LLMs as learning-preference amplifiers in first-year higher education: a conceptual affordance framework and design implications

Purpose This paper theorizes large language models (LLMs) as learning-preference amplifiers and introduces the AI Amplification Zone, specifying when LLM use strengthens productive learning strategies (e.g. strategic practice and reflection) versus entrenches maladaptive ones (e.g. avoidance and shallow processing). Design/methodology/approach A conceptual, theory-building synthesis integrating Affordance Theory with cognitive and motivational research. The analysis models a mechanism of preference detection → adaptation → ossification and derives testable propositions (P1–P3) and practice-oriented heuristics for first-year higher education. Findings The framework shows how desirable difficulties, clarified human–AI feedback roles, and UDL-aligned task structures can shift learners into the Amplification Zone, while guardrails (e.g. transparency, calibrated prompting and AI-free checkpoints) mitigate over-reliance. The paper advances three propositions: (P1) embedding desirable difficulties with LLM support promotes deeper learning strategies; (P2) scaffolded prompting plus human–AI co-teaching reduces over-reliance among learners with a strong performance orientation; and (P3) periodic AI-free checkpoints disrupt ossification loops and preserve transfer. Research limitations/implications As a conceptual synthesis, the framework requires empirical testing; the paper outlines falsifiable designs and evaluation measures (behavioral, performance and transfer) aligned to P1–P3. Practical implications Provides course-team heuristics: embed desirable difficulties, specify human/LLM feedback roles, schedule AI-free checkpoints and align tasks with UDL while monitoring over-reliance. Social implications Highlights equity and ethics considerations around profiling, transparency and bias; recommends guardrails to ensure AI augments rather than replaces productive struggle. Originality/value Reframes LLMs from generic assistants to preference amplifiers and offers a mechanism-level lens explaining why the same affordances can deepen learning for some students yet narrow strategy use for others.