Generative AI use before medical visits: disclosure-item responses, trust, and care-seeking behaviors in a cross-sectional social-media survey in Poland.

BACKGROUND: Patient-facing generative artificial intelligence (AI) tools such as ChatGPT are increasingly used for health information seeking, yet their role in patient-physician communication remains insufficiently characterized. Using the Network Episode Model as the primary framework, this study examined self-reported pre-consultation AI use, whether respondents reported disclosing that use to a physician, and self-reported AI-associated health-behavior change in a convenience sample of Polish adults recruited via social media. Findings are exploratory. METHODS: Cross-sectional online survey, anonymous with respect to the investigators (N = 1,044), administered in Polish via Google Forms with programmed routing; recruitment used a multi-account social-media strategy. A 24-item de novo questionnaire assessed health-related and pre-consultation AI use, disclosure-item responses, trust, self-reported behavioral consequences, and sociodemographics. Q13 responses were analyzed among respondents reporting AI use before a medical visit; the survey did not confirm that a consultation subsequently occurred. Exploratory analyses included chi-square, Mann-Whitney U, Kruskal-Wallis, Spearman correlations, and logistic regression restricted to definite Yes/No responses; the originally submitted coding and Benjamini-Hochberg adjustment were retained as sensitivity analyses. RESULTS: Health-related AI use was reported by 84.3% (n = 880; 95% CI 82.0-86.4). Among respondents reporting AI use before a medical visit (n = 519), 82.5% (n = 428; 95% CI 79.0-85.5) selected 'No' when asked whether they had informed a physician about that use. Because attendance was not confirmed, this proportion cannot be interpreted as a rate of what patients withheld during completed clinical encounters. Selection of 'No' was similar across visit-avoidance strata (79.2-86.8%). Among those selecting 'No' who reported no AI-related visit avoidance (n = 235), the most frequent reasons were fear of a negative physician reaction (86.4%) and reluctance to undermine physician authority (74.9%). In primary definite-response models, male gender and greater AI trust were associated with the outcomes; AI trust was strongly negatively associated with age. Apparent discrimination and explained variation were modest. CONCLUSIONS: Most respondents reporting pre-consultation AI use selected 'No' on the disclosure item. Because consultation attendance and a defined index episode were not confirmed, the actual disclosure rate within completed clinical encounters cannot be estimated from this survey. The findings advance a hypothesis about a possible communication barrier requiring prospective, episode-based study.