How family medicine physicians and trainees in israel perceive generative artificial intelligence: a qualitative study of challenges and support needs for its use in work and training

Generative artificial intelligence (GenAI) tools are increasingly being used in clinical practice, yet little is known about how family medicine physicians and trainees explain how these tools work, the challenges they experience, and the support they consider necessary for responsible use. This study qualitatively examined participants’ written descriptions of GenAI across the family medicine training continuum. A qualitative descriptive study was conducted among 87 participants across the family medicine training continuum in Israel, including medical students in family medicine rotations ( n = 28), interns ( n = 13), residents ( n = 16), and specialists or senior physicians ( n = 30). Data were collected through three open-ended prompts embedded in an online Hebrew-language questionnaire. The prompts addressed how GenAI produces responses, challenges associated with its use, and support needed for effective and responsible use. Written responses were analyzed thematically using ATLAS.ti. The analysis remained close to explicitly stated content because the brief written responses did not permit probing or follow-up. Coding was non-exclusive, and all 87 participants provided a usable response to each prompt. Information-aggregation descriptions were identified in 58 participants’ responses (66.7%), and 44 participants (50.6%) likened GenAI to a search engine. Only 10 participants (11.5%) explicitly mentioned probabilistic text generation. Output trustworthiness was the most frequently coded concern: 39 participants (44.8%) mentioned accuracy or reliability, and 20 (23.0%) mentioned hallucinations or fabricated information. Regarding support needs, GenAI literacy and source verification were each mentioned by 18 participants (20.7%), followed by formal training ( n = 10, 11.5%) and access to medical-specific GenAI tools ( n = 9, 10.3%). Because coding was non-exclusive, the counts overlap and should be interpreted descriptively. These findings point to a possible need for targeted GenAI literacy education within family medicine training programs, for clinical governance frameworks, and for GenAI tools that meet the evidentiary standards of medical practice. Because the material consists of brief written answers, these are implications generated by an exploratory sample rather than established requirements.