The "is ChatGPT making you dumber?" question has been asked enough that it deserves an answer that's neither the alarmist headline nor the defensive shrug. The published evidence supports a careful, narrow position. This post walks through what the research actually shows, what it doesn't, and what to do about it.
The strong claim — and why it's not supported
The strong version of the question — "does ChatGPT make people stupider in general?" — has no supporting evidence in the published literature. There is no study showing ChatGPT use causes IQ change, no study showing it causes any clinical cognitive condition, no study showing it produces lasting cognitive harm. Anyone making that claim has gone past what the evidence supports.
What "dumber" would even mean is the first problem. Cognitive psychology distinguishes between fluid and crystallised intelligence, between specific working-memory capacity and broad reasoning, between learned skills and underlying mechanisms. The headline framing usually collapses all of this into one word, and the published research doesn't.
The narrower claim — and what supports it
The narrower version — "does heavy AI use weaken specific cognitive habits?" — has more substantive support.
The cleanest piece of recent empirical work is the 2025 MIT Media Lab preprint by Kosmyna et al. (arXiv 2506.08872; TIME coverage). The study compared essay writers in three conditions — brain-only, search-engine-assisted, and LLM-assisted — using EEG, recall tests, and linguistic analysis. The LLM-assisted group showed the weakest neural connectivity during composition, the lowest self-reported ownership of the essays they produced, and the worst recall of their own arguments afterward. The paper introduces the phrase "cognitive debt" to describe the gap between using a tool to produce a result and being able to do that same thinking yourself afterward.
The Stanković et al. 2026 commentary on the Kosmyna study raises methodological concerns about sample size and EEG interpretation. Both are preprints. Neither is the final word. The honest posture is to read them together.
The broader cognitive-offloading literature provides the older context. Risko & Gilbert's 2016 review in Trends in Cognitive Sciences is the foundational reference. Sparrow, Liu & Wegner's 2011 Science paper on the Google effect showed that when participants believed information would be available online later, they remembered the information itself less reliably. The 2024 MDPI Societies study extended the framework to AI tools and found an inverse correlation between AI usage frequency and self-reported critical-thinking engagement.
The pattern across these sources: AI doesn't make you dumber in general. It changes which cognitive acts get daily reps. The acts that get fewer reps tend to weaken, in line with the older skill-decay literature.
What this means in plain English
Three claims the careful research supports:
- Practising a cognitive skill keeps the skill in regular use.
- AI tools can substitute for cognitive acts you used to do yourself — drafting, summarising, calculating, recalling, reasoning.
- Acts that are substituted-away tend to get less practice. Less practice tends to mean weaker performance on those specific acts.
Three claims the careful research does not support:
- AI use causes general cognitive decline.
- AI use produces measurable IQ change.
- AI use causes any clinical condition.
The distance between these two sets is what the headlines collapse. The honest framing sits in the gap.
What to do — without quitting AI
The practical advice across the 2024-2026 literature is consistent. Keep using AI when it genuinely helps. Keep daily reps on the cognitive surfaces you want to remain individually capable of.
Concretely:
- Write 200 words a day without AI. Not for a deliverable — for voice continuity.
- Read at least one full article a day without AI summary. Sustained attention is a habit and habits respond to use.
- Do one calculation in your head per day. Estimation, ballparking — the muscle of number sense.
- Make one decision per day before asking AI to help. Even small ones. The reasoning is the rep.
These are tiny commitments deliberately. The point isn't volume — it's preserving the daily existence of these acts. Senwitt's daily Set is one delivery mechanism; the daily-set page lays out the design.
The 2026 reading
If you only read one thing on this question, the research/your-brain-on-chatgpt page is our deep-dive treatment of the Kosmyna study with the Stanković critique alongside. The research/cognitive-debt page covers the broader cognitive-debt framing. The research/ai-overreliance page extends the framing to the workplace.
For the question this post is actually titled around — whether ChatGPT is making you dumber — the careful answer is: probably not in the way the headlines say, but in a narrower way that's worth taking seriously. The fix isn't quitting AI; it's keeping daily practice on the side.
Why the headline framing keeps coming back
A note on what the headline writers are reaching for that the research does not support. The strong claim — "AI is making us collectively dumber" — is a great headline, and it slots into a much older anxiety about new technologies and human capability that has been recurring since Plato's critique of writing. The same shape of claim has been made about television, calculators, search engines, smartphones, and now AI. The pattern is consistent enough to be its own genre. Each cycle has produced some narrow legitimate findings (cognitive offloading is real, recall shifts when a tool is available, sustained attention has a habit-component that responds to use) inside a broader cultural narrative that has gone past the evidence at every turn. The honest read on AI is the same as the honest read on each of the predecessors. Specific cognitive habits respond to use; the technology changes which habits get used; the people who keep the practice intentional tend to keep the underlying surface intact while still getting the productivity benefit of the tool. The strong-claim headline is wrong about each previous technology and is likely to be wrong about this one. The narrow claim is right about each previous technology and is likely to be right about this one.
What this implies for the next few years
A final practical note. The current AI-and-cognition research literature is thinner than the technology adoption curve suggests it should be. The 2025 Kosmyna preprint and the 2026 Stanković critique are early signals; the broader empirical base is still being assembled. We should expect the next two to three years to produce both stronger evidence and stronger contestation, with the public-facing summaries oscillating between the alarmist and the dismissive readings. The careful posture is to weight the consistent direction of the older offloading literature (where the evidence is robust) more heavily than any single new preprint (where it is not yet settled), and to build the daily practice around what the durable older finding supports. The strong claim that AI is making everyone dumber will be wrong. The strong claim that AI changes nothing about how cognitive habits develop will also be wrong. The truthful position is the smaller, more practical one: keep the daily reps on the cognitive surfaces you want to remain individually capable of, and let the research and the technology continue to evolve around that discipline.
