There is no good evidence in the studies reviewed here that ChatGPT causes general or lasting cognitive decline. Some research does raise a narrower concern: when people hand substantial thinking over to an AI, they may engage less deeply with a task or retain less of what they produce. Other studies find better work with AI, and one recent classroom experiment suggests that AI support and explicit reasoning instruction can work together. What matters is the task, how the tool is used and what outcome is measured.
What the MIT “Your Brain on ChatGPT” study found—and what it did not
The headline-grabbing study, “Your Brain on ChatGPT,” by Nataliya Kosmyna and coauthors at the MIT Media Lab, compared people writing essays with an LLM, a search engine or no external tool. The researchers used EEG to measure brain activity during writing and also analyzed and evaluated the essays. In the first three sessions, there were 54 participants; 18 completed a fourth crossover session.
The authors reported the strongest and most widely distributed EEG connectivity in the brain-only group, moderate engagement among search-engine users and the weakest connectivity in the LLM group. They also reported lower essay ownership among LLM users and more difficulty quoting their own work.
That is a finding about measured activity and performance during a particular writing task—not proof of brain damage, permanent loss of ability or reduced intelligence. The work is a small preprint, not a population-wide assessment, and its observed group differences do not by themselves establish lasting causation. The fourth session also had a smaller participant group than the first three.
Recommended Free Tools
Why the other studies do not give one simple answer
Studies ask different questions: whether AI improves the product someone submits, whether the person learns or transfers knowledge, how much effort they report, or how they manage AI-assisted work. Those outcomes can diverge. A polished answer is not necessarily evidence that its author learned more, just as lower reported effort on one task is not proof of a general loss of critical thinking.
| Study and participants | Task and design | Reported result | What it can tell us |
|---|---|---|---|
| Kosmyna et al., MIT Media Lab preprint (2025): 54 participants across the first three sessions; 18 in the fourth crossover session | Essay writing with an LLM, a search engine or no external tool; EEG and essay measures | The LLM group had the weakest reported connectivity and lower essay ownership; the brain-only group had the strongest and most distributed connectivity. | A preliminary, task-specific result; it does not establish lasting harm or general cognitive decline. |
| Fan et al., British Journal of Educational Technology (2025): 117 university students | Randomized writing-task comparison of ChatGPT, a human expert, writing analytics and no extra tool | The ChatGPT group improved essay scores, while knowledge gain and transfer did not differ significantly. | Output quality and learning are distinct outcomes; this short-term task cannot establish permanent effects. |
| Lee et al., Microsoft Research / CHI 2025: 319 knowledge workers who used generative AI at least weekly; 936 reported use examples | Survey and analysis of reported workplace use | Respondents described critical-thinking work shifting toward checking information, integrating AI responses and overseeing task execution. | Self-reported observational evidence describes how work was experienced, but cannot show that AI caused skill loss. |
| OpenAI Economic Research article (August 27, 2026): more than 1,000 first-year Bocconi University students | Randomized business-case experiment, including GPT-4o access and a causal-reasoning exercise | The researchers reported better rubric-rated work and coherence with GPT-4o; the reasoning exercise was associated with a wider range of ideas. Students receiving both showed benefits across multiple measures. | This specific classroom task suggests tool support and reasoning instruction can complement one another; it does not answer the question of long-term effects. |
| “Cognitive ease at a cost,” Computers in Human Behavior (2024): 91 university students | Research on a socio-scientific issue using ChatGPT 3.5 or Google | The study reported lower mental effort with the LLM and raised concerns about inquiry depth. | A focused result from one task and sample, not a general measure of cognition. |
What “dulling your brain” can mean in practice
The phrase can bundle together several different concerns that should not be treated as interchangeable:
- Less effort during a task: a person may report that using a chatbot takes less mental work than another approach. That alone does not show a lasting change in ability.
- Less ownership or recall: if an AI supplies much of the wording or reasoning, a user may find it harder to explain or remember the result. The MIT essay study reported lower ownership and difficulty quoting work in its LLM group.
- Weaker learning or transfer: someone can complete a task successfully without gaining more knowledge or becoming better at a related task. The 117-student writing study found improved essay scores but no significant difference in knowledge gain or transfer.
- Changes in how thinking is spent: workers may spend more attention checking, integrating and supervising AI output instead of generating every part themselves. The Microsoft survey describes reported shifts in work, not a measured decline in workers’ abilities.
These distinctions help explain why one study can raise a concern while another reports a benefit without necessarily contradicting it.
How to use ChatGPT without outsourcing the learning
The studies do not prove that any particular personal technique prevents cognitive effects. But if your goal is to learn or strengthen your own reasoning—not just finish a task—keep the parts you want to develop active:
- Try the problem yourself first. Draft an outline, identify the question or make an initial recommendation before asking ChatGPT for an answer.
- Use it to challenge your thinking. Ask for counterarguments, missing considerations or questions to investigate rather than a finished response to submit unchanged.
- Check claims against evidence. Verify important factual claims and sources yourself; treating fluent output as automatically reliable would replace evaluation with trust.
- Explain the result without looking. Summarize the reasoning in your own words and note what evidence supports it. If you cannot explain a key step, return to that step before relying on the answer.
- Match the tool to your goal. For a quick draft or routine task, assistance may be the priority. For practice, assessment or durable learning, do enough unaided work to test what you can do independently.
This is a practical distinction between using AI as support and letting it perform the thinking you intended to practice. The Bocconi experiment is consistent with the value of combining AI access with reasoning instruction in its particular classroom task, but it does not validate every technique or guarantee learning for every user.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What researchers still need to establish
The available studies do not settle whether frequent ChatGPT use changes cognition over months or years, or whether any effect differs by age, subject, task or pattern of use. The MIT work measures activity during constrained essays; the student experiments focus on specific classroom tasks; the Microsoft study relies on workers’ reports. None alone provides a robust estimate of how many people experience lasting harm.
Rank #4
Research directions include longer-term knowledge-retention measurement and studies of chatbot feedback, learning outcomes and brain activity. A 2026 systematic review describes mixed findings on critical and creative thinking, with instructional framing and task design among the relevant factors. Those broad observations do not resolve long-term causation. Stronger answers will require follow-up studies that track learning and independent performance over time, not just the quality or effort associated with one AI-assisted task.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
