By Eric, BlinkBench Founder · Last reviewed: July 2026
You take a reaction time test, you get 247 ms, and underneath it the site says you are faster than 63% of people. That second number is doing all the emotional work. It is also, on most sites, unsourced.
Not necessarily fabricated — a site with a million visitors really can rank you against its own visitors. The problem is that ranking you against them means something quite different from what it appears to mean, and almost nobody explains the difference. Four questions will usually do it.
1. Which people?
A percentile is meaningless without a described population. “People” is not a population. The people who take an online reaction time test are people who went looking for an online reaction time test: younger than average, more likely to be gamers, and — critically — often taking it more than once, having already read the instructions and warmed up.
Reaction time also moves with age at a measured rate. Woods and colleagues found about 0.55 ms per year across a sample of 1469 adults aged 18 to 65. So a percentile calculated over an unknown age mix is comparing you against a moving target nobody has described.
2. On what hardware?
This is the one that ought to end the conversation. Everyone in that visitor sample was using their own screen, their own mouse, their own browser — and those differ from each other by more than the thing being measured varies between people.
The display alone accounts for about 6.3 ms between a 60 Hz screen and a 240 Hz one (the arithmetic is here). At 0.55 ms per year, that is roughly 11 years of aging, available for the price of a monitor. Input devices, USB polling and OS event handling add more on top, and none of it is visible in the score.
A percentile computed across mixed hardware is therefore partly a ranking of equipment. It is presented as a fact about you.
3. Filtered how?
Every reaction time test has to discard something. A click before the stimulus appeared is not a reaction. A response forty seconds later is somebody who went to make tea. The question is not whether trials get thrown out — they must be — but whether you are told.
Silent filtering is where a comparison quietly stops being a comparison: if the reference sample dropped its slow trials on one rule and your run is scored on another, the two numbers were never on the same scale. This is why the tests here state what was discarded and why, on the page, in the run you just did.
4. Measuring the same task?
“Reaction time” covers several genuinely different tasks. Responding to a single expected signal is not the same as choosing between two signals, which is not the same as hitting a moving target. They produce different numbers, in that order, reliably. A percentile borrowed from one and printed under another is simply the wrong number.
What a defensible comparison looks like
There are real published reference ranges. The Woods study is one: 1469 adults, a stated age range, a described task, published means and standard deviations — and unusually, the authors measured their own equipment with external hardware and reported 17.8 ms of it, which is how their mean of 231 ms becomes 213 ms once the rig is subtracted.
That is enough to sit beside your score as context. It is still not enough to turn into a percentile, because their apparatus is not your apparatus and their sample was not recruited the way you arrived here. So the reaction time test shows you the range and lets you place yourself in it, rather than computing a rank that would imply a precision neither of us has.
And where no comparable study exists at all, the honest output is nothing. The visual memory test cites no reference range, because the norms for that task are locked in a printed manual and wouldn't transfer to a browser if they weren't. What it prints instead is the exact probability of passing each round by guessing — a number that can be computed rather than claimed. The chimp test reaches the same dead end by a different route, and says so.
A missing percentile is not a missing feature. It is the site declining to make one up, which is the whole point of the place.