Seven AI chatbots scored higher than dentists, doctors and members of the public on a 25-question radiation-safety quiz, according to a study reported by Bioengineer.org. The result is striking, but it measures answers to a quiz, not clinical judgement or patient outcomes.
Watch Desk analysis
What happened
The study compared seven chatbots with 344 human participants using the same multiple-choice questions. The chatbots scored between 80% and 90.67%; the human groups scored between 32% and 60%. Microsoft Copilot had the highest score among the AI systems, while Perplexity had the lowest.
The systems tested were Claude, DeepSeek, Microsoft Copilot, Gemini, Perplexity AI, ChatGPT-5 and ChatGPT-4o. The study, published in Scientific Reports on 26 October, also found that ChatGPT-5’s scores varied significantly across morning, midday and evening assessments. Most of the other systems were more consistent across those testing windows.
Why it matters
Radiation safety is relevant both to everyday questions and to medical settings, where misunderstandings can have consequences. The results suggest chatbots may be useful as educational aids, but the quiz cannot show whether they give safe advice in real conversations or improve decisions in practice.
The researchers explicitly cautioned against treating higher quiz accuracy as evidence of superior clinical judgement. That distinction is doing important work here: a good score is not a digital qualification to advise patients.
Our read
This is a worthwhile result about what chatbots can answer under controlled conditions, not a contest that settles whether they should be trusted with health decisions. The variation in ChatGPT-5’s results is a reminder that performance can depend on when a system is tested, too.
Use the findings as a reason to ask better questions about AI health information, not to swap professional advice for a chatbot with a strong multiple-choice record.
What to watch
- Whether follow-up research tests the systems on a broader range of radiation-safety questions.
- Whether the reported time-of-day variation for ChatGPT-5 appears in further testing.
- Whether researchers evaluate how people use chatbot answers in real health-information settings.
Discussion spark: Should a strong quiz score make a chatbot more useful for health education, or is real-world testing the only result that counts?
Sources and evidence
- AI chatbots outperformed dentists and doctors on radiation safety quiz – Bioengineer.org (27 October 2026, 00:18 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.