Evidence, trust, and objectivity with generative AI: a qualitative interview study of pre-service science teachers’ truth-assessment practices

Generative artificial intelligence (GenAI) systems can present information in a persuasive, scientific register while still producing subtle errors or making unsupported claims. For pre-service science teachers (PSTs), this creates a practical dilemma: determining what is sufficiently reliable to use, particularly when the intended audience is future students. This qualitative interview study examined how 20 PSTs enrolled in a science teacher education program at a public undergraduate university in Sichuan Province, China, evaluated the truthfulness of GenAI-style explanations. Participants completed a vignette-based think-aloud task and a semi-structured interview exploring evidence standards, trust calibration, conceptions of objectivity, and verification stop rules. Using Framework Analysis, we identified five themes: (1) situational evidence standards and locally embedded authority infrastructures; (2) trust calibration shaped by familiarity, perceived risk, and time pressure; (3) dual conceptions of objectivity-as rhetorical neutrality and as a justificatory process; (4) hierarchical verification strategies with explicit thresholds for "enough checking"; and (5) prudent instructional decisions under epistemic uncertainty, including revising, qualifying, or rejecting GenAI outputs. These findings position truth assessment as a situated professional practice rather than a decontextualized skill. They also suggest priorities for teacher education: making verification routines teachable, foregrounding objectivity-as-process, and connecting AI literacy to pedagogical responsibility.