Seven chatbots answered radiation-safety questions more accurately than human comparison groups in a study from Mersin University in Turkey. The researchers used the same 25-item multiple-choice questionnaire for the AI systems and 344 dentists, medical doctors and members of the public. Their result concerns knowledge of radiation risks and protection, rather than decisions about a patient’s imaging.[1]
The tested systems were Claude, DeepSeek, Microsoft Copilot, Gemini, Perplexity, ChatGPT-5 and ChatGPT-4o. Their reported accuracy ranged from 80% to 90.67%, compared with 32–60% among the human groups. Copilot scored highest and Perplexity lowest among the evaluated chatbots. Dentists and doctors performed better than the public, with no statistically significant difference between the two professional groups.[1]
What the questions can reveal
Radiation safety is a set of decisions about justification, dose, shielding, equipment, operator protection and patient communication. In dental and medical practice those decisions matter because X-rays can be clinically useful while unnecessary exposure should be avoided. A multiple-choice quiz can check declarative knowledge of these principles; it cannot recreate the competing factors in deciding whether a particular patient requires an image.[1]
The format isolates a particular ability: selecting an answer to a written knowledge question. Clinical judgement also depends on the patient, the reason for imaging and the available alternatives. The study did not test whether a chatbot could combine those factors safely or improve care.[1]
Reported accuracy on the same questionnaire
| Respondent group | Accuracy range |
|---|---|
| Seven chatbot systems | 80–90.67% |
| Human comparison groups | 32–60% |
The investigators repeated chatbot assessments in the morning, at midday and in the evening. Most systems showed stable performance across those windows, while ChatGPT-5 showed a statistically significant temporal difference (p < 0.001). That observation does not identify why its answers changed. It does show why the tested product and evaluation conditions belong alongside a benchmark score.[1]
An education finding
The human results give educators a reason to examine which radiation-safety concepts need reinforcement. The chatbot results suggest a possible way to help learners practise, provided answers are checked against reliable professional material. A useful next evaluation would ask whether that practice improves clinicians’ subsequent unaided knowledge and their handling of realistic cases. This study establishes the knowledge gap; it does not yet show how best to close it.[1]
Sources & further reading
- AI chatbots showed higher response accuracy than dentists, medical doctors, and the general public on radiation safety questions Scientific Reports · 26 Oct 2026 · Accessed 26 Oct 2026