Navigating the unknown: a qualitative descriptive study of ophthalmic nurses’ trust negotiation with generative artificial intelligence in China

Background Generative artificial intelligence (GenAI) tools such as ChatGPT and DeepSeek are increasingly used in ophthalmic clinical practice. Ophthalmic nursing has three trust-relevant features: the irreversibility of visual outcomes, the precision-sensitivity of ophthalmic medication, and the relational dependence of visually impaired patients on nurse-mediated care. These features create distinctive challenges for nurses establishing, calibrating and maintaining trust in GenAI. The existing literature is dominated by static attitude surveys and rarely characterises trust as a dynamic, context-embedded process. Methods In this qualitative descriptive study, we conducted semi-structured in-depth interviews with 26 ophthalmic nurses recruited by maximum-variation sampling from tertiary, secondary and primary eye-care settings in eastern, central and western China. Interviews were guided by the Critical Incident Technique, and transcripts were analysed using Braun and Clarke’s reflexive thematic analysis. Reporting follows COREQ. Results We developed three themes: (1) tentative approach — competence appraisal from curiosity to conditional trust; (2) guarding professional boundaries — trust anchoring amid ethical tension and relational reconfiguration; and (3) trust negotiation within the organisational ecology. Across themes, nurses calibrated their trust in a graded, task-specific manner — a pattern we tentatively term ‘gradient calibration’ — while ethical obligation, the nurse–patient relationship and organisational context shaped how trust was negotiated. Conclusions Ophthalmic nurses’ trust in GenAI appeared neither as static acceptance nor refusal, but as a dynamic, context-embedded negotiation shaped by the irreversibility of visual outcomes, medication-precision demands and visually impaired patients’ relational dependence on nurses. Findings may inform specialty-specific AI training, institutional guidance and tool design; transferability beyond this sample requires further study.