Can AI use improve short-term performance while lowering students' confidence in their own learning?

Why AI can both boost and erode confidence

AI chatbots can reduce anxiety in the short term, which may make students feel more confident during speaking activities [2]. However, this anxiety reduction does not necessarily translate into lasting confidence in their own abilities. In a longitudinal study of an AI-supported mathematics environment, sustained interaction with AI was associated with increased confidence over time [3]. This suggests that how long and how deeply students engage with AI matters: brief exposure may calm nerves, while extended use may build genuine self-assurance.

Yet confidence gains are not universal. For multilingual and international students learning English, generative AI can boost motivation and self-confidence, but it can also create dependency [4]. That dependency may lead to short-term performance gains while causing longer-term stagnation in language learning [4]. In other words, students might feel more confident because AI helps them perform, but that confidence may not reflect their own growing competence.

Reflection may be the difference between help and hindrance

One proposed mechanism is that reflective conversations with generative AI can support metacognition—students' awareness and regulation of their own learning [1]. This narrative review argues that future research should move beyond short-term performance outcomes and examine how AI affects confidence calibration, which is how well students' confidence matches their actual ability [1]. If AI simply provides answers, students may become overconfident without real understanding; if it prompts reflection, it might help them calibrate confidence more accurately.

The distinction between using AI as a mirror for thinking versus using it as an answer machine appears central. When AI encourages students to explain their reasoning or evaluate their own work, it may strengthen self-regulated learning [1]. When it short-circuits that process, it may undermine the very confidence it initially boosts.

What to watch for in AI-supported learning

Postgraduate students in Morocco reported that AI influenced their achievement, confidence, and critical thinking, though the study focused on perceptions rather than measured outcomes [5]. This highlights that students themselves notice shifts in confidence, but it remains unclear whether those shifts reflect genuine learning or just performance gains. The risk is that short-term performance improvements mask a decline in students' belief in their own ability to learn without AI.

For educators, the practical takeaway is to monitor not just whether AI helps students perform better today, but whether students feel more or less capable of learning on their own over time. The evidence suggests that AI's effect on confidence depends on how it is used: as a tool for reflection and metacognitive growth, it may support lasting confidence; as a shortcut, it may create a fragile confidence that fades when the AI is removed.

Sources used in this answer

Related research questions