Two questions hide inside the word accurate, and confusing them causes most of the argument on this subject. The first is whether an assessment measures consistently — whether the same person sitting it twice gets a similar answer. The second is whether what it measures corresponds to anything meaningful outside the test room. A test can be excellent on the first count and questionable on the second, and intelligence testing sits in exactly that awkward position.
By the standards of psychological measurement, the major batteries are impressively consistent. Reliability coefficients for full-scale composites on well-constructed instruments generally sit above 0.90, which means the great majority of variation between people reflects stable differences rather than measurement noise.
That figure is high, but it is not perfect, and the residual matters. A composite is best read as a band rather than a point. Publishers report confidence intervals for exactly this reason: a reported score of 108 typically means the underlying value probably lies somewhere between roughly 103 and 113, and treating 108 as an exact quantity misrepresents what was measured.
Test-retest studies confirm the picture. Sit the same battery twice a few months apart and the two results will usually land within a handful of points of each other, but rarely on the same number. Movement of five or six points is ordinary variation. Anyone treating a three-point difference as evidence of change is reading noise as signal.
Individual subtests are less reliable than composites, which is why a single index score should never be quoted in isolation. Reliability improves with length: the more items sampled, the more the random component averages out. This is the mathematical reason a twelve-question web quiz cannot deliver what a ninety-minute battery does, regardless of how the quiz is presented.
The second question is harder and more contested. Assessment scores do correlate with outcomes that people care about — academic attainment, training success, and job performance in complex roles show consistent positive relationships across a very large body of research.
The correlations are real but moderate. They describe tendencies across large groups, not predictions about individuals. Knowing a person's score narrows the range of likely outcomes slightly and tells you very little about what any particular person will do. Plenty of people with high scores accomplish little, and plenty with modest scores build significant careers, because persistence, judgement, opportunity and social skill all contribute independently.
There is also the question of what the composite excludes. Every item on a standard battery has one defensible answer, which means divergent thinking is essentially invisible to the instrument. Practical competence, emotional insight and domain expertise built over years are similarly unmeasured. The existence of broader models of ability reflects long-standing dissatisfaction with a single composite standing in for the whole.
The phrase official IQ test appears constantly in search results and is almost always misleading. No government body certifies intelligence tests, and no organisation holds a monopoly on the concept. There is no single official instrument.
What does exist is a meaningful distinction between supervised and unsupervised assessment. A result carries formal standing when a qualified professional administered a recognised, properly normed instrument under controlled conditions and documented the session. That result can support an educational assessment, an occupational decision or an application to an organisation with a testing requirement.
A result produced at home, in a browser, with nobody present, cannot do any of those things — not because the items are necessarily poor, but because the conditions are unverifiable. Nobody can confirm that you worked alone, without pauses, without assistance, on a first attempt. That is the entire difference, and it is not a formality. The admission requirements used by high-IQ societies illustrate the point clearly: they accept a range of instruments but insist without exception on supervision.
Several factors move scores around without any change in underlying ability, and most of them are controllable.
The last point deserves emphasis because it is invisible to the person sitting the test. Norms decay. Any instrument that has not been restandardised in twenty years is quietly generous, and the participant has no way of knowing.
Web-based assessment is not uniformly bad; it is uniformly unverifiable. Within that constraint, some are considerably better built than others, and the signals are visible before you begin.
Look for a stated number of items and a stated time limit, both disclosed up front. Look for a description of the norm sample — who it was drawn from and how large it is. Look for a reported confidence interval rather than a bare figure, and for subtest breakdown rather than a single composite. Look for the pricing model stated before you start rather than revealed after.
Warning signs are equally clear: no time limit at all, results that arrive impossibly fast, extravagant claims about precision, scores that seem flattering to everyone, and any suggestion that the outcome carries formal recognition. A fair free iq test is honest about being an estimate rather than presenting itself as something it is not. The broader assessment of what free online instruments can deliver covers the trade-offs in more detail.
A related question is whether a score measured once continues to describe the same person decades later. The evidence here is better than for most psychological measures, and more nuanced than the popular framing suggests.
Long-running follow-up studies that retested the same individuals across many years find substantial stability from late childhood onwards. Rank order within a group tends to hold: people who scored above average as adolescents mostly remain above average later. That is a real finding and it is why the measure is taken seriously at all.
But stability of rank is not the same as constancy of capacity. Different components move in different directions with age. Accumulated verbal knowledge and reasoning built on experience tend to hold up or improve well into later life, while processing speed and working memory decline steadily from early adulthood. A composite can stay flat while its underlying parts diverge sharply, which is why age-matched norms are essential and why comparing a figure obtained at twenty with one obtained at sixty requires care.
Circumstances also shift results. Extended education, serious illness, sustained sleep deprivation and significant life stress all move scores measurably. The number is stable enough to be useful and mobile enough that treating it as a permanent characteristic misreads the evidence.
Beyond reliability and validity sits a third issue that has driven reform for a century: whether an instrument measures the same thing equally well across different groups.
Early tests were plainly compromised on this count, relying on cultural knowledge that varied enormously with background and using results to support conclusions the data could not carry. Modern development takes the problem seriously. Items are screened statistically for differential functioning, culture-reduced formats have been developed to lower the linguistic and cultural load, and norm samples are stratified to reflect the population.
Improvement is real but incomplete. Educational exposure influences performance on any instrument, and no test can fully separate reasoning capacity from the conditions in which it developed. Any interpretation that ignores a person's circumstances is doing something the instrument does not support.
Treat a supervised composite as a reasonably reliable estimate of reasoning efficiency, expressed as a range rather than a point, useful alongside other information and dangerous on its own. Treat an unsupervised online result as a rough indication with wider error bars, useful for tracking your own performance over time under consistent conditions and useless as a claim about yourself.
The practical value of both improves when the profile is read rather than the headline. Uneven index scores identify specific strengths and difficulties, which is actionable; a single figure identifies nothing you can work on. That is also why deliberate problem-solving technique and understanding of different reasoning types matter more in practice than the composite ever will. Anyone about to sit an assessment will get a cleaner result by following the straightforward conditions set out in the material on how to test your IQ.
Continue with the rest of the series on assessment and reasoning: