inter-rater reliability was not assessed. 4.7. Future Directions: Bridging the Gap Future initiatives must focus on bridging the gap between digital advice and physical veterinary access. One immediate solution is the promotion of telemedicine services that are available in Japan. Online veterinary consultations, while not free, are significantly more affordable than physical hospital visits and eliminate the stress of transport for the animal (mirpet, 2025 ; Pet-online hospital24, 2025 ). For owners of "low-value" species who are hesitant to pay full clinic fees, telemedicine offers a "middle-ground" triage step that is safer than free Q&A sites such as Yahoo! Chiebukuro but more accessible than full veterinary visit. Platforms and AI tools should be programmed to suggest legitimate telemedicine providers as a viable alternative to home remedies. 5. Conclusions This study provides the first quantitative evidence of a digital "infodemic" affecting the welfare of pet reptiles in Japan. Our analysis confirms that while online peer-to-peer platforms offer community support, they are structurally inadequate for medical crisis management. This inadequacy is heavily stratified by species, revealing a pervasive "valuation hierarchy" where endemic species like the Japanese grass lizard are significantly less likely to receive veterinary referrals than higher cost "commodity" pets. The recommendation of high-risk home remedies over professional care for lower-value species highlights a "care gap" where the low economic replacement cost of an animal may lead to a dangerous devaluation of its medical needs. Conversely, Large Language Models (LLMs) demonstrated a significant "AI Welfare Safety Net" effect. Across all models—Claude, ChatGPT, and Gemini—the systems maintained a high-safety baseline that effectively ignored the economic or cultural biases compromising human advice. While significant internal differences in referral strictness exist, the overall performance of LLMs suggests they are superior tools for preliminary clinical triage. However, the "0.14% success rate" in proactive biosafety guidance identifies a critical One Health blind spot. Neither human nor AI sources proactively addressed the risks of zoonotic transmission, such as Reptile-Associated Salmonellosis (RAS), unless specifically prompted. We conclude that while online peer-to-peer platforms provide social support, they are structurally unreliable for managing suspected infectious disease in reptiles, and their safety performance varies by species. LLMs consistently recommended veterinary referral more frequently than community-selected human answers, suggesting a potential welfare “safety net” effect for undervalued or endemic species. Future risk-reduction strategies should prioritize (1) context-aware platform warnings that discourage high-risk home remedies while reinforcing veterinary triage, (2) improved visibility of legitimate telemedicine pathways as a financially accessible bridge to clinical care, and (3) AI outputs that integrate proactive One Health guidance (e.g., hygiene and waste-handling precautions) alongside animal-focused recommendations. Declarations Supporting information captions: The data presented in this study are derived from publicly accessible content on Yahoo! Chiebukuro ( https://chiebukuro.yahoo.co.jp/ ). Due to copyright restrictions regarding the redistribution of raw user-generated content, the full-text dataset cannot be shared directly. However, a complete list of URLs for all Q&A threads analyzed in this study, along with summarized English translations focusing on the clinical signs and the corresponding "Best Answers," are available in Supplementary Material 2 (bearded dragon), Supplementary Material 3 (leopard gecko), and Supplementary Material 4 (Japanese grass lizard). Note that accessing Yahoo! Japan websites from outside Japan currently requires a VPN set to a Japan location. References Abuhaloob L, Purnat TD, Tabche C, Atwan Z, Dubois E, Rawaf S (2024) Management of infodemics in outbreaks or health crises: A systematic review. Front Public Health 12:1343902. https://doi.org/10.3389/fpubh.2024.1343902 Anicom (2025) Search for animal hospitals in Tokyo for lizards. https://www.anicom-ah.com/tokyo/lizard Broughton C, Webb KL (2022) Diagnostic clinical pathology of the bearded dragon (Pogona vitticeps). Veterinary Clin North America: Exotic Anim Pract 25(3):713–734. https://doi.org/10.1016/j.cvex.2022.06.002 CalooPet (2025) Reptile hospital search (Tokyo). https://pet.caloo.jp/hospitals/search/13/a10/all D’Angelo A (2023), February 1 Poe. Quora Blog. https://quorablog.quora.com/Poe-1 Digirolamo R (2025a) Understanding pet reptile preferences in Japan: An analysis using Yahoo! Chiebukuro and Google Trends. Herpetological J 35(2):146–154. https://doi.org/10.33256/35.2.146154 Digirolamo R (2025b) Accuracy and completeness of AI chatbot responses for leopard gecko (Eublepharis macularius) husbandry: An exploratory comparison of ChatGPT-4o and Gemini 2.5 Pro. J Appl Anim Welfare Sci. https://doi.org/10.1080/10888705.2025.2606681 EPARK Pet Life (2025) Reptile animal hospital search in Tokyo. https://petlife.asia/hospital/search/tokyo/reptiles/ Gallotti R, Valle F, Castaldo N, Sacco P, De Domenico M (2020) Assessing the risks of ‘infodemics’ in response to COVID-19 epidemics. Nat Hum Behav 4(12):1285–1293. https://doi.org/10.1038/s41562-020-00994-6 Garcia MR et al (2025) One welfare: Bibliometric review of world literature. Front Veterinary Sci 12:1627981. https://doi.org/10.3389/fvets.2025.1627981 Google (2025), November 18 A new era of intelligence with Gemini 3. https://blog.google/products/gemini/gemini-3/#gemini-3 Informatics Research Data Repository (2024) Yahoo! Dataset (in Japanese). https://www.nii.ac.jp/dsc/idr/yahoo/chiebkr3/Y_chiebukuro.html Irvine L (2003) The problem of unwanted pets: A case study in how institutions ‘think’ about clients’ needs. Soc Probl 50(4):550–566. https://doi.org/10.1525/sp.2003.50.4.550 Koieyama H (2019) The complete book of leopard gecko health and disease (in Japanese). SEIBUNDO SHINKOSHA Publishing Krumpal I (2011) Determinants of social desirability bias in sensitive surveys: A literature review. Qual Quant 47(4):2025–2047. https://doi.org/10.1007/s11135-011-9640-9 Lai N, Khosa DK, Jones-Bitton A, Dewey CE (2021) Pet owners’ online information searches and the perceived effects on interactions and relationships with their veterinarians. Veterinary Evid 6(1). https://doi.org/10.18849/ve.v6i1.345 Leupen BTC, Wakao K, Asakawa Y, Eaton JA, Bruslund S (2024) Live owls in Japanese pet stores and cafés: Volumes, species, and impediments to effective trade monitoring. J Asia-Pacific Biodivers 17(3):513–524. https://doi.org/10.1016/j.japb.2024.03.006 LY Corporation (2024) Yahoo! Chiebukuro starts providing ‘Minna no Chiebukuro’ AI answer function. https://www.lycorp.co.jp/ja/news/release/009337/ mirpet (2025) Online veterinary services. https://mirpet.co.jp/ Miyazaki T (ed) (2022) Bearded dragon care bible (in Japanese). MATES UNIVERSAL CONTENTS Co., Ltd. Nakadai A et al (2005) Prevalence of Salmonella spp. in pet reptiles in Japan. J Vet Med Sci 67(1):97–101. https://doi.org/10.1292/jvms.67.97 OpenAI (2025) Tokenizer. https://platform.openai.com/tokenizer Paré JA, Lentini AM (2010) Reptile geriatrics. Veterinary Clinics of North America. Exotic Anim Pract 13(1):15–25. https://doi.org/10.1016/j.cvex.2009.09.003 Pees M et al (2023) Salmonella in reptiles: A review of occurrence. Front Cell Dev Biology 11:1251036. https://doi.org/10.3389/fcell.2023.1251036 Pet-online hospital24 (2025) Online pet consultations. https://pet-online24.com/ Pienaar EF, Sturgeon DJE (2024) Exotic pet owners’ preferences for different ectothermic taxa are based on species traits and purchase prices in the United States. NeoBiota 91:1–27. https://doi.org/10.3897/neobiota.91.109403 Richartz ER, Hodgkiss BA, Black-Ocken NC, Fuentes RA, Looper JS, Withers SS (2024) Characterization of the dissemination of canine cancer misinformation on YouTube. Vet Comp Oncol 22(3):359–366. https://doi.org/10.1111/vco.12977 Schilliger L, Vergneau-Grosset C, Desmarchelier MR (2020) Clinical reptile behavior. Veterinary Clin North America: Exotic Anim Pract 24(1):175–195. https://doi.org/10.1016/j.cvex.2020.09.008 Shi R, Jia X, Hu Y, Wang H (2024) The media risk of infodemic in public health emergencies: Consequences and mitigation approaches. PLoS ONE 19(9):e0308080. https://doi.org/10.1371/journal.pone.0308080 Sigaud M, Kitade T, Sarabian C (2023) Exotic animal cafés in Japan: A new fashion with potential implications for biodiversity, global health, and animal welfare. Conserv Sci Pract 5(2). https://doi.org/10.1111/csp2.12867 Silk Script Inc (2025) Web scraping services. https://www.silk-s.jp/sw75.html Springer S, Lund TB, Corr SA, Sandøe P (2024) Does ‘Dr. Google’ improve discussion and decisions in small animal practice? Front Veterinary Sci 11:1417927. https://doi.org/10.3389/fvets.2024.1417927 StatCounter Global Stats (2025) AI chatbot market share worldwide. https://gs.statcounter.com/ai-chatbot-market-share#monthly-202503-202508 Takenaka S (2023) Beginner’s guide to Kanahebi: Proper care and rearing (in Japanese). MATES UNIVERSAL CONTENTS Co., Ltd. Tanaka A, Erzincioglu TS, Ward SJ, Groves G (2025) Unveiling welfare concerns in Japanese exotic animal cafes. J Vet Med Sci 87(12):1499–1508. https://doi.org/10.1292/jvms.24-0257 Trendeconomy (2023) Commodity data: Reptiles global market. https://trendeconomy.com/data/commodity_h2/010620 Ushine N, Kamitaki A, Suzuki A, Hayama S-I (2024) Assessment of captive environment for oriental small-clawed otters (Aonyx cinereus) in otter cafés in Japan. Animals 14(16):2412. https://doi.org/10.3390/ani14162412 Wenzel SG, Coe JB, Long T, Koerner S, Harvey M, Shepherd ML (2023) Qualitative analysis of small animal veterinarian–perceived barriers to nutrition communication. J Am Vet Med Assoc 262(1):79–87. https://doi.org/10.2460/javma.23.05.0281 World Organization for Animal Health (2024) One Health. https://www.woah.org/en/what-we-do/global-initiatives/one-health/ Additional Declarations The authors declare no competing interests. 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-8735785","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":582731522,"identity":"b63e7f0e-107d-4e2f-a0bd-492062cb6fa3","order_by":0,"name":"Richard Digirolamo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuUlEQVRIiWNgGAWjYFCCxAaGBAYbIIOx8QApWtJAWhqI1ZIAIg6DmcRpMW9Pbn7xcM95u7Xth4G21NhEE9Qic+Zhm0XCs9vJ284kArUcS8ttIKRFQiKxzSDhwO1kswNALYwNh4nWci7Z7PxD4rU0P0g4cMDO7AbRtvA8bGNIOJCcYHYDaEsCUX5hT3/88ccBO3uz8+kPH3yosSGsBQjYJBjAEcoAjSMiAPMHIGFPpOJRMApGwSgYiQAA9lZLmOkugroAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0004-6424-1118","institution":"Independent Researcher","correspondingAuthor":true,"prefix":"","firstName":"Richard","middleName":"","lastName":"Digirolamo","suffix":""}],"badges":[],"createdAt":"2026-01-30 00:46:05","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8735785/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8735785/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101639502,"identity":"feec236f-d596-4216-8b13-ee7e2b652243","added_by":"auto","created_at":"2026-02-02 07:18:08","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":99975,"visible":true,"origin":"","legend":"\u003cp\u003eComparative distribution of medical advice categories provided by LLMs versus human community respondents across three reptile species. The \"AI\" columns represent the calculated mean frequency of advice generated by three LLMs (ChatGPT, Claude, Gemini). The \"Human\" columns represent the \"Best Answer\" selected by users on the Yahoo! Chiebukuro platform. Advice was classified into five hierarchical categories: Category 1 (Vet Referral), Category 2 (Husbandry Check), Category 3 (Home Treatment), Category 4 (Wait \u0026amp; See), and Category 5 (No Actionable Advice).\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8735785/v1/fe96942600fea55afc69b1e4.jpg"},{"id":101754042,"identity":"b14cd452-96d9-4746-8d73-9cc0b11a60d0","added_by":"auto","created_at":"2026-02-03 10:41:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1150869,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8735785/v1/1d93ab10-f42c-4d86-b974-b0787dcb2178.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eAI as Animal Welfare Safety Net: Large Language Models Favor Veterinary Referral Over Online Community Advice for Suspected Reptile Infections\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe landscape of companion animal ownership has diversified significantly, with non-traditional pets\u0026mdash;particularly reptiles\u0026mdash;experiencing a surge in global popularity. Japan ranks as the second-largest reptile importer globally, accounting for 24% of the global market share in 2023 (Trendeconomy, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This trend is exemplified by species such as the leopard gecko (\u003cem\u003eEublepharis macularius\u003c/em\u003e) and the central bearded dragon (\u003cem\u003ePogona vitticeps\u003c/em\u003e) (hereafter bearded dragon), which have become staples in the international pet trade and are top two popular non-native pet reptile species in Japan (Digirolamo, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e). Concurrently, endemic reptile species are also popular pets. The Japanese grass lizard (\u003cem\u003eTakydromus tachydromoides\u003c/em\u003e) is the most popular endemic pet lizard species in Japan (Digirolamo, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e), frequently kept due to its widespread availability and ease of capture or low purchase cost.\u003c/p\u003e \u003cp\u003eHowever, this diversification in herpetoculture has outpaced the availability and awareness of specialized veterinary care. Owners of these species face significant hurdles, including the high cost of exotic animal medicine and a scarcity of qualified practitioners. According to hospital search databases, only 3\u0026ndash;5% of animal hospitals in the Tokyo area treat reptiles, with even fewer options available in non-urban regions (Anicom, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; CalooPet, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; EPARK Pet Life, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eCompounding this infrastructure deficit is a cultural context where exotic animal welfare is often compromised. Japan has faced scrutiny for policies and practices that commodify wildlife, most notably the proliferation of \"exotic animal cafes\". These establishments house species ranging from otters (Ushine et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), owls (Leupen et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) to various reptiles (Sigaud et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Tanaka et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) in conditions often characterized by poor welfare, restricted mobility, and forced human interaction (Tanaka et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This environment potentially normalizes the perception of exotics as \"living toys\" rather than patients requiring medical care, further widening the gap between ownership and responsible veterinary utilization.\u003c/p\u003e \u003cp\u003eThis care gap, combined with the lack of reliable institutional guidance, has necessitated a large-scale behavioral shift. In the absence of accessible professional care, owners are increasingly turning to online communities and social media as their primary source of information (Digirolamo, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e; Lai et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Pienaar \u0026amp; Sturgeon, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Springer et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). While these platforms democratize information by offering rapid access to anecdotal experience, they remain unregulated and un-vetted. This environment is conducive to the propagation of an \"infodemic\"\u0026mdash;an overabundance of information, both accurate and inaccurate, that impedes the ability of individuals to locate trustworthy guidance during critical times.\u003c/p\u003e \u003cp\u003eIn human and animal health, infodemics are a recognized threat (Abuhaloob et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Gallotti et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Richartz et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Shi et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wenzel et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the context of pet reptile welfare, they represent an unquantified but significant multidimensional risk. This risk is most acute when owners seek advice for symptoms indicative of infectious disease/infections. While chronic nutritional or metabolic disorders (e.g., metabolic bone disease) constitute a significant portion of reptile morbidity, this study explicitly limits its scope to suspected infectious etiologies. This focus was selected due to three distinct factors that amplify the potential for welfare compromise.\u003c/p\u003e \u003cp\u003eFirst, infectious states in small ectotherms are often acute and rapidly fatal. Unlike chronic conditions, the \"time-to-intervention\" is critical; delays caused by misinformation carry a disproportionately high welfare cost. Reptiles are stoic and adept at concealing signs of illness, often until they are so severely physiologically compromised that they can no longer maintain the appearance of health. Consequently, observable clinical signs often reflect advanced disease, making timely intervention critical (Broughton \u0026amp; Webb, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Par\u0026eacute; \u0026amp; Lentini, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Schilliger et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This biological trait creates a diagnostic lag, where visible symptoms typically indicate the late stages of a disease rather than its onset. Second, suspected infections are more likely than husbandry issues to prompt owners to attempt unregulated pharmaceutical interventions (e.g., human ointments), creating a risk of iatrogenic harm.\u003c/p\u003e \u003cp\u003eFinally, this issue must be viewed through the lens of \"One Health\"\u0026mdash;a collaborative framework recognizing that the health of people is closely connected to the health of animals and their shared environment (Garcia et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; World Organization for Animal Health, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In herpetoculture, animal welfare and public health are intrinsically linked; for example, the mishandling of waste like diarrhea could increase the risk of zoonotic diseases, most notably reptile-associated salmonellosis (RAS) (Nakadai et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Pees et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Inaccurate online advice that lacks proper guidance on quarantine and hygiene constitutes not only a \"herd health\" failure for the animals but a direct biosafety risk to the human handler and the wider household.\u003c/p\u003e \u003cp\u003eThe welfare impact is, therefore, multidimensional. It begins with the physical dimension of the pathogen but is immediately compounded by an informational dimension. The advice an owner receives online becomes a direct effector of animal welfare. Medically sound advice (e.g., \"See a veterinarian immediately\") could lead to positive outcomes. Conversely, harmful advice (e.g., \"Scrub the mouth with iodine\" or \"Just wait and see\") prolongs suffering, delays in professional treatment, and can lead to fatal outcomes.\u003c/p\u003e \u003cp\u003eDespite anecdotal acknowledgment of this issue, scientific literature lacks a quantitative assessment of this infodemic for pet reptiles. We currently lack data on the prevalence of harmful advice versus sound veterinary referrals on these platforms. Furthermore, it is unknown if the quality of information differs between captive-bred species (like leopard gecko and bearded dragons) and endemic, often wild-caught species (like the Japanese grass lizard), where husbandry data may be sparser.\u003c/p\u003e \u003cp\u003eTo address this gap, this study utilizes data from Yahoo! Chiebukuro, Japan's largest question-answering website (Informatics Research Data Repository, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The platform offers a unique longitudinal dataset of owner concerns across the country. Crucially, its \"Best Answer\" feature\u0026mdash;where the questioner selects the most helpful response\u0026mdash;provides a proxy for the \"accepted\" community consensus, allowing us to evaluate the specific advice that owners are most likely to follow.\u003c/p\u003e \u003cp\u003eAdditionally, the rise of Large Language Models (LLMs) and AI-integrated search introduces a new variable to this ecosystem. As Q\u0026amp;A sites begin to offer AI-generated answers (D\u0026rsquo;Angelo, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; LY Corporation, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), it is crucial to understand whether these automated systems mitigate or exacerbate the welfare risk compared to human community advice. Previous research indicates that LLMs (ChatGPT and Gemini) provided desirable answers for clean, best-case scenario health questions regarding leopard gecko (Digirolamo, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e). This study expands that investigation by using raw, natural-language questions posted to Yahoo! Chiebukuro to evaluate three specific LLMs: Claude Sonnet 4.5 (hereafter Claude), ChatGPT-5.2 (hereafter ChatGPT), and Gemini 3 Pro (hereafter Gemini).\u003c/p\u003e \u003cp\u003eThese models were selected to represent different capabilities of the AI landscape: Claude is currently integrated directly into Yahoo! Chiebukuro as a user-selected option; ChatGPT represents the most globally popular LLM (StatCounter Global Stats, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) and represents one of the highest benchmarks including scientific reasoning (GPQA Diamond: 92.4%) and multilingual Q\u0026amp;A (MMMLU: 89.6%); and Gemini represents a state-of-the-art multimodal model, also consistently achieving top-tier benchmarks including scientific reasoning (GPQA Diamond: 91.9%) and multilingual Q\u0026amp;A (MMMLU: 91.8%) (Google, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis study aims to fill the knowledge gap regarding the safety of online veterinary advice by systematically analyzing a longitudinal dataset of owner questions and community-provided answers from a major Japanese Q\u0026amp;A website. We quantify the prevalence and nature of treatment advice given for symptoms of suspected infectious disease across three popular reptile species. Our primary objectives are:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTo classify the quality and safety of \"accepted\" community advice for infectious symptoms.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTo generate advice using three prominent LLMs and compare their safety classification against human advice.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTo compare the distribution of advice quality across the three different species.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eWe hypothesize that a significant proportion of community-sourced advice for suspected reptile infections is ethically and medically inappropriate and poses a direct, quantifiable risk to animal welfare compared to the safety baselines established by modern LLMs.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Data Source and Acquisition\u003c/h2\u003e \u003cp\u003eData were obtained from Yahoo! Chiebukuro (Yahoo! JAPAN), the largest publicly accessible Japanese question-and-answer (Q\u0026amp;A) platform. All threads were retrieved from the \u0026ldquo;Pet\u0026rdquo; category using commercial web-scraping software (Silk Script Inc., 2025). Species-specific subcorpora were constructed using commonly used Japanese terms and abbreviations: bearded dragon (\u0026ldquo;フトアゴ\u0026rdquo;), leopard gecko (\u0026ldquo;ヒョウモントカゲモドキ\u0026rdquo; and \u0026ldquo;レオパ\u0026rdquo;), and Japanese grass lizard (\u0026ldquo;カナヘビ\u0026rdquo;). The collection window spanned 01 January 2010 to 30 September 2025, corresponding to the earliest retrievable posts available through the scraping software. For each thread, the question text, posting metadata, and the community-selected \u0026ldquo;Best Answer\u0026rdquo; were extracted. Only publicly available text was collected; no attempt was made to identify users, and all data were analyzed in aggregate.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Filtering and eligibility criteria\u003c/h2\u003e \u003cp\u003eA multi-stage screening procedure was applied to derive the analytical sample. First, each species subcorpus was filtered using a curated dictionary of Japanese keywords suggestive of potential infectious disease, infected wounds, or localized infections (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), developed from husbandry and clinical references for the focal species (Koieyama, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Miyazaki, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Takenaka, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Second, duplicate threads and threads without an original-poster-selected \u0026ldquo;Best Answer\u0026rdquo; were removed. Third, the remaining threads were manually reviewed to confirm that the original post described current health signs and was actively seeking management advice. Threads focused primarily on breeding, baseline husbandry without illness signs, or retrospective discussion of death causes were excluded. Threads in which an infection had already been diagnosed, but the user requested additional guidance on care or treatment, were retained, as these posts represent decision-making regarding infectious disease management.\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 \u003cp\u003eKeyword Dictionary for Infectious Disease Filtering\u003c/p\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\" colname=\"c1\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKey terms in Japanese\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEnglish translation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGeneral Signs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e腫れ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSwelling\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e爛れ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSore / Erosion\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e死*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDeath\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eぐったり\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLimp / Lethargic\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e痩せ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThin\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e病気\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDisease/Illness/Sickness\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePathogens\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e菌*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBacteria/germs\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eカビ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFungus\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eダニ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMites\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e寄生虫\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eParasite\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eウイルス\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVirus\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e感染\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInfection\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExcretions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e膿\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePus\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e下痢\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDiarrhea\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e吐*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVomit\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e血便\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBlood in stool\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e鼻水\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRunny nose\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eよだれ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDrooling/Saliva\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eコクシ*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCoccidia\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eクリプト*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCryptosporidiosis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eマウスロット\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMouth rot\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e炎*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInflammation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e口を|口が\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMouth (sentence start)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e呼吸\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBreathing\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eくしゃみ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSneezing\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e*The term \u0026ldquo;死\u0026rdquo; covers death/dying\u003c/p\u003e \u003cp\u003e*The term \u0026ldquo;菌\u0026rdquo; covers general term of germs including bacteria\u003c/p\u003e \u003cp\u003e*The term \"クリプト\" covers \"クリプトスポリジウム\"(Cryptosporidiosis)\u003c/p\u003e \u003cp\u003e*The term \"吐\" covers all common forms of vomit/vomiting/nausea\u003c/p\u003e \u003cp\u003e*The term \"コクシ\" covers \"コクシジウム\"(Coccidia)\u003c/p\u003e \u003cp\u003e*The term \"炎\" covers various conditions, including general inflammation, pneumonia, and eye infections\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Advice classification framework\u003c/h2\u003e \u003cp\u003eThe primary unit of analysis was the \u0026ldquo;Best Answer\u0026rdquo; selected by the original poster. Each answer was coded into one of five mutually exclusive categories representing the primary recommended action:\u003c/p\u003e \u003cp\u003e(1) Vet-direct: explicit recommendation to seek professional veterinary assessment or treatment as the primary action.\u003c/p\u003e \u003cp\u003e(2) Husbandry-check: advice to evaluate or correct environmental or routine husbandry parameters (e.g., temperature, humidity, lighting, feeding schedule) within standard care.\u003c/p\u003e \u003cp\u003e(3) Home-treat: recommendation of active non-veterinary interventions (e.g., topical human medications or home remedies).\u003c/p\u003e \u003cp\u003e(4) Wait-and-see: recommendation to observe without intervention or minimization of the condition (e.g., \u0026ldquo;it will resolve on its own\u0026rdquo;).\u003c/p\u003e \u003cp\u003e(5) No action: non-actionable or non-substantive guidance (e.g., sympathy-only responses or irrelevant content).\u003c/p\u003e \u003cp\u003eWhen multiple recommendations were present, a hierarchical decision rule was applied to assign the category representing the most immediate action likely to influence welfare outcomes. Answers were coded as Vet-direct only when veterinary consultation was advised as the first-line action. If veterinary consultation was explicitly contingent on prior non-veterinary steps (e.g., \u0026ldquo;try home treatment first and see a veterinarian if it fails\u0026rdquo;), the response was coded according to the initial action (Home-treat, Husbandry-check, or Wait-and-see). This approach distinguishes between immediate referral and delayed referral from an animal welfare risk perspective.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. AI response generation\u003c/h2\u003e \u003cp\u003eFor each eligible thread, the question text (excluding human answers) was submitted to three large language models (LLMs): ChatGPT (GPT-5.2 Thinking; OpenAI), Claude (Claude Sonnet 4.5; Anthropic), and Gemini (Gemini 3.0 Pro; Google). All models were accessed via their standard web interfaces using free-tier accounts to approximate public accessibility. To reduce carry-over effects, new isolated sessions were used for each batch under Temporary Chat function for ChatGPT and Gemini; Incognito Chat for Claude. Questions were processed in batches (approximately 80 questions) to remain within context window constraints, with total token counts checked using the OpenAI Tokenizer (OpenAI, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). A standardized Japanese prompt instructed the model to provide the most important advice and to return outputs in a two-column Markdown table (ID, Answer). The full prompt text (Japanese original with English translation) is provided in supplementary material 1. AI outputs were subsequently classified using the same coding framework applied to human Best Answers. The dates of model access were from 15\u0026ndash;26 December 2025.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Statistical Analysis\u003c/h2\u003e \u003cp\u003eAnalyses focused on the safety-relevant outcome of veterinary referral. The primary endpoint was binary: Vet-direct (Category 1) versus non-referral (Categories 2\u0026ndash;5). For each species, descriptive statistics were used to summarize the distribution of advice categories for human Best Answers and for each LLM (ChatGPT, Claude, and Gemini).\u003c/p\u003e \u003cp\u003eBecause each thread yielded matched observations (one human Best Answer and one response from each LLM), human\u0026ndash;AI comparisons of Vet-direct recommendations were conducted at the thread level. For each species and for each model separately, paired differences in Vet-direct proportions between human Best Answers and LLM responses were evaluated using two-sided McNemar\u0026rsquo;s tests with continuity correction, using discordant pairs only. Statistical significance was defined as α\u0026thinsp;=\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003eAmong human Best Answers, Vet-direct proportions were compared across species using a chi-square test of independence, and effect size was summarized using Cram\u0026eacute;r\u0026rsquo;s V. All analyses were performed in Microsoft Excel.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eFollowing the scraping, filtering, and cleaning process, the final dataset comprised 168 threads for bearded dragons, 431 for leopard geckos, and 93 for Japanese grass lizards.\u003c/p\u003e \u003cp\u003eAcross all species, veterinary referral (Category 1: Vet-direct) was recommended more frequently by large language models (LLMs) than by human Best Answers. Model-specific veterinary referral rates are shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e; paired thread-level comparisons are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Bearded dragon\u003c/h2\u003e \u003cp\u003eFor bearded dragons (N\u0026thinsp;=\u0026thinsp;168), human Best Answers recommended veterinary consultation in 41.07% of cases (69/168). The most common alternative human recommendation was husbandry correction (Category 2: 34.52%, 58/168), followed by home treatment (Category 3: 11.31%, 19/168). In contrast, LLM responses showed higher vet-direct rates across all models (Claude: 61.90%, 104/168; ChatGPT: 79.76%, 134/168; Gemini: 90.48%, 152/168).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Leopard gecko\u003c/h2\u003e \u003cp\u003eFor leopard geckos (N\u0026thinsp;=\u0026thinsp;431), human Best Answers recommended veterinary referral in 34.34% of threads (148/431). Non-referral advice was distributed across husbandry checks (27.15%, 117/431), home treatment (14.62%, 63/431), and watchful waiting (13.92%, 60/431). LLM outputs recommended vet-direct action more frequently than humans (Claude: 61.25%, 264/431; ChatGPT: 75.41%, 325/431; Gemini: 77.96%, 336/431).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Japanese grass lizard\u003c/h2\u003e \u003cp\u003eFor Japanese grass lizards (N\u0026thinsp;=\u0026thinsp;93), veterinary referral was recommended in 23.66% of human Best Answers (22/93), with comparatively high use of home treatment (31.18%, 29/93) and no actionable guidance (Category 5: 17.20%, 16/93). LLM responses were more risk-averse, recommending vet-direct action in 66.67%\u0026ndash;95.70% of cases (Claude: 62/93; ChatGPT: 82/93; Gemini: 89/93).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Paired human\u0026ndash;AI comparisons\u003c/h2\u003e \u003cp\u003eBecause each thread generated matched observations (one human Best Answer and one response per LLM), human\u0026ndash;AI differences in veterinary referral were evaluated using McNemar\u0026rsquo;s tests at the thread level (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Across all three species and all three models, LLM outputs were significantly more likely than human Best Answers to recommend immediate veterinary consultation (Vet-direct) (all p\u0026thinsp;\u0026le;\u0026thinsp;9.58\u0026times;10⁻⁶). Discordant-pair counts consistently showed more threads where the LLM recommended Vet-direct when the human Best Answer did not (c) than the reverse (b), indicating a consistent shift toward veterinary referral in LLM-generated guidance compared with community-selected human responses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Comparison of human advice across species\u003c/h2\u003e \u003cp\u003eAmong human Best Answers, the probability of receiving Vet-direct advice differed by species (χ\u0026sup2;(2, N\u0026thinsp;=\u0026thinsp;692)\u0026thinsp;=\u0026thinsp;8.05, p\u0026thinsp;=\u0026thinsp;0.018; Cram\u0026eacute;r\u0026rsquo;s V\u0026thinsp;=\u0026thinsp;0.108). Bearded dragons and leopard geckos received veterinary referral advice more often (41.07% and 34.34%, respectively) than Japanese grass lizards (23.66%), which also showed the highest proportion of home-treatment recommendations (31.18%).\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 \u003cp\u003eComparative distribution of advice categories (%) between LLMs (mean) and Human respondents.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecies \u0026amp; Source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCat. 1\u003c/p\u003e \u003cp\u003e(Vet)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCat. 2\u003c/p\u003e \u003cp\u003e(Husbandry)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCat. 3\u003c/p\u003e \u003cp\u003e(Home-treat)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCat. 4\u003c/p\u003e \u003cp\u003e(Wait\u0026amp;See)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eCat. 5\u003c/p\u003e \u003cp\u003e(No action)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBearded Dragon\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLLMs Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuman (Yahoo!)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e5.95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLeopard gecko\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLLMs Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e431\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuman (Yahoo!)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e431\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e9.98\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eJapanese grass lizard\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLLMs Mean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e83.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e3.58\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuman (Yahoo!)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e26.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e31.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e17.2\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=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of veterinary referral rates (Category 1) across three LLMs.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecies\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClaude\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChatGPT\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGemini\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAI Mean\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBearded Dragon (N\u0026thinsp;=\u0026thinsp;168)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e61.90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e79.76%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e90.48%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e77.38%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeopard gecko (N\u0026thinsp;=\u0026thinsp;431)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e61.25%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75.41%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e77.96%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e71.54%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapanese grass lizard (N\u0026thinsp;=\u0026thinsp;93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.67%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e88.17%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e95.70%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e83.51%\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=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThread-level paired comparison of Vet-direct recommendations between human Best Answers and LLM outputs (McNemar\u0026rsquo;s test).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecies\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eComparison\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHuman Vet-direct n (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLLM Vet-direct n (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHuman Yes / LLM No (b)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHuman No / LLM Yes (c)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMcNemar χ\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBearded dragon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Claude\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69 (41.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e104 (61.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e19.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e9.58E-06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBearded dragon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs ChatGPT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69 (41.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e134 (79.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e53.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e3.02E-13\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBearded dragon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Gemini\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69 (41.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e152 (90.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e77.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.48E-18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeopard gecko\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e431\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Claude\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e148 (34.34)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e264 (61.25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e81.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.64E-19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeopard gecko\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e431\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs ChatGPT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e148 (34.34)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e325 (75.41)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e192\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e149.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2.08E-34\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeopard gecko\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e431\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Gemini\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e148 (34.34)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e336 (77.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e199\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e166.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e4.26E-38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapanese grass lizard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Claude\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (23.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e62 (66.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e29.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e6.36E-08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapanese grass lizard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs ChatGPT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (23.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e82 (88.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e49.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.77E-12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJapanese grass lizard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHuman vs Gemini\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (23.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e89 (95.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e65.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e7.43E-16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003eNotes:\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eb\u0026thinsp;=\u0026thinsp;discordant threads where the human Best Answer recommended Vet-direct but the LLM did not.\u003c/p\u003e \u003cp\u003ec\u0026thinsp;=\u0026thinsp;discordant threads where the human Best Answer did not recommend Vet-direct but the LLM did.\u003c/p\u003e \u003cp\u003eContinuity-corrected McNemar statistic: χ\u0026sup2; = (|b\u0026thinsp;\u0026minus;\u0026thinsp;c| \u0026minus; 1)\u0026sup2; / (b\u0026thinsp;+\u0026thinsp;c) (df\u0026thinsp;=\u0026thinsp;1).\u003c/p\u003e \u003cp\u003eP-values are computed using the chi-square distribution with 1 degree of freedom.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.6. Model-to-model variation\u003c/h2\u003e \u003cp\u003eWithin LLM outputs, Gemini and ChatGPT generally produced higher Vet-direct rates than Claude across all species (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). However, even the least risk-averse model (Claude) recommended veterinary referral more frequently than human Best Answers for every species\u0026rsquo; veterinary care. Claude produced the lowest Vet-direct proportions across species (61.25%\u0026ndash;66.67%), but still exceeded human Best Answers in every species.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.7. Distribution of advice categories\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e Comparative distribution of medical advice categories provided by LLMs versus human community respondents across three reptile species. The \"AI\" columns represent the calculated mean frequency of advice generated by three LLMs (ChatGPT, Claude, Gemini). The \"Human\" columns represent the \"Best Answer\" selected by users on the Yahoo! Chiebukuro platform. Advice was classified into five hierarchical categories: Category 1 (Vet Referral), Category 2 (Husbandry Check), Category 3 (Home Treatment), Category 4 (Wait \u0026amp; See), and Category 5 (No Actionable Advice).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.8. Evaluation of Zoonotic Risk and Biosafety Guidance\u003c/h2\u003e \u003cp\u003eA qualitative review was conducted on all human \"Best Answers\" and AI-generated responses to evaluate the prevalence of zoonotic safety and hygiene guidance. Guidance regarding human biosafety (handwashing/disinfection, waste handling) was found to be rare across all platforms and models.\u003c/p\u003e \u003cp\u003eAmong the 692 human-sourced \"Best Answers,\" only 0.43% (n\u0026thinsp;=\u0026thinsp;3) mentioned handwashing or hygiene for the purpose of safeguarding human health, and only 0.14% (n\u0026thinsp;=\u0026thinsp;1) explicitly identified the risk of salmonella. Similarly, across the 2,076 AI-generated responses (n\u0026thinsp;=\u0026thinsp;692 questions x 3 models), the mention rate for human-centric hygiene protocols was 0.14% (n\u0026thinsp;=\u0026thinsp;3). Notably, the cases for the AI were limited to a single specific query that appeared in both the human and AI datasets regarding whether a diagnosed pinworm infection in a bearded dragon could be transmitted to humans. In this isolated instance, both the human community and all three LLMs correctly recommended handwashing. None of the LLMs proactively cautioned against the risk of salmonella.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.1. The \"Care Gap\" and the Informational Infodemic\u003c/h2\u003e \u003cp\u003eTo our knowledge, this is the first quantitative assessment of the \"information ecosystem\" available to reptile owners in Japan during medical crises. The results confirm that community-sourced advice frequently fails to recommend veterinary intervention for acute infectious symptoms, posing a significant welfare risk. This failure is heavily stratified by species. While LLMs maintained a consistent \"safety baseline\" (recommending veterinary care in 71.54%\u0026ndash;83.51% of cases regardless of species), human advice quality fluctuated markedly. The analysis reveals a valuation hierarchy: owners of the endemic Japanese grass lizard were significantly less likely to receive veterinary referrals (23.66%) compared to owners of Leopard geckos (34.34%) and Bearded dragons (41.07%).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Economic barriers and species-based differences in referral advice\u003c/h2\u003e \u003cp\u003eThe stark disparity in advice quality suggests a deep-seated economic bias. Even for the bearded dragons\u0026mdash;the \"best performing\" species in the human dataset\u0026mdash;fewer than half of the owners were advised to seek professional medical assistance for potentially fatal symptoms. This low baseline is likely driven by an \"economic inversion\" of value, where the cost of veterinary intervention often exceeds the replacement cost of the animal. In the Japanese market, a juvenile leopard gecko or bearded dragon of a standard morph can frequently be purchased for 5,000 to 15,000 JPY (approx. \u003cspan\u003e$\u003c/span\u003e35\u0026ndash;\u003cspan\u003e$\u003c/span\u003e100 USD). In contrast, a single veterinary consultation, combined with diagnostics (e.g., bloodwork, X-rays) and treatment, can easily exceed, or be comparable to, these prices. For the Japanese grass lizard, often sold for ~\u0026thinsp;1,000 JPY or caught freely in the wild, this economic disparity is more extreme. When a pet\u0026rsquo;s market value is lower than the price of its medical treatment, it risks being treated as readily replaceable, leading owners (and advice-givers) to prefer low-cost home remedies (Irvine, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). This is evidenced by the high prevalence of Category 3 (Home Treatment) advice for Japanese grass lizards (31.18%), which exceeded the rate of veterinary referrals.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.3. AI as a Welfare \"Safety Net\"\u003c/h2\u003e \u003cp\u003eA critical finding of this study is the inverse relationship between human and AI performance regarding the endemic species. While human advice quality degraded for the Japanese grass lizard, LLMs effectively compensated for this bias, recommending veterinary care in 83.51% (AI Mean) of cases. This suggests that LLMs, trained on vast global datasets, do not share the localized economic biases that devalue native wildlife. Consequently, the integration of AI-generated answers into Q\u0026amp;A platforms could serve as a vital \"welfare safety net,\" guiding novice owners of undervalued species toward professional care when human communities fail to do so.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e4.4. The One Health Blind Spot: Zoonotic Risks and Proactive Triage\u003c/h2\u003e \u003cp\u003eA critical finding of this study is the near-total absence of proactive biosafety guidance. While the LLMs and human respondents correctly identified handwashing as a necessity when explicitly queried about pinworm transmission, they failed to offer similar warnings for symptoms that may increase the risk of Salmonella transmission or other enteric pathogens. In the context of \"One Health,\" this represents a dangerous failure in triage logic. Within herpetoculture, animal welfare and public health are inextricably linked; the mishandling of infectious waste during a medical crisis is a primary driver of Reptile-Associated Salmonellosis (RAS) (Nakadai et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Pees et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIt could be argued that the lack of biosafety warnings in the AI-generated responses was a byproduct of the specific prompt used in this study, which instructed the models to \"emphasize the most important advice\" regarding the reptile's health. Under a narrow clinical interpretation, the AI prioritized the survival of the animal (the patient) over the safety of the handler. However, we contend that the open-ended nature of this prompt was appropriate and necessary for two reasons.\u003c/p\u003e \u003cp\u003eFirst, it accurately reflects the \"natural language\" behavior of novice owners, who typically seek solutions for the animal\u0026rsquo;s visible distress rather than inquiring about their own potential exposure. Second, a truly robust medical AI should operate under a \"safety-first\" paradigm that recognizes the human-animal interface as a singular clinical unit. If an LLM recognizes an infection risk to the animal but fails to trigger a corresponding warning for the handler, its utility as a public health tool remains limited. The \"0.14% success rate\" observed here suggests that current LLMs are reactive rather than proactive regarding zoonosis; they provide safety advice only when the user possesses enough prior knowledge to ask a specific question. This creates a \"knowledge trap\" where the most vulnerable owners\u0026mdash;those unaware of RAS risks\u0026mdash;are the least likely to receive the hygiene guidance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e4.5. The Paradox of Choice and Confirmation Bias\u003c/h2\u003e \u003cp\u003eIntegrating AI answers alongside human responses introduces the risk of conflicting advice. If an LLM suggests a veterinary visit while a human user suggests a convenient \"home remedy,\" owners may be psychologically predisposed to follow the human advice due to confirmation bias\u0026mdash;the tendency to favor information that aligns with one's desire to avoid costs/effort. Therefore, simply displaying AI answers is insufficient. Q\u0026amp;A platforms should implement Context-Aware Warning Systems. If a thread contains keywords indicative of illness, an automatic banner should explicitly warn users that \"home remedies for these symptoms carry high risks,\" prioritizing clinical guidelines over anecdotal suggestions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e4.6. Limitations\u003c/h2\u003e \u003cp\u003eSeveral limitations must be acknowledged. First, because this was an observational study using publicly posted threads, we could not control unmeasured confounding (e.g., owner resources, severity of clinical signs, or prior veterinary access). Second, the data is derived exclusively from \u003cem\u003eYahoo! Chiebukuro\u003c/em\u003e. While the anonymity of the platform likely yields more honest data regarding \"home remedy\" attempts than face-to-face interviews (mitigating social desirability bias\u0026mdash;answering in a way they believe is \"correct\" or \"ethical\" (Krumpal, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2011\u003c/span\u003e)), the population may not fully represent all reptile owners. Third, our classification focused on safety (process) rather than diagnostic accuracy (outcome). We did not attempt to verify if the \"home remedy\" suggested might have incidentally worked, nor did we diagnose the animals ourselves. However, this distinction is critical: treating a suspected infectious disease requires professional diagnostics. Even if a layperson's guess (\"wait and see\") happens to be correct in a specific instance, the process of relying on unverified internet advice is inherently dangerous. The welfare risk lies in the lack of professional triage, not solely in the outcome of a specific case. Finally, regarding LLM consistency, while all models outperformed humans, Gemini and ChatGPT were generally stricter than Claude. Thus, usage of models not tested or paid tiers or future iterations may show different outcomes. However, even the \"least strict\" model consistently outperformed the human average, confirming that the \"AI Safety Net\" effect holds true despite model variation. Additionally, all advice categorization was conducted by a single coder; inter-rater reliability was not assessed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e4.7. Future Directions: Bridging the Gap\u003c/h2\u003e \u003cp\u003eFuture initiatives must focus on bridging the gap between digital advice and physical veterinary access. One immediate solution is the promotion of telemedicine services that are available in Japan. Online veterinary consultations, while not free, are significantly more affordable than physical hospital visits and eliminate the stress of transport for the animal (mirpet, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Pet-online hospital24, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). For owners of \"low-value\" species who are hesitant to pay full clinic fees, telemedicine offers a \"middle-ground\" triage step that is safer than free Q\u0026amp;A sites such as Yahoo! Chiebukuro but more accessible than full veterinary visit. Platforms and AI tools should be programmed to suggest legitimate telemedicine providers as a viable alternative to home remedies.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThis study provides the first quantitative evidence of a digital \"infodemic\" affecting the welfare of pet reptiles in Japan. Our analysis confirms that while online peer-to-peer platforms offer community support, they are structurally inadequate for medical crisis management. This inadequacy is heavily stratified by species, revealing a pervasive \"valuation hierarchy\" where endemic species like the Japanese grass lizard are significantly less likely to receive veterinary referrals than higher cost \"commodity\" pets. The recommendation of high-risk home remedies over professional care for lower-value species highlights a \"care gap\" where the low economic replacement cost of an animal may lead to a dangerous devaluation of its medical needs.\u003c/p\u003e \u003cp\u003eConversely, Large Language Models (LLMs) demonstrated a significant \"AI Welfare Safety Net\" effect. Across all models\u0026mdash;Claude, ChatGPT, and Gemini\u0026mdash;the systems maintained a high-safety baseline that effectively ignored the economic or cultural biases compromising human advice. While significant internal differences in referral strictness exist, the overall performance of LLMs suggests they are superior tools for preliminary clinical triage. However, the \"0.14% success rate\" in proactive biosafety guidance identifies a critical One Health blind spot. Neither human nor AI sources proactively addressed the risks of zoonotic transmission, such as Reptile-Associated Salmonellosis (RAS), unless specifically prompted.\u003c/p\u003e \u003cp\u003eWe conclude that while online peer-to-peer platforms provide social support, they are structurally unreliable for managing suspected infectious disease in reptiles, and their safety performance varies by species. LLMs consistently recommended veterinary referral more frequently than community-selected human answers, suggesting a potential welfare \u0026ldquo;safety net\u0026rdquo; effect for undervalued or endemic species. Future risk-reduction strategies should prioritize (1) context-aware platform warnings that discourage high-risk home remedies while reinforcing veterinary triage, (2) improved visibility of legitimate telemedicine pathways as a financially accessible bridge to clinical care, and (3) AI outputs that integrate proactive One Health guidance (e.g., hygiene and waste-handling precautions) alongside animal-focused recommendations.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eSupporting information captions:\u003c/h2\u003e \u003cp\u003eThe data presented in this study are derived from publicly accessible content on Yahoo! Chiebukuro (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://chiebukuro.yahoo.co.jp/\u003c/span\u003e\u003cspan address=\"https://chiebukuro.yahoo.co.jp/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e).\u003c/span\u003e Due to copyright restrictions regarding the redistribution of raw user-generated content, the full-text dataset cannot be shared directly. However, a complete list of URLs for all Q\u0026amp;A threads analyzed in this study, along with summarized English translations focusing on the clinical signs and the corresponding \"Best Answers,\" are available in Supplementary Material 2 (bearded dragon), Supplementary Material 3 (leopard gecko), and Supplementary Material 4 (Japanese grass lizard). Note that accessing Yahoo! Japan websites from outside Japan currently requires a VPN set to a Japan location.\u003c/p\u003e \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAbuhaloob L, Purnat TD, Tabche C, Atwan Z, Dubois E, Rawaf S (2024) Management of infodemics in outbreaks or health crises: A systematic review. Front Public Health 12:1343902. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpubh.2024.1343902\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2024.1343902\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnicom (2025) Search for animal hospitals in Tokyo for lizards. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.anicom-ah.com/tokyo/lizard\u003c/span\u003e\u003cspan address=\"https://www.anicom-ah.com/tokyo/lizard\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBroughton C, Webb KL (2022) Diagnostic clinical pathology of the bearded dragon (Pogona vitticeps). Veterinary Clin North America: Exotic Anim Pract 25(3):713\u0026ndash;734. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cvex.2022.06.002\u003c/span\u003e\u003cspan address=\"10.1016/j.cvex.2022.06.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalooPet (2025) Reptile hospital search (Tokyo). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pet.caloo.jp/hospitals/search/13/a10/all\u003c/span\u003e\u003cspan address=\"https://pet.caloo.jp/hospitals/search/13/a10/all\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD\u0026rsquo;Angelo A (2023), February 1 Poe. Quora Blog. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://quorablog.quora.com/Poe-1\u003c/span\u003e\u003cspan address=\"https://quorablog.quora.com/Poe-1\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDigirolamo R (2025a) Understanding pet reptile preferences in Japan: An analysis using Yahoo! Chiebukuro and Google Trends. Herpetological J 35(2):146\u0026ndash;154. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.33256/35.2.146154\u003c/span\u003e\u003cspan address=\"10.33256/35.2.146154\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDigirolamo R (2025b) Accuracy and completeness of AI chatbot responses for leopard gecko (Eublepharis macularius) husbandry: An exploratory comparison of ChatGPT-4o and Gemini 2.5 Pro. J Appl Anim Welfare Sci. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10888705.2025.2606681\u003c/span\u003e\u003cspan address=\"10.1080/10888705.2025.2606681\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEPARK Pet Life (2025) Reptile animal hospital search in Tokyo. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://petlife.asia/hospital/search/tokyo/reptiles/\u003c/span\u003e\u003cspan address=\"https://petlife.asia/hospital/search/tokyo/reptiles/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGallotti R, Valle F, Castaldo N, Sacco P, De Domenico M (2020) Assessing the risks of \u0026lsquo;infodemics\u0026rsquo; in response to COVID-19 epidemics. Nat Hum Behav 4(12):1285\u0026ndash;1293. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41562-020-00994-6\u003c/span\u003e\u003cspan address=\"10.1038/s41562-020-00994-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGarcia MR et al (2025) One welfare: Bibliometric review of world literature. Front Veterinary Sci 12:1627981. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fvets.2025.1627981\u003c/span\u003e\u003cspan address=\"10.3389/fvets.2025.1627981\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoogle (2025), November 18 A new era of intelligence with Gemini 3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://blog.google/products/gemini/gemini-3/#gemini-3\u003c/span\u003e\u003cspan address=\"https://blog.google/products/gemini/gemini-3/#gemini-3\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInformatics Research Data Repository (2024) Yahoo! Dataset (in Japanese). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nii.ac.jp/dsc/idr/yahoo/chiebkr3/Y_chiebukuro.html\u003c/span\u003e\u003cspan address=\"https://www.nii.ac.jp/dsc/idr/yahoo/chiebkr3/Y_chiebukuro.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIrvine L (2003) The problem of unwanted pets: A case study in how institutions \u0026lsquo;think\u0026rsquo; about clients\u0026rsquo; needs. Soc Probl 50(4):550\u0026ndash;566. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1525/sp.2003.50.4.550\u003c/span\u003e\u003cspan address=\"10.1525/sp.2003.50.4.550\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoieyama H (2019) The complete book of leopard gecko health and disease (in Japanese). SEIBUNDO SHINKOSHA Publishing\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrumpal I (2011) Determinants of social desirability bias in sensitive surveys: A literature review. Qual Quant 47(4):2025\u0026ndash;2047. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11135-011-9640-9\u003c/span\u003e\u003cspan address=\"10.1007/s11135-011-9640-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLai N, Khosa DK, Jones-Bitton A, Dewey CE (2021) Pet owners\u0026rsquo; online information searches and the perceived effects on interactions and relationships with their veterinarians. Veterinary Evid 6(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18849/ve.v6i1.345\u003c/span\u003e\u003cspan address=\"10.18849/ve.v6i1.345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeupen BTC, Wakao K, Asakawa Y, Eaton JA, Bruslund S (2024) Live owls in Japanese pet stores and caf\u0026eacute;s: Volumes, species, and impediments to effective trade monitoring. J Asia-Pacific Biodivers 17(3):513\u0026ndash;524. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.japb.2024.03.006\u003c/span\u003e\u003cspan address=\"10.1016/j.japb.2024.03.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLY Corporation (2024) Yahoo! Chiebukuro starts providing \u0026lsquo;Minna no Chiebukuro\u0026rsquo; AI answer function. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.lycorp.co.jp/ja/news/release/009337/\u003c/span\u003e\u003cspan address=\"https://www.lycorp.co.jp/ja/news/release/009337/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003emirpet (2025) Online veterinary services. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://mirpet.co.jp/\u003c/span\u003e\u003cspan address=\"https://mirpet.co.jp/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiyazaki T (ed) (2022) Bearded dragon care bible (in Japanese). MATES UNIVERSAL CONTENTS Co., Ltd.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakadai A et al (2005) Prevalence of Salmonella spp. in pet reptiles in Japan. J Vet Med Sci 67(1):97\u0026ndash;101. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1292/jvms.67.97\u003c/span\u003e\u003cspan address=\"10.1292/jvms.67.97\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOpenAI (2025) Tokenizer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://platform.openai.com/tokenizer\u003c/span\u003e\u003cspan address=\"https://platform.openai.com/tokenizer\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePar\u0026eacute; JA, Lentini AM (2010) Reptile geriatrics. Veterinary Clinics of North America. Exotic Anim Pract 13(1):15\u0026ndash;25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cvex.2009.09.003\u003c/span\u003e\u003cspan address=\"10.1016/j.cvex.2009.09.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePees M et al (2023) Salmonella in reptiles: A review of occurrence. Front Cell Dev Biology 11:1251036. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fcell.2023.1251036\u003c/span\u003e\u003cspan address=\"10.3389/fcell.2023.1251036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePet-online hospital24 (2025) Online pet consultations. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pet-online24.com/\u003c/span\u003e\u003cspan address=\"https://pet-online24.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePienaar EF, Sturgeon DJE (2024) Exotic pet owners\u0026rsquo; preferences for different ectothermic taxa are based on species traits and purchase prices in the United States. NeoBiota 91:1\u0026ndash;27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3897/neobiota.91.109403\u003c/span\u003e\u003cspan address=\"10.3897/neobiota.91.109403\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRichartz ER, Hodgkiss BA, Black-Ocken NC, Fuentes RA, Looper JS, Withers SS (2024) Characterization of the dissemination of canine cancer misinformation on YouTube. Vet Comp Oncol 22(3):359\u0026ndash;366. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/vco.12977\u003c/span\u003e\u003cspan address=\"10.1111/vco.12977\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchilliger L, Vergneau-Grosset C, Desmarchelier MR (2020) Clinical reptile behavior. Veterinary Clin North America: Exotic Anim Pract 24(1):175\u0026ndash;195. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cvex.2020.09.008\u003c/span\u003e\u003cspan address=\"10.1016/j.cvex.2020.09.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi R, Jia X, Hu Y, Wang H (2024) The media risk of infodemic in public health emergencies: Consequences and mitigation approaches. PLoS ONE 19(9):e0308080. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0308080\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0308080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSigaud M, Kitade T, Sarabian C (2023) Exotic animal caf\u0026eacute;s in Japan: A new fashion with potential implications for biodiversity, global health, and animal welfare. Conserv Sci Pract 5(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/csp2.12867\u003c/span\u003e\u003cspan address=\"10.1111/csp2.12867\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilk Script Inc (2025) Web scraping services. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.silk-s.jp/sw75.html\u003c/span\u003e\u003cspan address=\"https://www.silk-s.jp/sw75.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpringer S, Lund TB, Corr SA, Sand\u0026oslash;e P (2024) Does \u0026lsquo;Dr. Google\u0026rsquo; improve discussion and decisions in small animal practice? Front Veterinary Sci 11:1417927. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fvets.2024.1417927\u003c/span\u003e\u003cspan address=\"10.3389/fvets.2024.1417927\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStatCounter Global Stats (2025) AI chatbot market share worldwide. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gs.statcounter.com/ai-chatbot-market-share#monthly-202503-202508\u003c/span\u003e\u003cspan address=\"https://gs.statcounter.com/ai-chatbot-market-share#monthly-202503-202508\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTakenaka S (2023) Beginner\u0026rsquo;s guide to Kanahebi: Proper care and rearing (in Japanese). MATES UNIVERSAL CONTENTS Co., Ltd.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTanaka A, Erzincioglu TS, Ward SJ, Groves G (2025) Unveiling welfare concerns in Japanese exotic animal cafes. J Vet Med Sci 87(12):1499\u0026ndash;1508. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1292/jvms.24-0257\u003c/span\u003e\u003cspan address=\"10.1292/jvms.24-0257\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrendeconomy (2023) Commodity data: Reptiles global market. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://trendeconomy.com/data/commodity_h2/010620\u003c/span\u003e\u003cspan address=\"https://trendeconomy.com/data/commodity_h2/010620\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUshine N, Kamitaki A, Suzuki A, Hayama S-I (2024) Assessment of captive environment for oriental small-clawed otters (Aonyx cinereus) in otter caf\u0026eacute;s in Japan. Animals 14(16):2412. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ani14162412\u003c/span\u003e\u003cspan address=\"10.3390/ani14162412\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWenzel SG, Coe JB, Long T, Koerner S, Harvey M, Shepherd ML (2023) Qualitative analysis of small animal veterinarian\u0026ndash;perceived barriers to nutrition communication. J Am Vet Med Assoc 262(1):79\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2460/javma.23.05.0281\u003c/span\u003e\u003cspan address=\"10.2460/javma.23.05.0281\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Organization for Animal Health (2024) One Health. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.woah.org/en/what-we-do/global-initiatives/one-health/\u003c/span\u003e\u003cspan address=\"https://www.woah.org/en/what-we-do/global-initiatives/one-health/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"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":"One Health, Infodemic, leopard gecko, bearded dragon, veterinary referral, artificial intelligence","lastPublishedDoi":"10.21203/rs.3.rs-8735785/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8735785/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe global trade in pet reptiles is substantial, with Japan ranking as the second-highest importer worldwide. However, access to specialized veterinary care remains limited, encouraging owners to seek advice on unregulated online forums. We analyzed 692 Yahoo! Chiebukuro Q\u0026amp;A threads (2010\u0026ndash;2025) describing suspected infections in three pet lizards\u0026mdash;central bearded dragon (\u003cem\u003ePogona vitticeps\u003c/em\u003e; n\u0026thinsp;=\u0026thinsp;168), leopard gecko (\u003cem\u003eEublepharis macularius\u003c/em\u003e; n\u0026thinsp;=\u0026thinsp;431), and endemic Japanese grass lizard (\u003cem\u003eTakydromus tachydromoides\u003c/em\u003e; n\u0026thinsp;=\u0026thinsp;93)\u0026mdash;and compared community-selected human \u0026ldquo;Best Answers\u0026rdquo; with outputs from three large language models (ChatGPT-5.2, Claude Sonnet 4.5, Gemini 3.0 Pro). Advice was classified into five hierarchical categories, with the primary safety endpoint being immediate veterinary referral (Vet-direct) versus all non-referral categories. Human referral rates differed by species (23.66\u0026ndash;41.07%), with the endemic Japanese grass lizard receiving the fewest Vet-direct recommendations and the highest rate of home-treatment advice. In contrast, all LLMs produced substantially higher Vet-direct rates across species (61.25\u0026ndash;95.70%); paired thread-level McNemar tests showed LLMs were significantly more likely than humans to recommend Vet-direct for every species\u0026ndash;model comparison (all p\u0026thinsp;\u0026le;\u0026thinsp;9.58\u0026times;10⁻⁶). However, biosafety guidance related to zoonotic risk and hygiene was rare in both human and AI responses. These findings suggest that online communities may reinforce a \u0026ldquo;valuation hierarchy\u0026rdquo; that disadvantages low-economic-value species, while LLMs can function as a welfare-oriented safety net; nevertheless, proactive One Health messaging remains an important gap.\u003c/p\u003e","manuscriptTitle":"AI as Animal Welfare Safety Net: Large Language Models Favor Veterinary Referral Over Online Community Advice for Suspected Reptile Infections","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-02 07:18:01","doi":"10.21203/rs.3.rs-8735785/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":"23b96e49-ad04-458a-8e1d-2ee10402174d","owner":[],"postedDate":"February 2nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":61998995,"name":"Animal Science"}],"tags":[],"updatedAt":"2026-02-02T07:18:02+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-02 07:18:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8735785","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8735785","identity":"rs-8735785","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}