[article]
| Titre : |
A comparative and temporal evaluation of autism information across ten AI platforms |
| Type de document : |
texte imprimé |
| Auteurs : |
Serife BALIKCI, Auteur |
| Article en page(s) : |
202965 |
| Langues : |
Anglais (eng) |
| Mots-clés : |
Autism spectrum disorder Artificial intelligence Brave ChatGPT Claude Copilot DeepSeek Gemini Grok Meta AI Perplexity Poe |
| Index. décimale : |
PER Périodiques |
| Résumé : |
Caregivers are increasingly using artificial intelligence (AI) platforms to obtain information about autism spectrum disorder (ASD), yet the consistency and usability of these responses remain unclear. This study examined the quality and temporal stability of responses generated by ten widely used, freely accessible AI platforms (Brave, ChatGPT, Claude, DeepSeek, Gemini, Grok, Meta AI, Microsoft Copilot, Perplexity, and Poe) when asked 15 autism-related questions. A descriptive research design was used. Responses were collected at two time points (August 2025 and February 2026) and evaluated across six dimensions including accuracy, readability, language framing, actionability, reference presence and format, and safety indicators. All responses reflected single-turn, first-pass outputs without follow-up prompting. Findings revealed considerable differences in performance across platforms, alongside consistent patterns over time. AI platforms generally provided accurate responses to autism-related questions; however, the readability of responses exceeded recommended guidelines. Most responses were presented using medicalized language, with limited use of neurodiversity-affirming wording. Actionable guidance was also limited, as only a small proportion of responses included clear next steps for families. Reference practices showed considerable variation, with some platforms including multiple sources (e.g., Gemini, Brave, Perplexity) while others offered few or none (e.g., ChatGPT, Meta AI). Across all platforms and both time points, no explicit misinformation, inappropriate reassurance, discouragement of evaluation or intervention, or unsafe recommendations were identified. Comparisons across time points showed that platform-level performance patterns were largely stable, with only minor fluctuations in scores across evaluation dimensions. These findings suggest that differences across AI platforms reflect consistent platform-level behaviors under default conditions rather than short-term variation in outputs. Overall, while AI platforms can provide broadly accurate information about ASD, they differ considerably in clarity, tone, usability, and transparency. The results highlight the need for cautious interpretation of AI-generated autism information and suggest that families may benefit from guidance when using these platforms. |
| En ligne : |
https://doi.org/10.1016/j.reia.2026.202965 |
| Permalink : |
https://www.cra-rhone-alpes.org/cid/opac_css/index.php?lvl=notice_display&id=592 |
in Research in Autism > 136 (August 2026) . - 202965
[article] A comparative and temporal evaluation of autism information across ten AI platforms [texte imprimé] / Serife BALIKCI, Auteur . - 202965. Langues : Anglais ( eng) in Research in Autism > 136 (August 2026) . - 202965
| Mots-clés : |
Autism spectrum disorder Artificial intelligence Brave ChatGPT Claude Copilot DeepSeek Gemini Grok Meta AI Perplexity Poe |
| Index. décimale : |
PER Périodiques |
| Résumé : |
Caregivers are increasingly using artificial intelligence (AI) platforms to obtain information about autism spectrum disorder (ASD), yet the consistency and usability of these responses remain unclear. This study examined the quality and temporal stability of responses generated by ten widely used, freely accessible AI platforms (Brave, ChatGPT, Claude, DeepSeek, Gemini, Grok, Meta AI, Microsoft Copilot, Perplexity, and Poe) when asked 15 autism-related questions. A descriptive research design was used. Responses were collected at two time points (August 2025 and February 2026) and evaluated across six dimensions including accuracy, readability, language framing, actionability, reference presence and format, and safety indicators. All responses reflected single-turn, first-pass outputs without follow-up prompting. Findings revealed considerable differences in performance across platforms, alongside consistent patterns over time. AI platforms generally provided accurate responses to autism-related questions; however, the readability of responses exceeded recommended guidelines. Most responses were presented using medicalized language, with limited use of neurodiversity-affirming wording. Actionable guidance was also limited, as only a small proportion of responses included clear next steps for families. Reference practices showed considerable variation, with some platforms including multiple sources (e.g., Gemini, Brave, Perplexity) while others offered few or none (e.g., ChatGPT, Meta AI). Across all platforms and both time points, no explicit misinformation, inappropriate reassurance, discouragement of evaluation or intervention, or unsafe recommendations were identified. Comparisons across time points showed that platform-level performance patterns were largely stable, with only minor fluctuations in scores across evaluation dimensions. These findings suggest that differences across AI platforms reflect consistent platform-level behaviors under default conditions rather than short-term variation in outputs. Overall, while AI platforms can provide broadly accurate information about ASD, they differ considerably in clarity, tone, usability, and transparency. The results highlight the need for cautious interpretation of AI-generated autism information and suggest that families may benefit from guidance when using these platforms. |
| En ligne : |
https://doi.org/10.1016/j.reia.2026.202965 |
| Permalink : |
https://www.cra-rhone-alpes.org/cid/opac_css/index.php?lvl=notice_display&id=592 |
|  |
A comparative and temporal evaluation of autism information across ten AI platforms in Research in Autism, 136 (August 2026)
