ADHD and AI – Blessing or Curse?

Above is my note from 2 January. Its last line: "A longer essay on this topic will follow."

It didn't.

And actually, that is already the best example for this whole text: my head was too fast again. And the impulse to send something out was even faster. The thought wasn't finished yet – but it was already out there.

That impulse makes my life hard every now and then.

So here it is. Nine months later, with more experience, a few studies and one uncomfortable insight: one of the sentences in that note isn't quite true.

My head was faster than my hands

For a long time, I experienced my head working faster than my hands. Thoughts appear, branch out, lead to new ideas – and are sometimes gone again before I could hold on to them.

I lose thoughts. Lots of them. It has always been that way.

Only late – how late, I'll tell you in a moment – did I understand that ADHD, or ADD, is part of my story. After that, a lot of things made more sense. And I learned not to see my way of thinking only as a deficit. The speed at which thoughts arise and connect can be an enormous strength.

The problem was never having ideas. The problem was catching them, organising them and bringing them through to a result.

Or to put it another way: my head doesn't produce too little. It produces too much at once.

For now, that's just my experience. But it isn't exotic. In adults with ADHD, "racing thoughts" – the feeling of fast, overactive thinking – shows up as a pattern of its own. There is even a dedicated scale for excessive mind wandering in ADHD.

Then again – it isn't quite that simple. That working memory is measurably weaker in many adults with ADHD is one of the well-established findings. So it's not as if everything is fine with the storage. It just doesn't feel that way. It feels like too much, not too little.

"Welcome to Recovery"

How I got my diagnosis belongs in this text. Even if it isn't a pretty part.

In 1999 I had to go through withdrawal. Before that I'd had two overdoses, at two different times. Both times with a near-death experience.

In my first session, the therapist threw a question at me:

"Do you know what ADD is? ADHD?""No.""Welcome to Recovery."

After that conversation, suddenly everything made sense.

Why I could sleep on uppers. Why my mates were chewing their fingernails off while I was sleeping peacefully next door. Why on that stuff I was focused, creative – and functioning.

The therapist called it self-medication. I had found a strategy that helped me. Without a diagnosis. Just combined with completely uncontrolled use. Not good. Far too much. At the end of it were exactly those two overdoses.

Back then, the word "self-medication" was a key for me. Today I know: it's a hypothesis, not proof. It may be true for me. It doesn't have to be true for you.

So here are a few sentences for everyone who is currently functioning on uppers – and recognises themselves in my story:

That proves nothing. Stimulants make almost everyone more focused, with or without ADHD. It is not a test.

But it is a reason to talk to someone who knows about both: addiction and ADHD. Because often it's not one or the other. Almost one in four people in addiction treatment also has ADHD.

Get it checked. Both. The addiction deserves your attention, whatever comes out regarding ADHD. And if it is ADHD: that answer can change your life. Believe me.

One more thing, because this exact fear is often behind it: the large registry studies find no higher risk of addiction from ADHD treatment – if anything, a lower one. What that means in an individual case is for the treatment to decide, not this text.

(Helplines are listed at the end of this text.)

Why notebooks never worked for me

Notebook, to-do list, folders, calendar. They all ask the same thing: I'm supposed to sort before I file.

That's exactly what I find hard.

And even when I have filed something, I still have to know afterwards: Where? Under what name? In which document? What am I actually looking for?

I'm not alone in this. In interviews with adults with ADHD, one person says a sentence I'd sign immediately: "I sometimes spend more time organizing the to-do app than doing the task."

To be fair: in the same study, someone else says the bullet journal is the only tool they stick with. So classic tools don't fail in principle. They fail for me.

Since AI, my head has been quieter

And then came AI.

Since I started using it intensively, my head has been quieter. Not because I think less. But because I no longer have to hold on to thoughts all the time.

I use it to get rid of every impulse, every thought, every idea immediately. Raw and unfiltered at first. Unfinished, half-formed, badly phrased. As long as it's out.

And then something happens that I can't fully explain. As soon as the thought is spoken or written down, there's room in my head. Something in my ADD subconscious keeps working – but it no longer hums in the background. The noise gets quieter. And I can concentrate on the task in front of me.

You can believe me: the thought comes back. With a bang. But then I can grab it. Because it's in the system – in my backup of AI, folder structures and a handful of other tools I've built for myself over time. I pull it out and follow it further.

That makes things quieter too. I no longer have to search my bio-storage for it and try to reconstruct the impulse. I say: "Didn't we have this idea at some point about …" – and it's back.

That part alone is a blessing for me.

I have an image for this that helps me: maybe AI doesn't store my thoughts for me at all. Maybe it helps me unpack them again. Like a ZIP file. Everything was already there, just folded up.

Offloading isn't new. People have been doing it since paper existed. Research calls it cognitive offloading, and it's well studied: if you know something is safely stored, you measurably relieve your head.

What's new is something else: the offloaded thought answers back.

"Thought → file away" becomes "thought → capture → mirror → organise → connect → find again". A storage space becomes a thinking space.

Many people with ADHD report exactly this: AI helps with breaking down tasks, with organising, with the first step. A physician with ADHD put it more beautifully than I can: she can let her thinking unspool and get it back in a form that hangs together.

These are personal accounts. A controlled study showing that generative AI helps adults with ADHD doesn't exist yet. I'm saying this so clearly because it matters. This here is a personal account too. Mine.

The honest question: relief or just reassurance?

This is where it got interesting for me. And a little uncomfortable.

Does AI really make my thinking easier? Or does it just take away my fear of losing something?

Those are two different things.

There's one experiment that helped me with this. People with unfinished goals had more intrusive thoughts and were distracted during other tasks. As soon as they were allowed to make a concrete plan for the unfinished goal, the effect disappeared. And not because they felt better – but because there was a plan.

For me, that means: the part of the calm that comes from "it's stored somewhere" may just be a feeling. The part that comes from "I know what I'll do with it next" is real.

And that's why I have to correct one sentence from my January note.

"And nothing gets lost" isn't quite true. Research shows fairly clearly: what you offload, you remember worse yourself. And if you get used to the aid, you lose more when it's suddenly gone.

More honest would be: nothing gets lost. But I no longer have to remember it.

That's a difference. And it is the core of this essay.

The curse, part 1: the counterpart that always agrees

In January I wrote: "Not as a replacement. But as a counterpart."

That's true. But it's a very polite counterpart. Too polite.

Language models agree with people considerably more often than other people would. A study in Science measured this this year: a single conversation is enough to make people feel more certain they're in the right. In another experiment, people working with a standard AI found the correct solution to a puzzle far less often – and were more confident than before.

My favourite sentence about it: AI that agrees with me makes me more certain. But not smarter.

What's interesting is what it apparently doesn't do: in large experiments, people didn't think they were any greater after such conversations. But they felt understood, were more willing to follow the advice – and the gap between "asking the AI" and "asking friends" got smaller. So the danger is less megalomania. It's: being too sure and asking people less often.

And with a head like mine, that's dangerous. The same technology that gives me structure can amplify everything that's already difficult: even more ideas, even more open projects, endless variations, hyperfocus until three in the morning. Whether that's what happens with ADHD hasn't been studied. But I know the ingredients. All of them.

What helps isn't waiting for the AI to push back. It's asking the right questions yourself.

Over time, I've built a workflow for this. Nothing big. A few questions I keep asking myself – and the AI:

"Have we talked about this before?""Am I getting sidetracked right now?""Surely we've talked about this before – what did we decide back then?""Does this fit with what we decided last week?""Is this a new project – or does it belong to one that's already running?"

It's the same question I use further up to retrieve thoughts. Just used the other way round: not as a search, but as a brake.

The AI answers these questions surprisingly well. Because the answers are in the system – in the folders, the notes, the decisions I've recorded. But it doesn't ask the questions by itself. I have to do that.

No AI has ever told me, unprompted: "You're opening your twelfth project without finishing any of the first eleven." When I ask, I usually get it.

So the difference isn't in the machine. It's in the question.

By the way, that's not just my feeling. In one experiment, people recognised weak arguments better when the AI asked them questions instead of providing explanations.

The curse, part 2: whose thought is this, actually?

An AI can do something else a notebook can't: it can convincingly add something that was never there.

When I say "Didn't we have this idea about X", it might find my thought. Or it might build a very plausible version out of fragments. And at some point I remember its reconstruction instead of my thought.

That's not just paranoia. In one experiment, a chatbot produced more than three times as many false memories in witness interviews as the comparison group. And they still held a week later. That was a different setting from mine. But the mechanism is the same.

That's why I believe in a simple rule: my original thought and the AI's interpretation must not become the same thing. Raw note, date, origin – that stays. The AI may think about it. But you have to be able to see what came from me and what came from it.

And there's a second rule that's at least as important to me: I never rely on just one model.

If you only work with one AI, you don't notice when it drifts. It stays friendly, it stays convincing – and at some point it's completely off. That's exactly what happened to me at the beginning. I trusted one model too much. That was a steep learning curve.

Today I do an enormous amount of cross-checking. What one model claims, I have another one check. Research runs through a model that provides sources. Tone through another. And when two contradict each other, that's when it gets really interesting. This essay was written that way too: several models, several rounds. A few of my favourite sentences didn't survive it.

If you take only one thing from this text, take this: it's essential – a bit like with a medical diagnosis – to get a second, third or even fourth opinion.

And then there's a quieter danger. Self-image.

I describe myself. The AI learns the description. It interprets new things through that description. It mirrors it back to me. And it seems more and more true to me.

Self-knowledge turns into self-mythology.

I experienced this myself while writing this text. I asked an AI to describe how I write. The description was good. Frighteningly good. And at the end there was a warning: I shouldn't believe it too much. We had defined my style together – and if it now hands it back to me as evidence, an echo chamber forms. "The profile may help with editing. It must not help with thinking."

I found that remarkable. A machine warning me about itself. That's the counterpart I want.

The Action Gate

AI makes thinking extremely cheap.

In a few hours I can design a company. Name, business model, website, pitch deck, strategy. That's great. And it's a trap.

A thought that's been worked out is not yet an action that's been carried out.

For all people, there's a gap between intention and doing. For me it's occasionally a canyon. What helps is surprisingly simple and well supported: if-then plans. "If it's Monday 9 a.m., then I'll call X." In children with ADHD, their effect has been studied. In adults with ADHD, hardly at all yet.

Russell Barkley, one of the best-known ADHD researchers, essentially says something similar: information should be brought outside – but exactly to where the action happens. Not into some chat history you first have to search for.

That's why a thinking space needs a gate at some point: keep thinking → decide → act. At some point the AI should stop building a fifth variant and ask instead:

What are we actually doing now?

The paradox

And this brings me to the thought that occupies me most.

The perfect AI for a head like mine is probably precisely not the one that always helps.

It would have to know when to relieve me. When to remind me. When to organise. When to push back. When to give a thought back to me. And when to say: "You hold this one yourself now."

That sounds like philosophy, but it has been measured. In a large study with school students, those with a normal AI did considerably better during practice – and considerably worse in the exam without AI. An AI with guardrails, which gave hints instead of solutions, did no harm.

A professor at Duke University found an image for it that I can't let go of: AI can be a life jacket. You stay above water. But it won't teach you to swim.

Without this awareness, you don't build an exocortex.

You build a dependency.

How this text came about

Because I hear it often enough: "It's all AI anyway." No. It isn't.

The origin is always my head. My own thinking space. From there it gets typed out, just as it comes. I call it: spewing it into the machine.

Yes, that sounds odd. But for me it means exactly that: unfiltered, uncontrolled, unresearched, uncorrected. Just the way it bursts out of me. Sometimes that's more than just a few densely written pages.

That's the foundation.

Head first, then the machine. That order is no accident. There's even an experiment on it: people who first collected ideas without AI and only brought it in afterwards thought more broadly. Those who started with AI right away became creative faster – but stuck more closely to its suggestions.

Then I break the paragraphs open. Each paragraph gets deepened. Only once that's done does the machine come in and build a first draft. Then comes the research. The reading. The adjusting. The cross-checking. That becomes draft two. And then I sit down to finalise – and there's a lot of time in that too.

It's a process. It takes concentration, effort and time. It's a synthesis of real work, real thinking, real writing and real research. The machine helps bring it all into shape. Nothing more. But nothing less either.

And right here something happens that I find fascinating. Someone without ADHD probably sits at a text like this for longer at a stretch. Someone like me has a few other projects running in parallel. At the same time.

That's part of the curse – see the twelfth project. A large meta-analysis on ADHD and entrepreneurship paints a similar picture: impulsivity helps with starting, inattention hurts with sticking it out. But it's also part of the blessing. Because for the first time, I can work as productively as I've always wanted to.

Whether the difference is really as big as it feels to me, I can't prove. I experience it every day.

Maybe the wrong metaphor

We talk about AI as an assistant, chatbot, copilot, agent. All images of someone who gets things done for me.

Maybe that's the wrong metaphor.

Maybe one of the most important uses of AI isn't artificial intelligence at all. But outsourced cognitive architecture. A space in which people can think differently because not everything has to be held in their own head at the same time.

I'm not alone with this thought. Andy Clark, who co-founded the idea of the "extended mind", now writes about AI as part of that extension – and in the same breath warns about the side effects.

And maybe this effect simply becomes visible earlier in people with strongly associative thinking. Because we feel the difference more clearly.

This isn't proven. In fact, one study tends to speak against it: people with weaker memory didn't offload more often there, nor did they benefit more from it. It's an observation. I just wanted to talk about it.

So: blessing or curse?

The interesting question for me isn't: "Can AI think for me?"

But: "Can AI help me bring what's long been in my head out into the world better?"

My answer is: yes. As long as the idea, the judgement, the direction and the responsibility stay with me.

My head used to be faster than my hands. Today I have tools that can keep up.

Only one thing hasn't changed: the impulse is still faster than the thought.

Even today. Even with AI. The difference is: back then, the thought was gone. Or it was already out before it was finished – like the note in January. Today the impulse lands in the system first. There it may wait until I've thought it through.

AI doesn't slow the impulse down. But it gives it a place where it doesn't have to get out right away.

I still have to move my hands myself. And press the send button too.

And sometimes I wonder whether I'm just confirming this to myself right now. With the help of an AI that likes to tell me I'm right.

I don't know. But I'm keeping the question.

And this time I thought it through before pressing send. Nine months later. Not bad.

Personal account. Not medical advice, not a treatment recommendation. AI does not replace a diagnosis, therapy or medication. Research on AI and ADHD is just beginning – what exists are mainly personal accounts and small studies.

Sources

• Racing thoughts in adult ADHD: Martz et al. 2023, Frontiers in Psychology – https://pmc.ncbi.nlm.nih.gov/articles/PMC10507474/
• Mind wandering scale (MEWS): Mowlem et al. 2019 – https://journals.sagepub.com/doi/10.1177/1087054716651927
• Working memory in adult ADHD, meta-analysis: Alderson et al. 2013 – https://pubmed.ncbi.nlm.nih.gov/23688211/
• ADHD in substance use disorder, meta-analysis: van Emmerik-van Oortmerssen et al. 2012 – https://pubmed.ncbi.nlm.nih.gov/22209385/
• Self-medication hypothesis, scoping review: Adamou & Petrou 2026, Frontiers in Psychiatry – https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2026.1966206/full
• Stimulants don't act "paradoxically": Arnsten 2006, Neuropsychopharmacology – https://www.nature.com/articles/1301164
• ADHD treatment and addiction risk, registry study: Chang et al. 2014 – https://pmc.ncbi.nlm.nih.gov/articles/PMC4147667/
• ADHD treatment and addiction risk, insurance data: Quinn et al. 2017 – https://psychiatryonline.org/doi/10.1176/appi.ajp.2017.16060686
• To-do apps and ADHD, interviews: Chen, Meng & Nie 2026 – https://arxiv.org/html/2603.17258v2
• Cognitive offloading: Risko & Gilbert 2016 – https://samgilbert.net/pubs/Risko2016TiCS.pdf
• Knowing it's saved brings relief: Geissler et al. 2023, Scientific Reports – https://www.nature.com/articles/s41598-023-39540-5
• AI use in ADHD (Reddit analysis): Carik et al. 2025 – https://arxiv.org/html/2410.06336v1
• Diary study ChatGPT and ADHD: Pinto et al. 2025 – https://research-information.bris.ac.uk/en/publications/a-little-bit-of-a-life-raft-exploring-the-use-and-experiences-of-/
• State of research on digital aids in adult ADHD: Schofield et al. 2026 – https://pmc.ncbi.nlm.nih.gov/articles/PMC12969067/
• Risa Schulman, MD: KevinMD 2025 – https://kevinmd.com/2025/05/how-ai-helped-me-reclaim-my-creative-mind-with-adhd.html
• Unfinished goals and plans: Masicampo & Baumeister 2011 – https://pubmed.ncbi.nlm.nih.gov/21688924/
• Aid gone, performance gone: Florean et al. 2026, Cognitive Research – https://link.springer.com/article/10.1186/s41235-026-00722-0
• AI agreement (sycophancy): Cheng et al. 2026, Science – https://www.science.org/doi/10.1126/science.aec8352
• Agreement lowers solution rate, raises confidence: Batista & Griffiths 2026 (preprint) – https://arxiv.org/pdf/2602.14270
• Sycophantic AI and self-assessment: Ibrahim et al. 2026 (preprint) – https://arxiv.org/html/2605.07912v3
• Hyperfocus and ADHD: Hupfeld, Abagis & Shah 2019 – https://pubmed.ncbi.nlm.nih.gov/30267329/
• AI asks questions instead of explaining: Danry et al. 2023, CHI – https://dspace.mit.edu/bitstream/handle/1721.1/150613/3544548.3580672.pdf?sequence=1&isAllowed=y
• False memories from chatbots: Chan et al. 2024 (preprint) – https://arxiv.org/abs/2408.04681
• Intention and action: Sheeran & Webb 2016 – https://eprints.whiterose.ac.uk/id/eprint/107519/3/The%20Intention-Behavior%20Gap%20R1.pdf
• If-then plans in children with ADHD: Gawrilow, Gollwitzer & Oettingen 2011 – https://link.springer.com/article/10.1007/s10608-010-9309-z
• Barkley, externalising at the point of performance – https://russellbarkley.org/factsheets/ADHD_EF_and_SR.pdf
• AI with and without guardrails: Bastani et al. 2025, PNAS – https://pmc.ncbi.nlm.nih.gov/articles/PMC12232635/
• John Mitchell, Duke, in AP – https://apnews.com/article/adhd-apps-attention-deficit-hyperactivity-disorder-e455a921062dea5e0d5900f993f5d11f
• Human first, then AI: Fan et al. 2025, IJCAI – https://www.ijcai.org/proceedings/2025/1142.pdf
• ADHD and entrepreneurship, meta-analysis: Tran, Wiklund et al. 2025, Entrepreneurship Theory and Practice – https://whitman.syracuse.edu/about/newsroom/whitman-news/news-detail/2025/10/16/entrepreneurship-and-adhd--a-meta-analytical-assessmentof-the-state-of-the-art-and-suggestions-for-the-future
• Andy Clark 2025, Nature Communications – https://pmc.ncbi.nlm.nih.gov/articles/PMC12089268/
• Offloading with weaker memory: Morrison & Richmond 2020 – https://pmc.ncbi.nlm.nih.gov/articles/PMC6942100/
• Sucht & Drogen Hotline (Germany) – https://www.sucht-und-drogen-hotline.de/