A Seven-Step Framework for Structured ChatGPT-Assisted Pharmacology Learning: Educational Outcomes and Attitudes...

Generative artificial intelligence is increasingly used by health professions students and by patients seeking health information before consultation. This creates two related educational challenges: supporting critical AI use during learning and preparing future clinicians to respond appropriately to hypothetical AI-informed patients. This study evaluated a teacher-designed seven-step ChatGPT-assisted framework for autonomic pharmacology learning and explored whether principles embedded in the framework were reflected in students' attitudes toward AI-informed patients.

In this matched-cohort educational study, 104 third-year dental students were included. Students who voluntarily participated in the seven-step ChatGPT-assisted framework (n = 52) were individually matched with nonusers (n = 52) by previous academic achievement and digital-tool familiarity. The primary outcome was performance on a three-item open-ended pharmacology quiz assessing mechanistic reasoning, scored independently by two blinded assessors. Framework participants also completed an eight-item learning-perception questionnaire and a ten-item questionnaire addressing hypothetical AI-informed patients. Exploratory Spearman correlations examined conceptually matched learning- and patient-facing statements.

Framework participants achieved higher postlearning pharmacology scores than matched nonusers (1.77 ± 0.30 vs. 1.45 ± 0.47; p = 0.0003). The largest difference was observed for questions requiring integration of receptor mechanisms and downstream signaling (+28%; p < 0.001), whereas factual drug-name recall did not differ (p = 0.47). Students generally recognized both the potential usefulness and limitations of ChatGPT and endorsed professional, nondismissive responses to hypothetical AI-informed patients. The perceived usefulness of ChatGPT for learning and writing was associated with recognition of the risk of patient overreliance on AI-generated diagnoses (ρ = 0.32, p = 0.019), while intention to use ChatGPT for learning was associated with endorsement of a serious and professional approach to patients presenting with ChatGPT-generated diagnoses (ρ = 0.40, p = 0.004).

Participation in the structured framework was associated with stronger postlearning mechanistic pharmacology performance. The novel cross-context analysis identified limited but potentially meaningful alignment between responsible AI use in learning and professional attitudes toward hypothetical AI-informed patients. These exploratory associations do not demonstrate transfer of clinical communication competence, but suggest that verification, contextualization, risk awareness, and human oversight may provide a shared educational foundation for addressing AI use by both learners and patients.