AI isn’t rotting your brain. It’s doing your thinking for you.

A few weeks ago I came across a claim that chronic use of AI makes your neurons diminish. I loved it. It’s exactly the kind of thing I want to be true, because it backs up how we built Clearahead. I was halfway through writing a post about it when I thought I’d better go and read the actual research.

It isn’t true. Or at least, nobody has shown it. There’s no study measuring brain cells shrinking because someone uses ChatGPT. But what the research does show is, to me, more worrying than the headline, because it’s happening in every company that’s rolled out AI and nobody’s talking about it.

What’s hype

Let’s get the scary stuff out of the way first. The study everyone shares is MIT Media Lab’s “Your Brain on ChatGPT”, from 2025. Researchers put EEG caps on 54 people and had them write essays over several sessions, either with ChatGPT, with a search engine or with nothing but their own head.

The people writing on their own had the strongest, most connected brain activity. The ChatGPT group had the weakest. And the ChatGPT writers struggled to quote a single line of essays they’d written minutes earlier.

That’s interesting. It’s also 54 students writing essays, it hadn’t been peer reviewed when it went viral, and the authors themselves have an FAQ that answers “are LLMs making us dumber?” with a flat “No!”. They ask people not to use words like “brain rot” or “brain damage”. Lower brain activity while you’re writing means you’re putting in less effort. It doesn’t mean you’re losing neurons.

The neuron claim itself comes from research on new brain cells, mostly in animals, which survive when they’re used for learning. Whether adult humans even grow new cells in that part of the brain is still being argued about by neuroscientists. Nobody has connected it to AI.

What’s true

Here’s the stuff that actually stands up, and it’s the bit I can’t stop thinking about.

The doctors who got worse. In 2025, the Lancet Gastroenterology and Hepatology published a study of 19 experienced endoscopists in Poland, every one of them with more than 2,000 colonoscopies behind them. Their clinics brought in AI to help spot polyps. A few months later, on the procedures they did without the AI, their detection rate had dropped from 28.4% to 22.4%. These aren’t students. They’re experts, and a few months of help was enough to blunt a skill they’d spent years building.

The knowledge workers who stopped checking. Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks they’d used AI for. The pattern was simple. The more someone trusted the AI, the less critical thinking they did on what it gave them. The more they trusted themselves, the more they did. The researchers also found the work itself changing shape, away from doing the thinking and towards checking and stitching together what the AI produced.

The drivers who forgot the way. A study in Scientific Reports found that people who’d used GPS more over their lives had worse spatial memory when they had to find their own way. When the researchers retested a small group three years later, the ones who’d leaned on GPS most had declined the most.

None of this is surprising if you think about it. Psychologists have a name for it, cognitive offloading. You hand a job to a tool so your head doesn’t have to hold it. That’s often smart. I don’t need to remember phone numbers. But the skill you offload is the skill you stop practising, and judgement is a skill.

The trap is trust, not laziness

The bit people miss is that this isn’t about lazy people. It’s about good tools. The better the AI gets, the more you trust it, and the more you trust it, the less you check. Researchers in aviation and medicine have been writing about this for decades. They call it automation bias: people go along with the machine’s answer, especially when it’s usually right and they’re busy.

That’s fine when the answer is right. The danger is the confident, fluent, wrong answer that nobody questions because it sounds exactly like all the right ones.

Why leadership teams should care

Now picture this inside a company. Your AI writes the weekly status update. It pulls the metrics, turns them into tidy paragraphs and sends them round. Everyone reads it. The dashboard is mostly green. And not one person in the leadership team can tell you why the number that went red actually went red.

That’s not alignment. That’s autopilot. You’ve saved everyone an hour and quietly removed the thinking that was the whole point of the update in the first place. When the goal slips, nobody saw it coming, because nobody was really looking.

I wrote about why alignment beats activity: a team that agrees on what matters will outrun a busier team that doesn’t. But you can’t agree on something you haven’t thought about. A shared understanding has to be built by the people who share it. An AI can’t do that part for you, however good its summaries are.

What I’d do about it

I’m not anti-AI. I run an AI company. I use it all day. But I’ve got a strong view on where it belongs. AI should take the admin, the chasing, the collating and the remembering. People should keep the thinking and the deciding. I set out that line in The machine asks, people decide, and the research above is the reason I’d never move it.

In practice that means three things for any leadership team:

• Never let a number travel without a reason. If a metric moved, a person should say why, in their own words.
• Make AI ask before it tells. The best prompt is often a question back to you.
• Keep the hard part human. Working out what’s really in the way of a goal is the most valuable thinking a team does. Don’t outsource it.

Use AI to save hours. Don’t use it to skip thinking. Next in this series: AI is making everyone think the same.

Questions people ask

Does using AI damage your brain?

There is no good evidence that using AI kills or shrinks brain cells. The evidence shows something different: people who trust AI tend to think less critically about its answers, remember less of work the AI did for them, and lose skills they hand over to it.

Does AI reduce critical thinking?

A 2025 survey of 319 knowledge workers by Microsoft Research and Carnegie Mellon found that the more people trusted an AI tool, the less critical thinking they applied to its output, and the more they trusted themselves, the more they applied. Other surveys find heavier AI use goes with lower critical-thinking scores, though they show a link rather than a cause.

What did the MIT “Your Brain on ChatGPT” study find?

MIT Media Lab researchers measured brain activity with EEG while 54 people wrote essays with ChatGPT, a search engine or no tools. The ChatGPT group showed the weakest connectivity and struggled to quote their own essays. The study is small, covers one task and had not been peer reviewed, and its authors ask people not to describe it as brain damage.