AI and ADHD: What Helps and What Does Not

A chatbot is genuinely useful for one part of ADHD and useless for another, and the two get confused constantly. It is good at removing a blank page, which is where starting most often fails. It cannot tell you whether you have the condition, and it is unusually convincing while failing to.

Our overview of ADHD covers what the condition involves and how it is assessed; this article is about the tool everyone has now, and about which of the things people use it for actually hold up.

What it is genuinely good at

Removing the blank page. The bottleneck in ADHD is frequently getting started rather than being able to do the thing, and a blank page is the worst possible starting condition because it demands the exact function that is hardest to summon on demand.

A chatbot changes the shape of that problem. It will break a vague task into steps, produce a rough draft you can react to, or turn a tangle of thought into an ordered list. Reacting to something imperfect is a different and much easier operation than generating from nothing, and this is a real match between what the tool does and where the difficulty sits. Our guide to ADHD paralysis covers what that stuck point actually is, which is worth reading first, because the tool only helps if the thing in your way is initiation.

What the evidence does and does not cover

Very little, so far, and it is worth knowing exactly what the good studies looked at. The most relevant published work gathered clinicians who work with children with ADHD, had them interact with a custom chatbot in simulated therapy scenarios, and asked them to rate it: they reported strengths in empathy, adaptability and communication, alongside concerns about privacy and the inability to read anything nonverbal. [berrezueta-2024-chatgpt-adhd]

Read that for what it is. It is expert opinion about a demonstration, which is a reasonable early signal and is not an outcome measured in patients over time. Nothing here establishes that a chatbot improves symptoms, and the honest position is that the question has not been properly asked yet.

Why there is no AI test for ADHD

Because the models that exist have almost never been tried on anyone new. A 2024 systematic review screened 7,764 records and included 100 published prediction models for ADHD, most of them aimed at diagnosis, and found that nearly 90 percent reported high accuracy. [salazar-2024-adhd-models]

The first view is the whole argument. Internal validation means a model was tested on the data it was built from, which tells you it fits those people; external validation means it was tried on a separate sample, which is the test of whether it works at all. Around 96 percent had the first and about 7 percent the second.

So the accuracy numbers are not fraudulent, they are just answering an easier question than the one that matters. That gap is why none of this has turned into something a clinician can order, and why an app promising an instant answer is selling something the field has not built.

The trap that is specific to this

A chatbot agrees with you, and that is the problem for the one question you most want to ask it. Describe your difficulties to one and it will produce a fluent, sympathetic account of how they fit ADHD. Describe the same difficulties with a different framing and it will do that equally well for something else.

Nothing in that exchange tests the alternatives, and testing the alternatives is most of what an assessment is. Difficulty concentrating is produced by anxiety, by not sleeping, by depression and by the aftermath of something difficult, and our guide to ADHD versus anxiety covers how much two of those overlap. Sorting between them is the job, and it is the part a conversation with a model skips entirely while sounding like it did.

The second trap is quieter. If the tool is doing the step you find hardest instead of getting you into it, the practice that would make starting easier never happens. Output stays fine and avoidance deepens underneath it, which is difficult to notice precisely because nothing is going wrong.

What this does not establish

None of this says chatbots are bad for people with ADHD, and none of it says they help. The evidence to support either statement does not exist yet, which is a different thing from evidence of no effect, and the sensible reading is that the tool is worth using thoughtfully rather than either adopted or avoided on principle.

The review of prediction models is also about research models rather than consumer products, and if anything the consumer ones have been tested less rather than more.

When to seek help

Speak to a doctor if difficulty with attention, organisation or starting things has been there since childhood, shows up in more than one setting, and is actually costing you at work, in study or at home. Ask directly about an ADHD assessment, and mention anything you have noticed about sleep, mood and anxiety, because those are what an assessment has to rule in or out.

Bring examples and a rough timeline rather than a conclusion, including anything a chatbot told you if it is on your mind. A clinician can work with what you noticed. What nobody can work with is a label that arrived without the reasoning behind it.

If you are having thoughts of harming yourself, treat that as urgent and contact your local emergency services or a crisis helpline.

How MyFreud can help

MyFreud is useful here for the evidence a conversation with a model cannot produce, which is what actually happened on the days you lost. Dated notes on sleep, mood and what you got stuck on give an assessment something real to work from, and they answer the question a clinician asks first and almost nobody can reconstruct from memory.

Download MyFreud and start today: App Store or Google Play.