Straight answer

Are online IQ tests accurate?

Accurate enough to place you in a broad band. Nowhere near accurate enough for anything official. The gap between those two sentences is where most of the marketing in this niche lives.

The short version

A well-built online reasoning test with thirty or more items can tell you, with reasonable confidence, which quarter of the distribution you fall into. That is a real piece of information and it is worth ten minutes.

It cannot tell you that your IQ is 123. It cannot produce a result any institution will accept. And it cannot be fixed by writing better questions, because the limits are in the administration rather than the items.

The four structural limits

1. Nobody is watching. A psychologist administering a clinical test controls the environment, reads standardised instructions, times what needs timing and records one attempt. A web page controls none of it. The person answering may be tired, interrupted, on their fourth attempt, or helping someone else with theirs.

2. The norm group is self-selected. Clinical tests are normed on samples stratified to match census demographics on age, sex, education and region. A web test is normed, at best, on people who searched for an IQ test and finished one. That group is younger, more online and more invested in the result than the population is, and percentiles against it are not percentiles against everyone.

3. Short tests are noisy. The standard error of measurement falls roughly with the square root of test length. This is arithmetic, not an opinion about any particular site. Ten items supports a quarter of the distribution; thirty narrows it usefully; getting the error down to a few points takes the hour or more that a clinical battery runs.

4. Unlimited retakes. Practice effects on reasoning formats are real and worth several points, concentrated on the first repeat. Anyone can take five tests and keep the best, and a personal-best score collected that way measures persistence.

Five signals that tell you what you are dealing with

You cannot audit an item bank from outside. You can read the site, and five things tell you most of what you need to know.

1. Does it report an error band? A bare number with no interval is a claim to precision that no unsupervised test can support. Its absence usually means the site either does not know about measurement error or decided a clean number converts better.
2. Does it describe its norm group? This is the single most informative section on any IQ site, and the one most often missing. Silence is the answer.
3. Is the score itself free? A result held behind a payment creates an incentive towards flattering numbers, because a disappointing score does not convert. Free-to-take and paid-to-see is a paid test with a long preamble.
4. Does it guarantee accuracy? Nobody can guarantee this. A guarantee is a marketing sentence and its presence is informative in itself.
5. Does it claim Mensa qualification? No online test qualifies anyone for Mensa. A site claiming otherwise is either uninformed or dishonest, and neither is reassuring.

What a free test is genuinely good for

Four things, all of them real.

A first read. If you have never taken a reasoning test, a free one tells you roughly where you sit without an appointment or a fee.

Format familiarisation. If you have a supervised test coming up and have never seen a matrix item, part of what that test will measure is your unfamiliarity. Removing that is fair and sensible.

Deciding whether to go further. If a free test puts you near a threshold that matters to you, that is a reasonable basis for booking a supervised one.

Curiosity, which is a perfectly good reason. Most people taking these tests are not making a decision. They want to know. There is nothing wrong with that as long as the site is honest about what it is handing over.

What we do about it

Our own test is ten items, which is shorter than the tests above and therefore noisier still. We deal with that in three ways, all of which cost us something in how impressive the result looks.

We report a band rather than a number, because ten items cannot support a number. We map raw scores onto the theoretical normal curve rather than onto our own visitors, because our visitors are not a norm sample and pretending otherwise would be the more flattering option. And we say on the result screen that the band is what the test can support, rather than putting that in a footer.

The full description is on the methodology page, including what we would need to do to claim more than we do.

When you need the real thing

If a result has consequences, an individually administered test is the only route. Specifically:

• School placement, gifted identification or special educational provision.
• A workplace or examination accommodation request.
• Any legal or forensic context.
• Diagnosis of a learning disability or intellectual disability.
• Neuropsychological assessment after injury or illness.

In all of these the instrument will be a WAIS, a WISC, a Stanford–Binet or a comparable battery, administered by a licensed psychologist, and the report will include a confidence interval, a profile across indices and an interpretation. That is what a result that counts looks like, and no website produces one at any price.