The AI Behavior Index is the research arm of OneChat AI. This page is unusual in that it requires triangulating across two very different bodies of evidence: self-reported survey data (HEPI for students, Elsevier / Oxford UP / Nature for researchers) and forensic-detection studies of actual published work (Desaire et al. for the MDPI corpus). The two methodologies produce strikingly different numbers, and both are valid measures of different things — stated use vs detected text generation.
Three caveats worth flagging. First, "use" is defined differently across surveys: some count any AI interaction during research, others count direct text generation in papers. Second, social-desirability bias likely pushes researcher self-reports in opposite directions depending on framing — some understate to avoid stigma, others overstate to appear current. Third, the Desaire et al. detection rate (1–3%) is now over two years old; given the steep acceleration in student use over the same period, undisclosed academic use has likely risen substantially since. Data is refreshed when major surveys publish. If you have a study to suggest or notice an error, contact us at research@aibehaviorindex.org.