ChatGPT, Perplexity, AI Overviews audited on mental health citations
What this answers
- Do ChatGPT, Perplexity and Google AI Overview cite the same sources for health questions?
- No — an audit of 1,140 responses to mental-health questions found the three platforms diverged sharply in citation consistency and in which source types they favored, even though overall source-category shares (government, commercial-health, academic) were similar.
- Does asking an AI chatbot to show its sources change what it cites?
- Only modestly. The audit found that explicitly requesting sources produced only modest changes in citation composition across the three platforms tested.
A cross-platform audit tested ChatGPT, Perplexity and Google AI Overview on 20 English mental-health questions plus queries in six additional languages, collecting 15,942 citations across 1,140 responses and 1,713 unique domains, per a preprint co-authored by Beth Israel Deaconess digital-psychiatry researcher John Torous and colleagues. The ten most-cited domains accounted for 43.6% of English-language citations, and government, commercial-health and academic sources each drew roughly 22% of citations overall — but the three platforms differed sharply in consistency and in which source types they favored. Non-English queries returned fewer citations and less language-appropriate routing, and explicitly asking a platform to show its sources produced only modest changes in what it cited.
Why it matters: It quantifies, for a sensitive topic, what a wire-record finding already suggested about health citations diverging by platform — now with a stated methodology and a 15,942-citation sample instead of a single vendor's dataset.
The record: ChatGPTPerplexityGoogle AI Overviews
Glossary: AI Overviews
Via arXiv AI-search citations ↗
Posted to the wire September 10, 2026. Edited by Joe Balewski.