Generative artificial intelligence in patient education: opportunities and caution for clinical implementation

Letter to: Generative artificial intelligence for patient education material on gastric cancer prevention Endoscopy 2026; 58(06): 669-677 DOI: 10.1055/a-2780-0664 10.1055/a-2780-0664 We read with great interest the study by Rizkala et al. evaluating the use of generative artificial intelligence (AI) to produce patient education materials on gastric cancer prevention [ 1 ]. The authors should be commended for exploring the potential of large language models to generate patient-friendly summaries of complex clinical guidelines. As guideline-based patient education becomes increasingly important for promoting participation in screening and surveillance programs, integrating AI into this process represents a promising innovation. However, an important practical consideration in real-world use is that outputs generated by large language models may be influenced by the user’s interaction context. Responses can be shaped by previous prompts, earlier conversations, or even inaccurate information introduced during prior exchanges [ 2 ]. Consequently, different clinicians or patients using the same prompt may obtain different summaries, and existing misinformation may inadvertently affect the generated interpretation. Providing practical guidance on how users can minimize such variability, such as standardized prompts or concise operational frameworks, may therefore improve the consistency and reliability of AI-assisted patient education. The responsible deployment of AI-generated medical information also requires careful monitoring of safety and accuracy [ 3 ]. In the present study, the absence of hallucinations was inferred primarily from expert ratings of accuracy; however, subjective scoring alone may not reliably detect subtle factual inconsistencies. Structured hallucination audits, predefined error taxonomies, and independent verification processes may therefore be necessary before patient-facing dissemination. Clearer governance frameworks and accountability mechanisms for AI-generated educational content also remain important considerations [ 4 ]. Finally, the study compared an AI-generated summary with material from a patient advocacy organization, without benchmarking either against an independently validated reference standard. Similar ratings between the two summaries may therefore reflect shared limitations rather than true equivalence in quality. Incorporating objective content validation methods may help determine whether AI-generated summaries truly meet the standards required for patient-facing medical communication. Publication History Article published online: 22 July 2026 © 2026. Thieme. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany