How accurate are online IQ tests?

This page separates the medium from the product: what the research says about testing over the internet, why most online IQ tests are nonetheless bad, the seven things that move a score by more than people expect, and what no online test can do at any price — including this one.

Three different things people mean by "accurate"

Almost every argument about online tests collapses because the word is doing three jobs at once. Psychometricians keep them apart, and once you do, the answer to the question becomes much clearer.

Reliability: does it give the same answer twice?

If the same person sits the same test on two ordinary days, do they get roughly the same score? A test can be highly reliable and still measure the wrong thing entirely — a bathroom scale that always reads eight kilos heavy is perfectly reliable.

Validity: is it measuring what it claims?

Does the score reflect general reasoning ability, or does it reflect how much trivia you happen to know, how fast you read English, or how familiar you are with puzzle conventions? This is where mixed-format tests fail badly, because each question type drags in a different contaminant.

Norming: is the comparison group real?

An IQ score is meaningless in isolation. It is a statement about where you sit relative to a comparison population. Professional tests are standardised on demographically stratified samples recruited to represent the population. Most online tests compare you against whoever else happened to visit the website — a self-selected group of people interested enough in IQ to seek out a test, which is not the general public.

What the research says about testing over the internet

Here is the part that surprises people who assume online means unserious. The medium itself has been checked carefully, and it holds up.

A 2012 study by Germine and colleagues directly compared cognitive and perceptual testing conducted over the web against the same tasks run in a laboratory, and found comparable performance across the two settings.

The stronger demonstration came in 2015. Hartshorne and Germine collected data from 48,537 online participants and compared their results against reanalysed norms from the Wechsler Adult Intelligence Scale and Wechsler Memory Scale — demographically stratified, professionally administered, paper-and-pencil standardisation samples collected two decades earlier. The age-of-peak-performance patterns converged across both. That convergence rules out the obvious objections: it was not an artefact of older participants being less comfortable with computers, nor of the internet sample being unrepresentative in some way that distorted the results.

So the medium is not the problem. A well-constructed test delivered over the internet can produce data good enough to publish in a peer-reviewed psychology journal. The problem is that almost no commercial online IQ test is a well-constructed test.

Why most online IQ tests inflate your score

The incentive structure explains nearly everything. A test that tells you your IQ is 127 gets shared, screenshotted, and paid for. A test that tells you 103 gets closed. Nobody has to be consciously dishonest for the whole market to drift upward — the tests that flatter simply survive.

Three mechanisms do the work:

• No real norm sample. If a test has never been standardised on a representative population, the number it prints is not an IQ score in any technical sense. It is a scaled version of how many questions you got right, dressed up in the vocabulary of psychometrics.
• Self-selected norms. Tests that do norm against their own visitors face a subtler version of the same problem in reverse, and honest ones disclose it.
• Easy items with a hard-looking presentation. A test whose questions are mostly solvable produces a compressed distribution near the top, which reads as "most people who take this are above average" — a mathematical impossibility that a surprising number of sites publish without noticing.

Add the paywall pattern — free test, price revealed only once you are emotionally invested in seeing your result — and you have the standard commercial model. The score is the bait.

Seven things that move your score more than you expect

Even a well-built test measures you on a particular day, in particular conditions. These are the factors that move results, roughly in order of how much damage they do.

The practical consequence is worth stating plainly: a single result from one online sitting is weak evidence. Not worthless — weak. If the number surprises you in either direction, that is a reason to sit it again properly, weeks later, rather than a reason to update your self-image.

What professional tests do that online tests do not

The gap is real and it is worth understanding rather than hand-waving at.

A supervised assessment is a battery, not a test. It samples verbal comprehension, perceptual reasoning, working memory and processing speed separately, then combines them — which is how a psychologist detects that someone is strong in one domain and weak in another, a pattern a single-format test cannot see at all. It is administered by a trained person who can tell the difference between not knowing an answer and not understanding the instruction. And it reports a score with a confidence interval attached, because no serious psychometrician reports a bare number.

That last point is the tell. Ask yourself how many online tests you have seen that report a range rather than a single figure.

What no online test can do, at any price

If you need a score for a formal purpose, the only route is an educational or clinical psychologist. Anything else wastes your money and your time.

How to tell a reasonable online test from a flattering one

Five questions, answerable in about a minute before you start.

1. Is it one question format or a grab-bag? Anagrams, arithmetic, riddles and pattern puzzles in the same test cannot produce a coherent score, because each measures something different.
2. Is it timed? Untimed reasoning tests measure persistence as much as ability.
3. Does it tell you what it compares you against? A site that will not say what its norm group is has told you something important.
4. Is the price visible before you start? A price revealed at the results screen is a design decision about your emotional state, not an oversight.
5. Does it claim more than it can deliver? Any test advertising itself as certified, clinically validated, or Mensa-accepted is misrepresenting itself.

What this test can and cannot tell you

It would be a poor page on this subject that exempted the site it sits on, so here is the same scrutiny applied here.

What it does reasonably well. It uses one format throughout — thirty matrix reasoning items, the task family most closely associated with the fluid, pattern-based core of general ability. It is timed. It is entirely non-verbal, so it does not quietly measure your English vocabulary or your general knowledge. Difficulty rises by layering rules rather than adding visual clutter. The price is stated before question one, and the test is free to take whether or not you buy the report.

What it cannot do. It is a single-format test, so it measures one thing and is blind to the rest of a cognitive profile. It is unsupervised, so it cannot verify the conditions you took it under. Scores are normed against this site's own test-taker population rather than a demographically stratified national sample, which means the comparison group is people who chose to take an IQ test online — not the general public. And it is thirty items, which is enough to place you within a broad region of the scale and not enough to resolve fine distinctions, particularly at the extremes.

Read honestly, a result here is a decent estimate of where your fluid reasoning sits relative to other people who took the same test. It is not a clinical measurement, and this site does not claim it is.

Getting the most accurate reading you can

If you are going to take one, these are worth more than anything else you can control.

• Sit it when you are rested and alert, not at the end of a long day.
• Use the largest screen available and close everything else.
• Give it the full uninterrupted block. A timed test taken in three chunks is not a timed test.
• Answer every question. Guessing carries no penalty here, so an unanswered item is a wasted one.
• Pace the opening. Early items should take seconds; the discrimination between scores happens at the end, and you want the clock there.
• If you retest, leave weeks between attempts. An immediate retake measures your memory of the items.

Frequently asked questions