Does using AI make students worse at thinking for themselves?
It can, when AI does the work for them. A 2025 PNAS study found students who practised with an answer-giving chatbot scored 17% lower once it was removed, while a hint-giving AI tutor largely avoided that drop. The risk comes from offloading the thinking, not from AI itself, so how students use it matters most.
AI can weaken students' thinking when it does the thinking for them, and the early research is fairly consistent on that point. When students use AI to get finished answers, they practise less and learn less. When AI is set up to question, hint and explain, the harm largely disappears. The design of the tool and the task decides which one you get.
What the research shows so far
The PNAS math study (2025)
The strongest evidence comes from a randomized field experiment with nearly 1,000 high school students in Turkey, published in PNAS:
- GPT Base, a standard chatbot, improved practice scores by 48%, but on the later exam without AI, those students scored 17% lower than students who never had it.
- GPT Tutor, a version prompted to give hints and teacher-provided guidance instead of answers, improved practice scores by 127%, and its users performed about the same as the control group on the exam.
The authors concluded that students used the unguarded chatbot as a "crutch". The study measured short-term outcomes, so long-term effects are still unknown.
Brain activity during essay writing (2025, not yet peer reviewed)
An MIT Media Lab preprint, "Your Brain on ChatGPT", followed 54 university-age participants writing essays with ChatGPT, with a search engine, or unaided. The ChatGPT group showed lower brain engagement on EEG and weaker recall of what they had written. It is small, adult and not yet peer reviewed, so treat it as a signal rather than proof.
Neither study looked at primary school children, but the mechanism (skipping the effort that builds skill) applies at any age, and arguably more so while basic skills are forming.
What this means for K-8 classrooms
Protect the practice. Reading fluency, handwriting, spelling, number facts and writing stamina are built by repetition. Keep AI out of that practice.
Use AI after the attempt, not instead of it. "Try it, then ask the AI to check your reasoning" keeps the student doing the work.
Prefer tools that ask questions back. The PNAS result suggests hint-based, explain-don't-answer design is what makes the difference.
Make students the critics. Have them find errors in an AI answer, compare it with a book or decide whether it is fair. Judging AI output is itself a thinking skill.
Talk about it openly. Older students can understand the study: "The students who let AI do the work did worse on the test."
Good uses that keep students thinking
- Asking follow-up "why" questions after a lesson
- Getting a concept explained in a different way
- Brainstorming, then choosing and justifying the best idea
- Role-play conversations (interviewing a historical figure) followed by fact-checking
Where Askie fits
Askie is built to explain rather than simply hand over answers, adapts responses to each child's age, and in Askie for Schools teachers can review students' conversation history to see how they are using it. For practical strategies, see Teaching kids critical thinking with AI and AI homework help: does it work?.
Bring safe AI into your classroom
Askie for Schools brings a safe, age-appropriate AI learning companion to your students, managed from your teacher account.