- Who it helps
- Researchers and teams evaluating the consistency of health-information assistants.
- Availability
- October 30 paper is public. The authors state that responses and analysis code are released; those artifacts were not checked.
The comparison
Six models answered four smoking-cessation questions for twelve personas, producing 2,880 responses. The authors found differences in factual coverage across the tested scenarios.
A chatbot audit finds persona-dependent answers to the same questions ↗
A proposed safeguard has limits
A prompt specifying core relevant facts reduced persona differences, but its overall effect varied across models. The authors caution that identical medication mention rates are not always clinically appropriate and that personas differ on several attributes at once.
A chatbot audit finds persona-dependent answers to the same questions ↗
What you can do
- Read the age-aware evaluation caveats and test any prompt intervention separately on each model. Open reference ↗
Where the evidence stops
Abstract checked November 4. The study measures model responses and does not measure clinical outcomes. Data and analysis were not independently reproduced.
Sources read. Primary source read November 4, 2026 UTC. No software installation, model inference or complete workflow executed.
Sources & further reading
Paper · paper
A chatbot audit finds persona-dependent answers to the same questions ↗
The audit distinguishes scenario-dependent fact coverage from clinical appropriateness. Its prompt intervention varies by model; no medical outcome or independent reproduction is established.
Artin Yousefi and coauthors · page checked 2026-11-04
Coverage note: Independent attention and end-to-end execution remain unverified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
A research audit examines how chatbot answers change when the same question comes from different fictional user personas.