Do People With ADHD Have Better Pattern Recognition? I Went Looking for the Study

Part of my Neurodivergent Systems series on tools and workflows for technical work.

This is an essay about pattern recognition in the everyday sense: noticing that two things in your work are the same thing wearing different names. It is not about pattern recognition in the machine-learning sense, and it is not clinical advice or a diagnostic claim about anyone. I have ADHD, I write software, and I wanted to know whether a thing I had been told about myself was true.

The question I actually had

There is a story that circulates about ADHD and technical work. It says that the same attention system that loses your keys is unusually good at spotting patterns: that the failure to filter is also a gift for connection, and that this is why so many people with ADHD end up in architecture, debugging, security, and design.

I have believed a soft version of that story about myself. It is flattering, it is consistent with days where I have genuinely seen something early, and it is repeated back to me often enough that it started to feel established.

That last part is what made me suspicious. A claim that everyone repeats and nobody cites is a claim I have learned to check.

What I found when I checked

I ran a search for the primary evidence on 2026-09-04. That was one search on one day, not a systematic review, and I am not qualified to run one. Weight the next paragraph accordingly.

The shape of what came back is the part I did not expect.

The sources asserting enhanced pattern recognition in ADHD were, with one exception, not primary research. They were a therapy practice page, a wellbeing magazine, and an essay on Medium reframing ADHD as a cognitive architecture. Good faith writing, mostly. But when I followed the claim upstream looking for the study underneath it, the study was not there. The claim cites itself.

The peer-reviewed material I found was mostly about something else entirely. Reviews of the cognitive neuroscience of ADHD describe dissociated multisystem impairments across the networks handling cognitive control, attention, timing and working memory. That literature carefully describes a mixed profile with documented deficits, and makes no case for a hidden strength.

One of the popular sources was honest enough to say the quiet part directly: the research into this specific link is limited. That is the accurate summary. As of today, “people with ADHD have better pattern recognition” is not a finding. It is a hypothesis with a large fan base.

I am labelling that as what it is. Unverified, not false. Absence of a study is not evidence of absence, and I checked for an afternoon, not for a career.

Why I still think there is something real underneath it

If the claim were invention, I would expect it to feel wrong from the inside. It does not. So the more interesting question is what people are actually noticing when they report it.

My best hypothesis, and I am flagging it clearly as a hypothesis I have not tested, is that the popular claim measures the wrong quantity. It says the accuracy of pattern detection is higher. I think what is actually different is the cost structure:

• More candidate patterns get generated, because less irrelevant input is filtered out.
• Fewer of those candidates get suppressed before reaching awareness.
• The hit rate per candidate is probably not better, and may be worse.

That model explains something the flattering version cannot. It explains the bad days. If ADHD made you better at seeing patterns and nothing else, you would not also be the person who confidently connects two unrelated things and spends a day chasing it. Under the cost-structure model, the brilliant catch and the confident wrong answer come out of the same mechanism, in the same proportion they always did. You just notice the catches, because the catches are the ones that get remembered.

Which is its own trap. Discounting your wins and only counting your wins are the same failure of instrumentation, pointed in opposite directions.

The part that changes what you should do

Take the cost-structure model seriously for a second and something practical falls out of it.

If your bottleneck were generating candidate patterns, then more ideas would help. But under this model you are not short of candidates. You are drowning in them. Your actual bottleneck is verification: the cost of finding out, quickly and cheaply, whether the connection you just saw is real.

And verification cost is a property of your tooling, not of the universe.

This is the honest version of the “ADHD advantage in the AI era” argument, and it is considerably less romantic than the version you usually see. AI tooling does not make anyone a better thinker. What it does is collapse the cost of checking. A hypothesis that used to cost an afternoon to test now costs four minutes. When checking is nearly free, generating a large number of mostly-wrong candidates stops being a liability and starts being a reasonable search strategy.

The cognitive style did not become an advantage. The environment around it got cheap enough for the style to pay off.

That is a real difference and it is worth naming, but notice what it implies: the advantage is entirely contingent on actually running the verification. A person with the same cognitive style who does not build the checking habit gets all of the candidate generation and none of the payoff. That is not a superpower. That is a tool requirement.

What this looks like in my own work

I want to ground this in things that exist rather than leave it as theory, so here are three instances from my own setup. All three are the same shape: I generated a candidate pattern, and the only reason I know whether it was real is that something outside my own judgment checked it.

One. I wrote a small Claude Code skill called mirror whose entire job is to counter the distortion of discounting my own progress. It does this by reading the git log. Not by reassurance, by evidence. The skill exists because my internal estimate of what I had done that week was unreliable in a consistent direction, and the repository was not.

Two. I now pre-register my own blog posts. The frontmatter of this page, which you can read in the page source, contains a primary metric, an expected outcome, an explicit falsifier, and the dates I have committed to read the result. I did this because I noticed I was very good at explaining after the fact why whatever happened was what I predicted. Writing the prediction down first is the cheapest instrument I know for catching that.

Three. In August 2026 I ran a piece of analysis that produced a confident, elegant conclusion about a spike in citations to one of my sites. It was wrong. I retracted it twice across five rounds of adversarial review before I got to something that held. What settled it was not the argument. It was reading the daily series instead of the total, at which point the beautiful curve I had been interpreting turned out to be eleven flat days after twelve days of zero.

That last one is the whole essay in miniature. The pattern I saw was vivid, coherent, and false. No amount of additional thinking would have fixed it. Looking at the data one level finer did, immediately.

So, do people with ADHD have better pattern recognition?

Not that I can show you, and not as of today. The claim is popular, under-evidenced, and repeated by sources that are mostly citing each other.

But I do not think the people reporting it are imagining things. I think they are describing a real difference in how many candidate patterns reach awareness, and then reaching for the nearest available compliment to explain it.

The version I would actually defend is smaller and more useful:

If you generate a lot of candidate patterns and filter them poorly, then your leverage is not in thinking harder. It is in making verification so cheap that being wrong often stops costing you anything.

That reframing survived contact with my own data. The superpower version did not.

A note on why this page exists

I am being transparent about the experiment, since it is relevant to anyone who writes on the internet.

The queries this page targets have almost no measurable search demand. Per Google Search Console for chudi.dev, query dimension, 2026-03-06 to 2026-09-02, read 2026-09-04: the entire identity-shaped ADHD cluster on this site is 264 impressions and 2 clicks across 180 days, against 52,304 impressions and 551 clicks for tool-shaped ADHD queries in the same window. By every normal content metric, this page is a bad idea.

I am writing it anyway, because the two most valuable conversations that have come out of this blog both started with somebody arriving on an identity question and not a tool question. Two is not a sample. It is a reason to run a test, which is what this is, with the falsifier written down in the frontmatter before the result is known.

If this reached you and something in it landed, I would genuinely like to know how you found it. That is not a call to action for a mailing list. It is the instrument. The measurement systems I have cannot see the queries that bring people like you here, and the only working sensor I have is you telling me.