Why Bad Sleep Hits ADHD 10x Harder Than Everyone Else
Is AI Making Your ADHD Brain Worse?
Every post on this site about AI has been optimistic. That’s been the honest call — AI scheduling tools, autonomous AI agents that handle triage, coaching apps that actually talk back. Most of them have real merit for ADHD brains.
But there’s a study sitting in the back of every conversation about this, and pretending it doesn’t exist isn’t a service to anyone.
In June 2025, MIT Media Lab published “Your Brain on ChatGPT” — a preprint studying what happens to brain activity when people use AI assistants versus traditional search versus writing on their own. Fifty-four participants. A 32-channel EEG across four sessions spanning four months. The finding that launched a thousand think-pieces: ChatGPT users showed “consistent underperformance at neural, linguistic, and behavioral levels” compared to search users and unaided writers. LLM assistance correlated with the weakest brain connectivity of the three groups. And 83% of ChatGPT users couldn’t quote from essays they had just finished writing.
The study isn’t peer-reviewed yet, and the authors are explicit about treating the findings as preliminary. But the mechanism it describes — “cognitive debt” — hits differently if your brain already struggles with executive function.
TL;DR
Question Honest Answer Does AI help ADHD? Yes, for specific problems — initiation barriers, triage, task breakdown Does AI have cognitive costs? MIT’s 2025 EEG data suggests reduced brain engagement and worse memory for AI-assisted work Are ADHD brains at higher risk? Possibly — the mechanisms involved (working memory, task ownership, executive control) are where ADHD already struggles Is AI dependency a real concern? Clinical psychologists think so. The setup dopamine loop is a documented ADHD trap Should you stop using AI tools? No. But the difference between intentional and reflexive AI use matters a lot One-sentence verdict: AI can be an accommodation or a crutch — the difference is whether you’re using it to do what your brain can’t, or instead of doing what it can.
Who should read this: Anyone who’s noticed they can’t think through a task without pasting it into ChatGPT first.
Who this isn’t for: People who use AI narrowly and deliberately and have clear stops built in.
What the MIT Study Actually Found
The MIT study isn’t about AI being “bad.” It’s more specific than that, and the specifics matter.
Participants were split into three groups: write essays using ChatGPT, write using Google search, or write with no tools at all. The EEG data tracked brain connectivity patterns across sessions. The finding wasn’t just that ChatGPT users showed lower engagement in one region — it was that the pattern held across the whole brain. The unaided writers had the strongest, widest-ranging connectivity networks. Search users were in the middle. ChatGPT users showed the weakest overall coupling.
The behavioral data was worse. ChatGPT-assisted essays used similar vocabulary, similar ideas, similar sentence structures — regardless of the individual writing them. The AI flattened variation. And the memory failure was striking: 83% of participants who wrote with ChatGPT couldn’t accurately quote from the essay they had just finished. They hadn’t processed the content as theirs.
The authors call this “cognitive debt” — the accumulation of long-term cognitive costs when AI handles tasks the brain would otherwise have to engage with fully.
Here’s the part nobody talks about enough: the study used neurotypical participants. The researchers weren’t studying ADHD. But the cognitive functions that took the biggest hits — working memory, executive control, linguistic ownership of output — are the exact functions that ADHD impairs most.
That’s the uncomfortable question clinical researchers are now asking.
Why ADHD Brains Might Be More Vulnerable, Not Less
The intuitive argument for AI and ADHD is straightforward: our executive function is impaired, so offloading executive function tasks to AI reduces the gap between what we need to do and what we can do. Clean logic.
But there’s a counter-argument that doesn’t get enough airtime.
Compensatory skill development is a real thing. ADHD brains, over years of struggling with executive function demands, build workarounds. Not perfect ones — but genuine adaptations. Systems for breaking down tasks. Strategies for catching memory gaps. Personal tricks for making initiation possible. These develop through repeated engagement with the problem, which is uncomfortable but also where learning lives.
The concern emerging in clinical practice: AI might eliminate the opportunity for those adaptations to form. If every time you hit the initiation wall you paste the task into ChatGPT and let it break things down, the wall never gets any shorter. You’re not building anything. You’re just going around it every time, and the going-around becomes the dependency.
One way researchers frame this: the brain, like muscle, requires regular exercise. That’s not a new metaphor. But it applies with particular force when the thing being outsourced is exactly the function you’d most benefit from strengthening.
None of this means AI help is useless. It means there’s a difference between using AI as an accommodation for tasks that genuinely exceed your current bandwidth, versus using it as a reflex to avoid any friction at all.
The second pattern is worth watching for.
The Dopamine Problem Is Different for ADHD Brains
There’s a specific trap ADHD brains fall into with AI tools.
Setting up AI systems feels productive. Genuinely, immediately rewarding. You’re building something clever, solving a meta-problem. The dopamine is real, and the sense of forward motion feels completely legitimate.
The task itself often doesn’t get done.
This is a documented ADHD pattern — the preparation phase becomes the work, and the work itself gets indefinitely deferred. With AI, the preparation phase has gotten much more elaborate and much more satisfying. Spending three hours building a ChatGPT prompt system that will “automatically help me process my inbox” is an extremely engaging activity. It’s also not processing the inbox.
Clinical psychologists working in ADHD specifically have started calling this out: the dopamine hit from designing automations can function as a substitute for task completion. You got the reward without doing the thing. Which, for ADHD brains already prone to dopamine-seeking detours, is a meaningful risk.
The hyperfocus rabbit hole has always existed. AI just gave it a new form that’s particularly easy to rationalize as work.
What Clinical Psychology Is Saying
In February 2025, Psychology Today published “ADHD, Executive Functions, and AI: A New Era in Treatment” — a piece that approached AI as potential treatment infrastructure for ADHD. The framing was positive: AI as “an agent that supports ineffective executive functioning skills.”
But the piece also surfaced a question that’s gotten less attention than the optimistic framing: whether AI use is intentional or reflexive.
Intentional AI use looks like this: you identify a specific executive function task that consistently exceeds your capacity, you use AI to handle that specific task, and you keep the rest of your cognitive engagement intact. You’re using AI the way you’d use a whiteboard — it holds information so your brain doesn’t have to, which frees your brain to do more of the actual thinking.
Reflexive AI use looks like this: any time friction appears, you open ChatGPT. Email you don’t know how to word. Decision you don’t want to make. Task you don’t want to start. The AI handles it, or at least delays the moment you have to. The friction goes away, along with the cognitive engagement the friction would have produced.
The distinction matters because reflexive use is how cognitive debt accumulates. Not from using AI at all, but from using it in response to any demand rather than in response to specific ones.
For ADHD brains, this is harder to maintain than it sounds. The ADHD brain is, by definition, less able to stop and ask “should I actually be handling this myself?” before reaching for the nearest tool. The reflex is faster than the metacognition.
The Research Gap (And Why It Matters)
The honest answer to “is AI making ADHD brains worse?” is: we don’t know yet.
The MIT study is a preprint studying neurotypical adults in a writing context. It’s meaningful, it’s concerning, and it’s preliminary. The specific question of how AI affects ADHD brains over time — whether it reduces symptoms or calcifies deficits, whether it supports skill development or prevents it — hasn’t been studied longitudinally in neurodevelopmental populations.
Clinical researchers are actively calling for that research. The existing literature on AI and ADHD is almost entirely about AI as a positive treatment adjunct, not about potential negative effects of long-term dependency. That’s a gap worth acknowledging.
What we can say: the evidence for evidence-based ADHD strategies — things like CBT, exercise, structured routines, body doubling — has decades of longitudinal data behind it. The cognitive costs and benefits are well-characterized. AI tools have six months of EEG preprints and a lot of anecdotal enthusiasm. Those aren’t equivalent levels of evidence.
That’s not a reason to avoid AI. It’s a reason to stay curious rather than committed.
A Framework for Not Outsourcing Your Brain
This isn’t “stop using AI tools.” Most of what we recommend on this site is still worth using, and used deliberately it genuinely helps.
But here’s a set of questions worth asking before you reach for AI on a task:
Is this task genuinely exceeding my capacity right now, or does it just feel uncomfortable?
There’s a real difference. If you’re in full cognitive shutdown and can’t formulate a coherent sentence, AI writing help is an accommodation. If you’re just avoiding the mild friction of figuring out how to word something yourself, you’re skipping the cognitive exercise, not the cognitive barrier.
Am I going to actually own the output?
The MIT finding about memory is the most actionable one. If you use AI to write something and then don’t deeply read and internalize what it produced, you’ve done less than you think. The output exists. You don’t own it. For tasks where ownership matters — decisions, responses to complex problems, anything you’ll need to explain or build on later — passing through AI without engaging fully with the result is often worse than doing it slower yourself.
Is this the fifth time this week I’ve used AI to avoid this exact type of task?
Pattern recognition matters here. A single instance of AI-assisted decision-making is a useful accommodation. A consistent pattern of using AI to avoid any task that requires sustained focus is worth examining. The decision fatigue tools that actually help work because they reduce unnecessary choices — not because they eliminate all the choices that require your judgment.
Am I in the setup phase or the work phase?
If you’ve spent more than 45 minutes building a system that will make a task easier, and the task itself still isn’t started, you’re in the setup loop. Finish the setup. Do the task. The system can be refined later.
What Intentional AI Use Actually Looks Like
For ADHD specifically, there are tasks where AI offloading is genuinely appropriate:
Email triage. Deciding which of 80 emails needs a reply today is an executive function task with low cognitive value. Tools like the ones we covered in best AI email apps for ADHD handle triage so you can focus on the replies themselves.
First draft of something you’ll rewrite. AI gives you something to react to, which is easier for ADHD brains than creating from nothing. But you have to actually rewrite it — in your words, with your thinking — not just clean it up.
Breaking a task down when paralyzed. The initiation barrier is real, and AI task decomposition helps at the specific moment when you can’t figure out what the first step is. Use it for that moment, then put it away.
Recurring administrative tasks with no cognitive value. Formatting, summarizing meeting notes you’ll review, filling in standard templates. Low cognition, high friction reduction. Reasonable trade.
Where it gets risky: any time AI is handling something that requires your judgment, your memory, your understanding of a situation — and you’re not deeply engaging with the output.
Our Take
The productivity internet is going to keep writing optimistic AI posts, including probably this site. Most of them will be accurate for specific use cases.
But the MIT “cognitive debt” finding describes a real mechanism, and ADHD brains are already in deficit on the functions it targets. That’s not a reason to avoid AI entirely. It’s a reason to be more deliberate about when and how.
The question to ask isn’t “should I use AI?” It’s “am I using AI to extend my capacity, or to avoid exercising it?” Those produce very different outcomes over time. And for ADHD brains already prone to avoidance patterns, the second one can be invisible until it’s already compounded.
The best tools we’ve covered on this site work because they handle the friction that ADHD genuinely can’t manage — not because they handle all friction, including the friction that’s supposed to be there.
There’s a difference. It’s worth keeping track of which side of it you’re on.
The irony of an AI-assisted site publishing this is not lost on us. The thinking behind it is still ours.
