Shannon Colville, Nadine Ostern
Interacting with generative artificial intelligence (GenAI) requires users to make ongoing trust judgements under uncertainty. This study examines how such judgements are constructed during use and the role metacognition plays in that process. Using ChatGPT as the case, we analysed 1,776 Reddit posts using grounded theory to examine how users’ reason about and regulate engagement with the system. The findings show that users draw on metaknowledge, expressed through anchored knowledge, interaction know-how, and delegation rules, and combine it with metacognitive regulation strategies, including guardrail use, active scrutiny, and reliance adjustment, to form trust judgements during interaction. From this analysis, we develop a conceptual model explaining how trust emerges through the interplay of metaknowledge and regulation, illustrating how users apply what they know and manage their engagement to navigate uncertainty. Building on this model, we introduce informed trust, a form of trust in which users rely on GenAI outputs only after evaluating them against what they know, recognising limits in their knowledge, and regulating their use in line with the task and stakes. This work advances theory by specifying the cognitive mechanisms that underpin trust in generative AI and offers practical levers for fostering more effective human–AI interactions.