ADHDHow Much Faster Does AI Actually Make You?
Somewhere on LinkedIn right now, a post is promising AI will make you "10x faster," without saying where the number came from. Some of those numbers are real research. Some are marketing copy dressed up as a stat. Below are the four real studies that actually measured this, ranked from most trustworthy to least, and then the one honest fact that changes how you should read all of them: none of these studies were done on ADHD adults.
The most rigorous number we have
MIT economists Noy and Zhang ran a preregistered randomized controlled trial and published it in Science in 2023. That combination matters. Preregistered means they wrote down what they were testing and how before they saw any results, so they couldn't quietly change the question to fit whatever the data showed. Randomized controlled means some of the 453 professionals in the study got ChatGPT and some didn't, and the two groups were otherwise similar, so the difference between them is a fair comparison.
The result: ChatGPT cut writing task time by 40%, and the quality of the writing went up 18%. The gain wasn't even across the board either. Workers who scored lower on the writing task without AI gained the most from using it. That's the single best-designed study we have on AI and speed, and it's worth remembering that even here, the gain is for writing tasks specifically, not "everything."
A bigger, messier real-world test
The next study down the rigor ladder comes from NBER, the same researchers who study labor economics for a living. Brynjolfsson, Li, and Raymond looked at 5,179 real customer support agents using AI assistance on the job, not in a lab. AI assist raised average productivity by 14%. But that average hides the real story: novice agents gained 34%, while experienced agents gained almost nothing.
That's a meaningful pattern, not just a number. AI assistance works best as a floor-raiser. It closes the gap between someone new to a skill and someone who already has years of it, rather than making the expert dramatically better. If you're newer to something, whatever it is, that's actually good news for you specifically.
The shipped-work catch
The next study looks like the biggest win and hides the most useful catch. A 2026 NBER paper tracked over 100,000 developers using AI coding tools and found their coding activity rose somewhere between 40% and 180%. That sounds enormous. But the same study found that actual shipped releases, the finished work that went out the door, only rose about 30%.
Activity went up 40 to 180%. Finished, shipped work went up about 30%. The gap between those two numbers is the whole lesson.
Activity isn't output. Writing more code faster doesn't matter if someone still has to review it, test it, and decide it's actually ready. Humans are the bottleneck in the parts of the process that AI doesn't touch, so raw speed at one stage doesn't translate one-to-one into more finished work. This is worth sitting with if you've ever felt like you got a ton done in a burst of AI-assisted activity and then realized barely any of it actually shipped. That gap is real, and it's been measured. It isn't just you.
The vendor number, read with a raised eyebrow
GitHub ran its own randomized controlled trial on its Copilot tool and found a 55% faster task completion time, with 95 developers in the study. Notice right away that this study was run by the company selling the product. That doesn't make the number false, but it does mean you should hold it more loosely than the independent studies above it. The sample size is also small, and the confidence interval, the range the true number probably falls in, is wide. A vendor-run study with a small sample and a wide range is still evidence. It's just weaker evidence than a preregistered academic trial with 453 or 5,179 people in it.
The real speed numbers, ranked by rigor
Reported percentage gains, most trustworthy study design at top
The number that doesn't exist
Every study above was run on general working adults. Not one of them screened for ADHD, measured ADHD traits, or reported results separately for ADHD brains. Zero rigorous studies exist that measure AI speed gains for ADHD adults specifically. That gap is real, it's not small, and it's worth naming out loud instead of quietly borrowing someone else's number and hoping it applies to you.
So when you hear "AI makes ADHD people 3x faster" or any specific percentage tied to ADHD, that number is either lifted from one of the studies above and relabeled, or it's a guess dressed up as data. That doesn't mean AI doesn't help. It means nobody has measured how much, for your brain, in a controlled way. Anecdote is not the same thing as evidence, and you deserve to know the difference before you build a plan around a number that was never actually about you.
What this means for how you use it
The real numbers point somewhere useful even without an ADHD-specific study. AI seems to help beginners more than experts, it seems to speed up activity more reliably than it speeds up finished output, and even the best-designed study only measured one kind of task. So there's no clean percentage you can promise yourself. The useful move is to stop chasing a speed number altogether and focus on the one thing that's actually in your control, which is whether your next step got smaller.
That's the whole idea behind the Yap Method. You YAP your mess into AI without organizing it first. AI MAPs it into five piles: Tasks, Worries, Ideas, Feelings, Questions. You TAP one tiny action from those piles, something that takes two to ten minutes. Then you SNAP a short restart note for next time. We don't promise you'll be 40% or 55% faster. We promise the next step gets smaller, and if it doesn't, you say SMALLER or SHORTER until it does.
Take the whole system with you
The free 43-page ebook: the brain science in plain words, the four steps, the master prompt, and your first seven days.
Prefer it live? Doug teaches this in libraries and meetups — book a 90-minute class →