How to spot a credible online IQ test

Searching for an online IQ test turns up dozens of offers that all look the same: a handful of tasks, a progress bar, a number, and often a paywall just before the result. What separates them is not inside the test. It sits around the test, on the pages nobody reads, and you can check it in a few minutes without any specialist knowledge.

The nine questions below are written so that you can answer them yourself on any site. At the end we apply them to our own test, including where the answer is not flattering. A checklist that exempts the provider who wrote it is not a checklist.

Three of the nine are specifically Swiss, and if you have moved here they are likely to be the unfamiliar ones: the Impressum, the way a price and a refund rule have to be stated, and the line a test may not cross without becoming a regulated medical device. Each of the three is explained where it comes up.

The nine questions at a glance

1. Is a norm stated, and does it say what that norm rests on?

An IQ value is not a measurement like a temperature. It says where somebody stands compared with a reference group. Without that group the number means nothing: 112 compared with whom, measured on which people, collected when?

So look for the answer to one question: what is this value measured against? Good answers name a sample with a size, a period and a country, or they say openly that a model assumption is doing the work. Bad answers do not exist, because the question is never addressed at all.

One yardstick that applies in Switzerland: the classification of intelligence scores published in 2023 by Schulpsychologie Schweiz, the joint body of the cantonal school psychology services, places the band from 85 to 114 in a category it calls the normal variant, corresponding to percentile ranks 16 to 83. Anyone who receives a value from a provider should be able to locate it in a classification of that kind.

2. Does the value come with a confidence interval?

This is the question with the sharpest discriminating power, and it costs you five seconds.

Every test result carries a measurement error. On a short test with 30 tasks it runs to five to ten IQ points, depending on which tasks are used. For our own test we measured 6.4 points across 4‘000 simulated runs, which makes the 95 per cent interval roughly 25 points wide. Someone who scores 112 lies, with 95 per cent certainty, somewhere between about 100 and 124. That is not the weakness of one particular procedure. It is a property of short tests as such.

A provider who reports a value to the point without a range is doing one of two things. Either he does not know his measurement error, in which case he has not calculated it. Or he knows it and keeps quiet, because a range looks less impressive than a number. Neither speaks in his favour.

3. Is the calculation method disclosed?

“Scientifically validated” is not information. It becomes checkable only when a provider writes down which model he uses and how he gets from your answers to a value.

Concretely, something should say: how is the difficulty of a task determined? Is it taken into account that with four answer options you are right by guessing in a quarter of cases? What happens to somebody who has everything right, or everything wrong? How wide is the range, and where does it come from?

A procedure that is not described cannot be criticised, and that is usually the point of not describing it. The reverse also holds: a provider who discloses his model makes himself attackable, and that is a mark of quality. It means somebody was willing to put forward a claim that can be checked.

4. Is there an Impressum with a named responsible person?

This is the check an English speaker who is new to Switzerland will not think of, because there is no equivalent word in British or American practice.

An Impressum is the operator identification that a Swiss website has to carry. Under the Federal Act against Unfair Competition (UWG Art. 3 para. 1 lit. s), anyone who offers goods or services in electronic commerce must state their name and address clearly and completely and must provide a means of electronic contact. A contact form on its own does not satisfy that. The page is normally linked in the footer, and it is normally labelled with that exact German word, whatever language the rest of the site is in.

Check whether a natural person or a company with a registered address is named there, and whether those details match the domain and the payment processor. A provider without an Impressum who nevertheless takes money has decided not to be reachable for his own product. That is exactly what matters when a payment goes wrong, or when you want to ask what has been stored about you.

This is not a formality. It is the one point on this list that decides whether you can reach anybody at all.

5. Is an author named?

A brand name is not an author. Words like institute, centre or academy in the page title say nothing about who wrote the text and who checked it.

What you are looking for is a statement that names a responsibility: a person, an editorial team, a described way of working. And alongside it, just as important, a statement about where the numbers in the text come from. A guide text without a single source is, in case of doubt, copied.

Pay attention to professional titles as well. The titles Psychologin and Psychologe are protected in Switzerland under the Psychology Professions Act (PsyG, SR 935.81, Art. 45), and only those who meet the statutory requirements may use them. A website counts as an advertisement for this purpose. A provider who hands such titles out generously without being able to substantiate them is taking a risk, and that says something about his care in general.

6. Does the price stay visible before the test starts?

The most widespread pattern in this market: the test begins free of charge, you work for twenty to forty minutes, and only once the result has been calculated does a paywall appear. Sometimes with a bar hinting at how well you did.

It works because invested time shifts the decision. Somebody who has spent forty minutes pays more readily than he would have beforehand. It is not forbidden, but it is a deliberate decision against your overview.

So check before the first click: does it say anywhere what is free and what costs money? Is there a pricing page you can reach without starting the test? Is the price in Swiss francs, and is the amount stated the amount you actually pay, with nothing added at the checkout? Swiss price disclosure rules require the final price for a service, so anything that appears later is a bad sign in itself.

There is a second thing to look for here, and people who have moved to Switzerland from the EU or the UK regularly get it wrong. Swiss law gives you no statutory right of withdrawal for an online order. The Code of Obligations lists the situations in which an order can be withdrawn exhaustively (OR Art. 40b: doorstep sales, sales in the street or on public transport, promotional excursions, telephone sales), and online shopping is not among them. A provider who nonetheless publishes a copied European cancellation notice has not checked which law applies to him. At the same time, silence is not allowed either: a Swiss provider has to say whether a return or a refund is possible at all, even when the answer is no. Our own refund rule gives you 14 days, and we say plainly what that is: a voluntary commitment we have written down and are bound by, not an entitlement the law gives you.

7. Are numbers claimed that nobody can check?

“Over two million tests taken.” “Recommended by more than 500 professionals.” “The most used platform in the German-speaking world.”

Ask three questions of every such number: what exactly is being counted, over what period, and who established it? A started run, a completed run and a purchased report are three different things, and the largest of the three is usually the one on the page.

Claims like these are rarely verifiable, and they are worthless for your decision anyway. That many people have taken a test says nothing about whether it calculates correctly. The same scepticism applies to rating stars without a link to the source and to recommendations without a name. In Switzerland the burden even runs the other way: unfair competition law lets a court demand proof of an advertising claim from the advertiser, and missing proof counts as inaccuracy.

8. Does the update date jump to the day you visit?

A date on an article is meant to say when the text was last checked. Some sites instead set it automatically to today, so that the content looks fresh.

You can check that in two minutes. Open the same page again on a different day, ideally in a private window with no stored data. If it shows that day’s date again, it is generated automatically. A second route runs through a web archive: compare an older stored version with today’s. If nothing in the text has changed but the date has moved, you know where you stand.

The point is not the date itself but what stands behind it. A provider who fakes currency does so because currency counts for visibility. He is optimising for search engines, not for you.

9. Is a statement about health derived from the result?

This is the sharpest point on the list, and almost nobody names it.

A test that claims its result says something about your state of health, or that offers to detect, monitor or predict states of health, has left what a cognitive test can do. In Switzerland, such a claim would also change what the software legally is.

The reason sits in the Medical Devices Ordinance. Software becomes a medical device as soon as it serves a medical purpose according to its manufacturer. What decides the matter is not the technology but the intended purpose, and the provider sets that purpose himself, through his product description, his advertising and the wording of his reports. Anyone who promises on his own site to derive something medical from a questionnaire has declared himself to be operating a medical device, with every obligation that follows: a notified body, a quality management system, a clinical evaluation, market surveillance.

So the question is not only whether a provider is right. It is whether he is allowed to say what he says at all. A provider who crosses that line either holds an approval, which in this market is close to unheard of, or does not know what he is doing.

This is also why our own texts stay well clear of it. Our test makes no statement about your state of health and is not intended to detect, monitor or predict states of health. What it produces is an orientation value for reasoning, which is a different thing entirely.

How Kognita scores against its own list

Point 3 also means publishing numbers that do not flatter us. Across 4‘000 simulated runs, the correlation between the true and the estimated value is 0.898 and the root mean square error is about 6.5 IQ points. The theoretical ceiling for a test of this length, given its standard error, is 0.905. In other words, this is as far as thirty tasks go. It is not an unusually good result, and we write it down that way.

Where our own test only partly meets the criteria

We do not fully meet point 1, and that belongs here rather than in a footnote on the method page.

Our reference distribution is provisional. The value comes from the model assumption that the underlying ability is normally distributed in the population, and from a growing sample of our own. It does not come from a representative Swiss norming sample of the kind that stands behind a standardised procedure.

There is a second problem on top of that, and it affects every online test: whoever voluntarily fills in an intelligence test on the internet is not a random draw from the population. As long as that holds, the reference group remains an assumption rather than a count. What we can do is collect our own data, recalculate the task difficulties against them, and document every change with a date instead of quietly adjusting the numbers. How that is meant to work is also set out on the method page.

And a third point that none of the criteria catch: our test is taken unsupervised. We do not know who is sitting at the device, whether somebody was interrupted, or whether aids were involved. That is why Mensa Switzerland, the Swiss branch of the international high-IQ society, which admits people whose result falls in the top 2 per cent, accepts neither unsupervised nor online results. That applies explicitly to ours. Mensa runs its own supervised tests, currently only in Zug in central Switzerland, and sends the result by post within about three weeks. Our number is not a document you should submit anywhere.

What an online test cannot do in principle

Even a provider who satisfies all nine points measures a section of the whole. What is captured is reasoning: recognising rules in unfamiliar material and applying them. What is not captured, as separate values, is working memory and processing speed, and nothing at all is captured of knowledge, stamina, care and everything else that decides success at school and at work.

Then there is the day itself. Tiredness, distraction, screen size, nervousness and an earlier run shift a result by several points, on top of the measurement error of the procedure. Anybody who takes the same test twice should expect two different numbers.

A value from a short test therefore describes a performance on a particular day under particular circumstances. It does not describe a person, and it is not a basis for a decision about a school path or a job.

If you need a result that holds up

Where a school placement or a career decision is at stake, the route runs through a standardised, supervised procedure with a qualified professional. It takes one to two hours, covers more areas, and comes with an interpretation by a person who has actually met the person tested. In Switzerland, IQ tests may only be administered by people with a recognised psychology qualification, which is a real quality filter and also explains why such an appointment is not cheap.

For children and young people in compulsory schooling there is a route most families who have moved here do not know about. Education in Switzerland is a cantonal matter: there is no national school system, no national curriculum for entry to upper secondary school and no national authority you could write to. Each of the 26 cantons runs its own Schulpsychologischer Dienst, the cantonal school psychology service, and these services assess pupils free of charge. It is the cheapest and at the same time the most solid route, and it is rarely mentioned because nobody earns anything from it. Your child’s school or the cantonal education department will tell you which office is responsible for your municipality.

For a first orientation an online test is enough, provided it satisfies the nine points above. Our test is free, takes 25 minutes and produces a value with a range. What it is not is set out in the disclaimer: an orientation value from a short online test, and nothing more.