Original title: Sources of Truth: A Multi-Platform, Multilingual Audit of Citations in AI Mental Health Information Queries
Phuong Anh Nguyen, Jill Noorily, Matthew Flathers, Haruka Notsu +6 · August 2026
People used to look up health questions by scanning lists of web links. Today, conversational artificial intelligence tools write a single answer and pick the sources to cite. This leaves users relying on the platform to choose good information. Researchers tested three free consumer tools: ChatGPT, Perplexity, and Google AI Overview. They asked them twenty mental health questions in English and translated three of those questions into six other languages. They studied what sources the tools cited. CitationsReferences to sources that support the information given. were heavily concentrated among a few places. The top ten websites made up 43.6% of all English citations. Academic, government, and commercial health sites each got roughly 22%. Non-English searches got fewer citations and sent users to language-appropriate resources far less often.
You are asking friends for restaurant recommendations. One friend always lists the same few popular chains, no matter what food you want. Another friend gives you shorter lists when you ask in a different language. Here, the researchers asked AI systems health questions and found that they also rely on a small set of sources, and that answers in other languages include fewer and less relevant references.
As people turn to AI for health advice, the sources these systems cite shape what users learn. This audit shows that citations are heavily concentrated in a few domains and that non-English queries get fewer relevant sources. That means information gaps are likely worse for non-native speakers.
The abstract does not directly state limitations. The audit covers only three free consumer products, twenty English mental health questions, and a subset translated into six languages. The findings may not apply to other platforms, topics, or language pairs. The classifier used is deterministic and validated against human coding, but the abstract does not report its accuracy.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Sources of Truth: A Multi-Platform, Multilingual Audit of Citations in AI Mental Health Information Queries — Phuong Anh Nguyen, Jill Noorily, Matthew Flathers et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.