Ace
Short answer: using ChatGPT to do the work of learning for you can leave you remembering less. In a 2025 randomised trial, students who studied with ChatGPT scored 57.5% on a surprise test 45 days later, against 68.5% for students who studied without it. There is no evidence it damages your memory permanently. The problem is skipped effort, so the fix is to use AI to test you rather than to give you answers.
The 45-day trial behind the headlines
Most of this year's coverage traces back to one study. André Barcaui (2025) randomly assigned 120 undergraduates learning about artificial intelligence to two groups. One group could use ChatGPT as a study aid. The other used only traditional methods. Forty-five days later, both groups sat a surprise retention test.
The ChatGPT group scored 57.5% correct. The traditional group scored 68.5%. That gap is moderate to large in research terms (Cohen's d of 0.68). The author's explanation is that unrestricted ChatGPT use reduced the mental effort that builds durable memory. Studying felt easier and went faster, and less of it stuck.
It is one study, in one institution, on one topic. But it is a randomised trial with a delayed test, which is exactly the kind of evidence that matters for exams.
What other studies found
High school maths. Bastani and colleagues (2025) ran a field experiment with nearly a thousand high school students. With GPT-4 open during practice, students did much better on the practice problems: grades improved by 48% with a standard ChatGPT-style tool and 127% with a version designed to tutor rather than give answers. But when access was removed for the exam, students who had used the standard tool scored 17% lower than students who never had it. The tutor-style version largely avoided that harm.
Essay writing and brain activity. A team led by Nataliya Kosmyna at the MIT Media Lab (2025) recorded EEG while 54 people wrote essays with an AI assistant, a search engine or no tools. The AI group showed the weakest brain connectivity and had trouble quoting their own essays afterwards. This is a preprint, the sample is small, and essay writing is not revision, so treat it as a signal rather than proof.
Put together, the pattern is consistent: AI that hands you finished answers helps the task in front of you and hurts what you can do later on your own.
Why AI help can mean less memory
Memory researchers have known for decades that learning which feels harder often lasts longer. Retrieving an answer from your own head, struggling with a problem before seeing the method, and spacing practice out all feel slower than reading a clear explanation. They also produce much better long-term retention.
A chatbot removes exactly that struggle. Ask it to summarise a chapter and you skip the work of figuring out what matters. Ask it to solve a problem and you skip the attempt. Researchers call handing mental work to a tool "cognitive offloading". It's fine for things you don't need to know later, like a phone number you'll look up again. It is a problem for things you'll need to recall in an exam hall with no tools at all.
How to use ChatGPT without forgetting
The research points to one rule: keep the effortful part for yourself. In practice:
- Attempt first, ask second. Try the problem or write your own answer before asking AI anything. Then use it to check or explain what you missed.
- Ask it to quiz you, not to tell you. "Ask me ten questions on this chapter one at a time and wait for my answer" turns the chatbot into retrieval practice. Our guide to getting ChatGPT to quiz you has prompts.
- Use tutor-style modes. Bastani's tutor version, which gave hints instead of answers, avoided most of the harm. ChatGPT's study mode and similar features work on the same idea.
- Close the chat and test yourself later. A day or two after an AI-assisted session, check what you can recall without it. If the answer is "not much", you offloaded too much.
- Don't let summaries replace reading. A summary is a map, not the territory. You still need to work through the material.
For more on fitting AI into revision properly, see how to study with AI.
This is also how Ace is built: it doesn't explain or summarise your material for you. It turns your own slides, PDFs and notes into practice questions you have to answer, and only then shows you the sentence from your file that backs the answer. The effort stays with you.
What the studies don't show
- Permanent damage. Some headlines talk about "permanent memory loss". None of these studies tested anything like that. They measured how much people remembered about specific material.
- That all AI use is harmful. In Bastani's study, the version built to guide students rather than answer for them largely avoided the drop.
- That heavy users are doomed. The effect is about how you use the tool in a given session, which you can change today.
Frequently asked questions
Does ChatGPT make you dumber?
There is no evidence that it lowers intelligence. The evidence is narrower: if ChatGPT does the thinking while you study, you remember less of that material later.
Is it bad to use ChatGPT for revision?
Not if it makes you do the work. Using it to quiz you, check your answers or explain a mistake fits the research. Using it to summarise everything and reading the summary does not.
What is cognitive offloading?
Handing mental work to an outside tool, such as a calculator, a note or an AI chatbot. It saves effort, but you don't build the memory you would have built by doing it yourself.
How big was the memory difference in the ChatGPT study?
In Barcaui's 2025 trial, 57.5% correct for the ChatGPT group against 68.5% for traditional study, on a surprise test 45 days later.
Sources
- Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Social Sciences & Humanities Open, 12, 102287.
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122.
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint 2506.08872.
