AI as a Cognitive Crutch: What Studies Say About Students and Heavy AI Use

Many students now use AI chatbots for notes, essays and problem sets. A fair question is whether that helps them learn or lets them skip the learning. The research is early, and the headlines have run ahead of it. This page looks at what four studies actually found, and what each one cannot show.

“AI addiction” is not a formal diagnosis. The American Psychiatric Association’s DSM-5-TR lists gambling disorder as its one behavioral addiction (APA), and the World Health Organization’s ICD-11 covers gaming disorder (WHO). Neither covers AI use. When a study uses the word “addiction” for AI, it is the authors’ own measure.

The Zimbabwe case study

A paper in Cogent Education, published in August 2025, studied 248 undergraduates at one university in Zimbabwe using surveys, interviews, usage tracking and academic records (Revesai 2025). We could read the abstract but not the full text. As reported in the abstract:

• 32.7% of the students showed “addictive patterns” of generative AI use, by the author’s measure.
• Those students averaged 18.3 AI interactions a day, and 65.8% reported failed attempts to cut back. They kept using AI even when they saw it harming their work.
• Heavily affected students had a mean GPA 0.41 points lower than students without those patterns.
• A regression pointed to three routes: weaker critical thinking, weaker writing development and less learning of course content.
• Interviews described limited internet, high costs and scarce course materials that made AI more appealing.

Read this carefully. It is one university. It shows an association, not that AI use caused lower grades, since students who are struggling may lean on AI more. It does not show that a third of students everywhere are dependent.

Practice with a chatbot versus practice without

A field experiment published in PNAS gave nearly a thousand high school math students access to GPT-4 tutors (Bastani et al. 2025). A tutor that behaved like standard ChatGPT improved practice grades by 48%. A version built to give hints and protect learning improved them by 127%. When access was taken away, the standard-ChatGPT group scored 17% lower than students who never had it, while the hint-based tutor largely avoided that drop. The authors say students used the standard version as a “crutch” during practice. The PNAS site later published a correction to the paper, so check the current version.

Self-reported effects on thinking at work

Microsoft Research surveyed 319 knowledge workers about 936 examples of using generative AI (Microsoft Research). People who had more confidence in the AI reported less critical thinking. People with more confidence in their own ability reported more. This is self-report from workers, not students, but it points the same way.

A small essay-writing study

A preprint from MIT Media Lab followed 54 participants writing essays with an LLM, a search engine or no tools (arXiv). The LLM group reported the least ownership of their essays and had more trouble quoting them. Only 18 people finished the final session, and the paper was not peer reviewed when we read it. It is a hint, not a conclusion.

What the studies suggest

None of these shows that AI harms learning in general. Taken together, they suggest the risk is in how it is used. Getting answers while practicing seems to cost more than getting hints, checking work or being quizzed. Local conditions matter too. Where textbooks are scarce or expensive, AI may be filling a real gap, and telling students to simply use it less would not help.

Practical steps for students

• Try the problem first. Open the chatbot after you are stuck, not before.
• Ask for hints, a quiz or feedback on your own draft rather than a finished answer.
• Close it and explain the idea out loud or on paper. If you cannot, you have not learned it yet.
• Read your course’s AI policy and ask your instructor what is allowed.
• Use office hours, tutoring centers and study groups too.

If AI use is tied to panic, loss of sleep or low mood, talk to your campus counseling service or a doctor. If you are in distress, call or text 988 in the US (988 Lifeline), or text HOME to 741741 (Crisis Text Line). In an emergency, call 911. SAMHSA’s helpline is 1-800-662-4357.

Is AI making students worse at thinking?

Nobody knows yet. Some studies link unguided use to weaker results when the AI is taken away, and one found a hint-based tutor avoided that. The studies are small or self-reported, and none follows students for years.

Does 32.7% mean a third of students are addicted to AI?

No. That figure comes from 248 students at one university in Zimbabwe, using the author’s own measure of “addictive patterns.” It cannot be applied to students in other places, and AI addiction is not a formal diagnosis.

How can I tell if I am leaning on AI too much for school?

Ask whether you can do the task without it, whether you open it before trying, and whether you feel anxious when you cannot use it. If yes to any, try the steps above for a week and see what changes.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.