How Accurate Are Online IQ Tests?

Search for an IQ test online and you will find hundreds. Some are built with care. Many are entertainment wearing a lab coat. The same person can score 104 on one site and 141 on another in the same afternoon, and both numbers arrive with the same confident certificate.

So the honest answer to "how accurate are online IQ tests?" is: it depends almost entirely on how the test was built and scored, not on the fact that it runs in a browser. Below we explain the three ideas that decide whether any IQ score means something, what a professional assessment does that a website usually does not, and how to tell a useful online test from a flattering one.

Reliability, validity and norms: three different questions

People use "accurate" loosely. Psychologists split it into three separate questions, and a test can pass one while failing another.

Reliability: would you get the same score again?

Reliability is consistency. If you took the test twice, a few weeks apart, with no practice in between, would the scores agree? It is usually reported as a coefficient from 0 to 1. Professional IQ batteries aim for above .90 for the full-scale score.

Reliability lets you calculate the standard error of measurement (SEM), the typical wobble around your score. On the IQ scale (mean 100, standard deviation 15), SEM is 15 multiplied by the square root of one minus the reliability. A test with reliability .98 has an SEM of about 2 points. A test with reliability .80 has an SEM of about 7 points, so a 95% band runs roughly 13 points either side of your score. That is the difference between "about 118" and "somewhere between 105 and 131".

Validity: does it measure reasoning ability at all?

A bathroom scale that always reads 10 kg heavy is perfectly reliable and still wrong. Validity asks whether the test measures what it claims to, and whether its scores relate to other things they should relate to, such as other established ability tests or school results. We cover this in depth in our guide to what IQ test validity means, and our older overview of IQ test reliability and validity covers the same ground for professional tests.

Norms: compared with whom?

An IQ is not a raw score. It is a position in a distribution. "115" means "one standard deviation above the average of the comparison group". So the comparison group matters enormously. Professional tests are normed on large samples chosen to match a country's population by age, sex, education and region. A website that compares you with its own visitors, who are self-selected and often unusually keen on puzzles, is using a different yardstick, and many sites use no measured comparison group at all.

What a professional IQ test does

The Wechsler Adult Intelligence Scale (WAIS) is the most widely used adult IQ test. Its fourth edition, published in 2008, was normed on 2,200 US adults aged 16 to 90, sampled to match census proportions for age, sex, ethnicity, education and region. An independent review of that edition reported internal consistency of .97 to .98 for the full-scale IQ, which puts the SEM at roughly 2 to 3 points. A fifth edition followed in 2024.

The WAIS is given one to one by a trained psychologist. It covers verbal comprehension, visual-spatial and fluid reasoning, working memory and processing speed across many subtests. The administrator follows strict rules on timing, prompts and when to stop. Scores are compared with people of your own age, and the report gives a confidence interval, not just a single number.

That process is expensive and slow, which is why most people never take one. It is also why a professional assessment is the only route to a score you could use for a diagnosis, an educational assessment or a legal question. If you only want to see how you handle a matrix test, the Raven's Progressive Matrices page explains how that classic format works.

Why many online IQ tests give high scores

We are not aware of any systematic audit of commercial online IQ sites, so treat this section as a description of common design choices, not a measured statistic. But each choice below pushes scores in a predictable direction.

• No real norms. If a site never tested a representative sample, its conversion from correct answers to IQ is guesswork. Guesswork drifts upwards when the business benefits from happy customers.
• Flattery sells. A result of 132 with the word "gifted" next to it gets shared and paid for. A result of 98 does not. Some sites appear to shift every score into the top quarter.
• No time limit. Untimed tests let you search for answers, ask a friend or sit with one puzzle for twenty minutes. That changes what the test measures.
• Easy items and low ceilings. If most items are easy, a ceiling effect squashes everyone above average into the same band, and the top of the scale becomes meaningless.
• Repeated or leaked items. If the same questions are reused and their answers circulate online, scores rise for reasons that have nothing to do with reasoning.
• Score as bait. Some quizzes collect your email, then send a dramatic score with an upsell. The number is part of the marketing.

Retaking the same test also inflates scores honestly, through practice. A 2018 meta-analysis of retest studies found an average gain of about a third of a standard deviation, roughly 5 IQ points, from the first to the second sitting, rising to about half a standard deviation by the third. That is one reason a second attempt on any test usually looks better.

Can a test taken online be good? What research shows

Yes. The medium is not the problem. Several studies show that a carefully built test works on a screen.

• Computer versus paper. Williams and McCord (2006) gave the Standard Progressive Matrices on paper and on computer and found no significant differences in mean scores or spread between the two formats.
• A public-domain online battery. The International Cognitive Ability Resource (ICAR) was developed and validated online with 96,958 participants aged 14 to 90 (Condon and Revelle, 2014). Its items loaded strongly on a general factor, and corrected correlations with an established test, the Shipley-2, were above .8.
• Online data that matches lab norms. Hartshorne and Germine (2015) tested 48,537 people online and found age curves for different abilities that lined up with normative data from standardised IQ and memory tests.

The lesson: an online test can be a reasonable measure of reasoning if it uses a fixed, tested set of items, enforces timing, and is honest about how scores are converted. What it cannot easily do is give you a properly normed IQ, because that needs a representative sample, which costs real money to collect.

Clinical tests, good online tests and typical quizzes compared

Feature	Clinical test (e.g. WAIS)	Good online test	Typical online quiz
Who gives it	Trained psychologist, one to one	Self-administered, fixed rules	Self-administered, no rules
Items	Many subtests, verbal and non-verbal	Fixed, tested item bank	Often small, reused, easy
Timing	Strict, per item	Enforced overall limit	Often none
Norms	Large representative sample, by age	Published method, often not fully normed	Usually unknown
Typical precision	About 2 to 3 points (SEM)	Wider, often stated	Unknown
Score display	Report with confidence interval	Clear method, no flattery	Big number, "genius" labels
Usable for diagnosis	Yes	No	No
Cost	High	Low	Often "free" with an upsell

How to judge an online IQ test in two minutes

Before you trust a number, check for these signs. None guarantees a good test, but each one missing is a warning.

1. A published method. The site should explain what the items measure, how raw scores become IQ points and what the score does not mean. Look for a page like our methodology.
2. A fixed item bank with a clear length. You should know how many questions there are before you start.
3. A time limit that is enforced. Timed reasoning is closer to how ability tests are designed.
4. Plain limits. Honest sites say whether the score is age-adjusted and whether it is normed. If a site claims a clinical-grade result from a 10-minute quiz, be sceptical.
5. No score held hostage for your email. Paying for a report is a normal business model. Hiding a number until you hand over contact details, then sending something dramatic, is a red flag.
6. No genius flattery. About 2.3% of people score 130 or above on a properly normed test. If a site tells most visitors they are gifted, its scale is broken. Our IQ percentile calculator shows how rare each score really is.

For a side-by-side look at well-known options, see our guide to the best IQ tests.

Test-day factors that move your score

Even a well-built test measures you on one day. Several things shift the result.

Sleep

A meta-analysis of 70 studies by Lim and Dinges (2010) found that short-term sleep deprivation hits simple attention hardest, with large effects on lapses, while the effect on reasoning accuracy was small and not statistically significant. Timed tests still lean on attention, so a bad night can cost you, even if your reasoning itself is intact.

Effort and motivation

A widely cited 2011 meta-analysis suggested that paying people to try harder raised IQ scores by a large amount. That paper was retracted in 2025. A newer set of six studies with 4,208 adults by Bates and Gignac (2022) found that monetary incentives had a much smaller effect, roughly 2.5 IQ points, and that self-reported effort was only modestly related to scores. Trying properly matters, but low effort is unlikely to explain a 20-point gap.

Test anxiety

A 30-year meta-analysis of 238 studies by von der Embse and colleagues (2018) found a consistent negative link between test anxiety and performance on standardised tests, exams and grades. If tests make you very nervous, your score may understate what you can do.

Environment and practice

Distractions, a phone screen instead of a laptop, and having seen similar puzzles before all change results. Take any test once, somewhere quiet, without help, and treat a retake as practice rather than a new measurement.

How accurate is the LifeScore IQ test?

We should be as clear about our own test as we are about everyone else's. The LifeScore IQ test is 25 visual reasoning questions in about 12 minutes, in five groups of five: pattern recognition, spatial reasoning, logical sequences, analogical reasoning and rule application. It uses a fixed item bank and an enforced time limit, and there is no sign-up to start.

Your score is a simplified conversion of your answers onto the IQ scale (mean 100, standard deviation 15). It is not normed against a representative population sample and it is not age-adjusted. It is not a clinical assessment, it is not given by a psychologist and it cannot diagnose anything. Twenty-five items also means a wider margin of error than a full battery. Treat the result as an estimate of your visual reasoning on the day, not a fixed fact about you. The full report is a paid unlock after the last question, and our methodology page explains the conversion in detail.

Questions people ask

Are online IQ tests accurate?

Some give a reasonable estimate of reasoning ability; many do not. The best ones use fixed, tested items, enforce a time limit and explain their scoring. None replaces a professionally administered, normed test such as the WAIS.

Why did I score higher online than on a real test?

Common reasons are inflated conversion tables, no time limit, easy items, having seen similar puzzles before, and being compared with an unclear group. Practice alone adds around 5 points on a second attempt at the same test.

Can an online IQ test be used for Mensa?

Generally no. Mensa accepts its own supervised tests or approved professional assessments. An online result is useful as practice, not as evidence.

How many questions does a reliable IQ estimate need?

More items usually means higher reliability. Professional batteries use dozens of items across several subtests. A short test can still be informative, but its margin of error is wider, and an honest site should say so.

Is a free online IQ quiz worth taking?

It can be fun practice. Just judge the result by the same checks: published method, time limit, fixed items and no flattery. If a quiz tells you that you are a genius after ten easy questions, the number says more about the quiz than about you.