The honest answer is: less accurate than a supervised assessment, more accurate than nothing, and wildly variable between sites. The difference between a useful online test and a decorated quiz has little to do with how the site looks and a lot to do with what it is willing to tell you.
What limits accuracy online
- Nobody is supervising: interruptions, second attempts and looking things up are all possible.
- Conditions are unknown — sleep, noise, screen size, whether you were rushing.
- Most online banks are small, so familiarity from a previous attempt inflates later scores.
- Self-selected visitors are not a representative population sample, so norms built purely on site traffic drift.
The self-selection problem is easy to underestimate. Consider what would happen if a gym scored "national fitness" using only the people who walk through its door on a January evening. Nobody lied, nothing was faked, and the answer is still wrong, because the sample is people who chose to show up. A site whose norms come only from its own visitors — usually people who like puzzles enough to search for a test — has the same problem baked in.
Warning signs a test is theatre
- A single confident number with no margin of error.
- A score you only see after paying, or after handing over an email.
- Suspiciously flattering results — sites that tell nearly everyone they are gifted are optimising for sharing, not measurement.
- Claims of accreditation, or borrowed authority from Mensa, universities or test publishers.
- No explanation of how the score was produced, and no way to see your own answers.
Two patterns are worth naming because you have probably met both. The first is the paywalled number: you answer thirty questions, a progress bar fills, and the result is hidden behind a card form. The test has been designed around the moment of maximum curiosity, not around measurement. The second is the flattery engine — screenshots of "your IQ is 137" spread on social media for the same reason quiz results about which historical figure you are spread. If almost everyone who posts a result is well above average, the scoring is a marketing decision.
Signals a test is being straight with you
- Results reported as a range with an explicit margin of error.
- A stated methodology, including what is inherited from published research and what is not.
- Plain limitations: what the result cannot be used for.
- Your result visible without payment, with paid options adding precision or depth rather than unlocking the number itself.
- Some fraud handling — runs answered impossibly fast, or with the same option every time, should be thrown out rather than scored.
A quick test of the test: answer the same option for every question, as fast as you can click, and see what comes back. A site that hands that run a respectable score is not measuring anything. A site that refuses to score it, and tells you why, is at least doing the basic hygiene.
So how close can an online result get?
Under decent conditions, a well-calibrated online reasoning test can usually place you in the right broad band — the right side of average, roughly which quarter of the distribution. It should not be expected to pin you to a specific number, and any site that claims otherwise is claiming precision it cannot have. If you need a result for a diagnosis or any formal decision, an online test is the wrong instrument entirely; see a licensed psychologist.
In practice that means treating the output the way you treat a home blood-pressure cuff rather than a hospital workup: good enough to tell you roughly where you sit and whether something is worth following up, not good enough to decide anything serious on its own.