Reframing Memories of Academic Uncertainty With a Constructive-Developmental AI Reduces Expressed Uncertainty in STEM Learners

Abstract Conversational AI can alter what people later remember, yet its effects on how students remember their own academic experiences are untested. We examined whether a large language model (LLM)’s reframing of memories of academic uncertainty changes how STEM undergraduates later describe those memories. Participants recalled five such memories and received an LLM-generated response to each: a reframe instructed to move them toward a self-authoring perspective according to Constructive-Developmental theory, or a restatement that elaborated the uncertainty. One week later, Self-Authorship participants used less tentative language when redescribing their memories, whereas restatement produced no reliable change. LLM-generated vocabulary reappeared in later descriptions of the specific memories it addressed in both conditions. The reduction in tentative language remained after novel reframe vocabulary was removed from the Session 2 descriptions. Emotional valence did not differ by condition. A brief interaction with a theory-constrained AI agent can change how learners later describe academic uncertainty.