Quantitative analysis:

To date, 109 questionnaires from Switzerland were analysed. The sample comprised 56% female participants. The age distribution was relatively homogeneous, with approximately 30% of participants in each of the following age groups: 18-39 years, 40-59 years, and 60-79 years. In terms of education, 44% of participants had attended university. Fifty five percent of the participants reported having undergone more than 10 radiological exams in their lifetime. About familiarity with chatbots, 55% of participants either did not know what a chatbot is or never used one. In terms of radioprotection awareness, 70% of the participants had never asked questions about radioprotection, while 20% had previously searched for radioprotection information online.

The median scores for the evaluated outcomes were as follows: public comprehension (human: 6 vs. ChatGPT: 7, p = 0.039), trust (human: 4 vs. ChatGPT: 6, p < 0.001), reassurance (human: 4 vs. ChatGPT: 6, p = 0.013), and overall satisfaction (human: 4 vs. ChatGPT: 6, p < 0.001). As opposed to expert, patients scores did not differ between human and ChatGPT-generated texts regarding perception of whether AI had generated the response (p ≥ 0.99). Patients correctly identified the source with high confidence in 14.2% (31 of 218) of human responses and 17.4% (38 of 218) of ChatGPT responses, respectively.

Qualitative Analysis:

Fifty patients wrote a comment about the human response, and 37 wrote a comment about ChatGPT.

• Human-generated response comments included only two clearly enthusiastic remarks, while the rest expressed suggestions for improvement, such as humanizing the text and adding empathy and reassurance for anxious patients. For instance, Participant 21 noted, "Adding humanity to the text, which is too cold and distant for a worried and stressed patient," and Participant 61 mentioned, "Response very factual and could be more reassuring." Participant 88 said, "Vague and unempathetic response. Not very reassuring. Seems degrading for the worried mother." Participants also suggested including more detailed, context-specific information, particularly from medical studies or expert sources. Participant 37 expressed, "Deserved to be more in-depth," while Participant 106 mentioned, "A more detailed development, more details, perhaps adapted for elderly people or non-native speakers."
• ChatGPT-generated response comments were more positive, with ten clearly enthusiastic remarks. Participant 105 described the response as "Very detailed and exhaustive," and Participant 32 felt that "No need to improve; it seems clear to me." However, most suggestions focused on structural improvements, such as simplifying the language and making responses clearer and more concise. For example, Participant 55 commented, "A bit more simplicity and shorter." A few also noted that the responses provided too many risks without enough balance. One participant remarked, "Avoid bringing so many negative details without balancing them afterward; the response is worrying." Some patients also seemed to have additional questions. Participant 61 said, "We don’t know exactly whether to consult a specialist or follow the recommendations," while Participant 69 asked, "Toward which specialists should we turn? Scientific studies, references?"