BRGHT IQ test: adaptive design, self-anchored norms

0 The Short Version

BRGHT is an adaptive 40 item test taken in 30 minutes, and its central limitation is not the design but the reference group: your score is your standing among BRGHT's own test takers, stated in its own documentation. The method page says it plainly. A raw score is compared to the entire population of test-takers to produce an IQ with a median of 100 and a standard deviation of 15. That is an honest description of a self-anchored scale, and it means the number is a rank inside a self-selected internet sample rather than a standing in the general adult population.

The good parts are real and should be said first. Adaptive administration is a better design than a fixed form of the same length, the item pool is calibrated from a large response corpus, the per item record shown after the test is more transparent than almost anything else in the consumer market, and BRGHT publishes an open offer of anonymised data to academic researchers. Those are choices a serious measurement team makes. Very few of its competitors made any of them.

The gaps are equally specific. There is no published reliability coefficient, no standard error of measurement, no factor analysis, and no validity study against any external instrument that we could locate on September 5, 2026. The country ranking on the site places the French Southern Territories third in the world with no cell size attached, and the official administration of that territory states that it has no permanent population. And the terms and conditions, last updated May 2023, contain no price, no renewal clause, no cancellation route, and no refund policy at all.

1 Who Operates BRGHT

The first question about any paid test is who you would be dealing with if something went wrong, and BRGHT answers it. Its terms and conditions name Assessly BV, a Dutch company registered under KVK number 76121224 at Kerkstraat 310, 1017 HC, Amsterdam, with the agreement governed by Dutch law. A named entity with a registration number in a jurisdiction that maintains a public company register is more than a large part of this market offers, and it deserves credit before anything critical follows.

Read the definitions section and something else becomes visible. The contract is written for a business buyer. It defines an Account Holder as the organization or individual named as the client on the Sign-up Form, a Candidate as a person who accesses the service to take a Test Session, and a Paid Plan as the paid service you subscribe to. It grants the company the right to use your company name and logo for marketing. That is the vocabulary of a pre-employment testing platform, and the consumer product at brght.org sits inside it.

The consequence is not cosmetic. A document drafted around Account Holders and Sign-up Forms does not contain the clauses a consumer needs, and in BRGHT's case it does not. The words refund, cancel, subscription, and trial do not appear anywhere in the terms, which were last updated in May 2023. Section 9 caps the company's total liability at the fees you paid in the six months preceding a claim.

One sentence in Section 4 deserves quoting because it is more candid than most competitors manage. The service is provided as-is, and the document states that testing is a statistical process with inherent randomness and that results may not accurately reflect reality. Providers rarely write that down. It is also, read carefully, the closest thing on the site to a statement about measurement error, which is a gap examined further below and covered in general terms on the reliability and validity page.

2 What the Test Actually Consists Of

Start with the concrete, because the numbers are published and they constrain everything else. The BRGHT start page, read on September 5, 2026, describes a test of 40 questions to be finished in 30 minutes, covering logical, numerical, and spatial reasoning, drawn from a pool it describes as over 5000 questions. Questions must be answered in sequence and skipping is not possible. Takers can retake the test with a new set of questions as often as they like.

The clock is unusual and worth understanding, because it changes what the test measures. The start page states that you get 45 seconds for the first question and that each next one adds another 45 seconds. That is a cumulative time bank rather than a per item limit: 40 items at 45 seconds each is exactly the 30 minute total, and time saved early is available later. The page says so directly, advising that answering early questions fast gives you more time for the harder ones later.

Two things follow from that design. The first is that BRGHT is a speeded test in effect, even though it is presented as a reasoning test. Anyone who works carefully will reach fewer items, and reaching fewer items lowers the score under the scoring rule described in the next section but one. The second is that speed and reasoning become entangled in the single number, which is exactly the confound that multi domain batteries separate by reporting processing speed apart from fluid reasoning. Why those two are worth keeping distinct is set out on the processing speed page and the page on measuring fluid reasoning.

The three named content areas map onto familiar ground. Logical, numerical, and spatial reasoning correspond roughly to induction, quantitative reasoning, and visualization, which sit inside fluid reasoning, quantitative knowledge, and visual processing in the Cattell-Horn-Carroll framework. What the form does not sample is verbal comprehension, working memory, or processing speed as separate measured constructs, which means the resulting figure is a reasoning composite rather than a full scale score in the sense a Wechsler battery uses the term. The domain map is on the cognitive domains page and the framework itself on the CHC model page.

3 The Adaptive Design Is a Genuine Advantage

Adaptive administration is the best thing about BRGHT, and pretending otherwise would make this review useless. A fixed form has to spend items establishing what it already knows. If you are well above average, the easy third of a fixed test tells the scorer almost nothing it did not learn from your first few answers, and those items are wasted. An adaptive test spends its items where they discriminate, which is why a well built adaptive form can reach a given precision in fewer items than a fixed one.

The principle is not controversial and it is not new. Computerized adaptive testing has been standard in high stakes assessment for decades, and the reason is exactly this efficiency argument. A consumer IQ product that implements adaptive administration at all is making a design choice most of its competitors did not, and BRGHT states that it was the first adaptive IQ test of this kind.

The item pool behind it is also real work. The research page states that the pool consists of approximately 3,000 questions in 10 categories, answered more than 75,000 times per day by around 2,500 unique daily users. The same page names the problem that motivated the pool, and names it well: public domain item sets are small, repeat exposure threatens validity, and multiple research groups administering the same items to the same online participants makes it worse. That is a sophisticated observation about item security, and it is correct.

Credit also belongs to what happens after the test. BRGHT's method page commits to sharing the metadata of every question you answered, including the average IQ of users who answered it, the average proportion correct by IQ range, and the median time taken. Publishing per item statistics to the person who just sat the item is a level of transparency that no major clinical publisher offers and almost no consumer competitor attempts.

4 How the Score Is Actually Computed

The scoring rule is published in full, and reading it changes what the word adaptive means here. The method page describes it in four steps. Questions are divided into 20 difficulty levels. All users begin with a level 1 question. For each correct answer the difficulty level increases by 1, and for each incorrect answer it decreases by 1. Points earned for a correct answer equal the difficulty level of that question, so a correct level 12 item earns 12 points, and incorrect answers earn nothing. The points are summed into a raw score.

That is a fixed step staircase with a weighted count as the score. It is a defensible procedure and it is easy to explain, which has value. It is not what a computerized adaptive test normally does, which is select each item to maximize the information it yields about the current ability estimate and then estimate ability by maximum likelihood or a Bayesian equivalent from the full response pattern. The distinction matters because the staircase discards information the response pattern contains.

Now run the arithmetic the rule implies. Starting at level 1 and climbing one level per correct answer, reaching level 20 takes 19 consecutive correct answers, which consumes 19 of the 40 items. For a taker whose true ability sits near the top of the pool, roughly half the test is spent climbing through items they will almost certainly pass, and those items carry very little information about where they actually stand. This is the specific inefficiency that maximum information item selection exists to remove, and the fixed step rule reintroduces it.

The scoring rule also explains the advice on the start page to try to answer all questions for the highest score. Under a sum of level points, an unanswered item is worth zero regardless of whether you would have got it right, so volume is rewarded directly. A person who reaches 28 items carefully cannot outscore an equally able person who reaches all 40, and the gap between them is a speed difference reported as an ability difference. How time limits change what a reasoning test measures is discussed on the page on time limits.

5 What the 2PL Claim Covers and What It Does Not

The item response theory claim is true of the calibration and not of the scoring, and the site does not draw that line. BRGHT's homepage states that it harnesses the advanced 2PL IRT model, meaning the two parameter logistic model, into the first adaptive IQ test, and says the methodology offers unparalleled accuracy by considering both the difficulty of questions and the likelihood of a correct response. The method page is more precise: the 20 difficulty levels are based on a Rasch and a 2PL item response theory model derived from the results of over 500,000 individual test-takers.

Read those two statements together and the architecture becomes clear. IRT was used to estimate item parameters and to bin items into 20 ordered difficulty levels. The score you receive is then a sum of level points from a staircase, not an IRT ability estimate. Both models are named, which is unusual candor, and the two are not interchangeable: the Rasch model constrains all items to a common discrimination while the 2PL estimates a discrimination parameter per item, and the site does not say which model produced the operational level assignments.

Why this matters in practice is the discrimination parameter. The whole point of the 2PL over the Rasch model is that items differ in how sharply they separate people, and a scoring rule that awards points equal to the difficulty level throws that difference away. Two level 15 items with very different discriminations contribute identically to the raw score. The model was used to sort items, then set aside when the number was produced.

None of this makes the score meaningless. A weighted sum of difficulty levels correlates strongly with an IRT ability estimate in most realistic conditions, and for a consumer product the simplicity may be worth the loss. It does mean that a reader who sees the phrase 2PL item response theory and assumes the reported figure is a calibrated latent ability estimate with a standard error attached has assumed something the published method does not describe. What a properly derived composite looks like, with its structure evidence and its per composite reliability, is set out in the ACIS technical manual.

6 Precision: What the Published Interval Implies

BRGHT does publish an uncertainty range on its shared result pages, which is better practice than most of the market, and the width of that range is the most informative number on the site. A publicly shared BRGHT result page in the sharing format, read on September 5, 2026, reports an IQ of 111 and states that the person's IQ is between 98 and 124. It also states that the person scored better than 44.00 to 94.00 percent of the people who ever finished this test. The percentile band matches the score band under a normal distribution with a mean of 100 and a standard deviation of 15, so the arithmetic behind it is internally consistent.

Now read the width. The band runs 13 points either side of the estimate. If it is intended as a 95 percent interval, the implied standard error of measurement is about 6.6 IQ points, which on a scale with a standard deviation of 15 corresponds to a reliability of about .81. If it is intended as a 90 percent interval, the implied standard error is about 7.9 points and the implied reliability is about .72. BRGHT does not state the coverage of the interval, so both readings are available and both are ours rather than the provider's.

For scale, the ACIS technical manual reports omega .9885 with a standard error of measurement of 1.61 IQ points for its fifteen subtest General Ability Index, and its weakest published composite, the two subtest Processing Speed Index, reports omega .8790 with a standard error of 5.22 points. A two subtest composite is the closest structural analogue to a short single session form, and even that sits above the range implied by BRGHT's published band.

The honest framing is not that BRGHT is imprecise for a bad reason. It is that 40 items in 30 minutes, spread across three content areas and partly consumed by a climb from level 1, cannot support a narrow interval, and the site's own display says so. A test that shows you a 26 point band is telling you something true about itself. What a confidence interval represents and why every professional report prints one is on the reliability page, and the score to percentile relationship is on the score versus percentile page.

7 The Norms Are Anchored to BRGHT's Own Users

This is the finding that determines what a BRGHT score means, and BRGHT states it rather than hiding it. The method page describes the final step without ambiguity: the raw score is compared to the entire population of test-takers to calculate the user's IQ score, with a median of 100 and a standard deviation of 15. The shared result page says the same thing in consumer language, reporting that a taker scored better than a given percentage of the people who ever finished this test.

Follow the consequence. The median BRGHT user scores 100 by construction, whoever the median BRGHT user happens to be. If the people who choose to take an online IQ test skew younger, more educated, more numerate, or simply more curious about their own cognition than the general adult population, then a BRGHT 100 is the middle of that group and not the middle of the population. The number is a percentile inside a self-selected sample wearing the clothing of a population standard score.

Compare the alternative. Canivez, reviewing the WAIS-IV for the Buros Center's Mental Measurements Yearbook in 2010, records a standardization sample of 2,200 people selected by stratified proportional sampling on age, sex, race and ethnicity, education level, and geographic region, matched against the October 2005 United States Census and split into 13 age groups. That machinery exists precisely so a percentile means something outside the group that produced it. Nothing sold at consumer prices reproduces it, and how the machinery works is explained on the norming page.

There is a second consequence that is easy to miss. A self-anchored scale drifts as the user base changes. If BRGHT's traffic shifts toward a different population, every previously issued score silently refers to a different reference group, and the site publishes no norm version, no freeze date, and no statement of when the reference distribution was last computed. Age is not mentioned in the scaling description at all, which matters because performance on speeded reasoning tasks varies systematically with age, a pattern set out on the average IQ by age page.

To be fair on the point BRGHT gets right: disclosing the reference frame at all puts it ahead of most competitors, several of which report a percentile with no statement of what group it ranks you against. Saying the frame is your own users is a real disclosure. It just answers a narrower question than the one most people came for.

8 The Country Table Is a Map of Who Visits the Site

BRGHT's ranking of countries by average IQ is the clearest available demonstration of what a self-selected sample does to a statistic, and it is published on the site's own pages. Read on September 5, 2026, the country ranking puts South Korea first at 117.61, Togo second at 116.06, the French Southern Territories third at 115.71, Taiwan fourth at 114.41, and China fifth at 112.99. Every one of the top 40 entries sits above 105.

The third entry settles the question. The official administration of the French Southern and Antarctic Lands describes the territory as one without a permanent population or elected representatives, administered by a prefect. A ranking that places a territory with no permanent residents third in the world by average IQ is not measuring national ability. It is measuring how many people typed a country name into a form, and how able those particular people happened to be.

The missing column is the whole story. Each country page repeats the same global figure, stating that the average is based on the 1,485,900 people who completed an IQ test at BRGHT, and none of them publishes how many of those people reported that particular country. A mean without an n is not a statistic anyone can evaluate, and for a territory with no permanent population the n is necessarily tiny, which is why it lands near the top.

BRGHT does attach a caution to these pages, noting that intelligence is complex and multifaceted and that its test measures logical, numerical, and spatial reasoning. That caveat is about construct coverage. It does not address the sampling problem, which is the one that produced the ranking. The same mechanism runs through the site's average IQ by field of study and by occupation tables, where Data Scientist appears at 119 and Bioengineering at 117, again with no cell sizes. Why occupational and educational averages behave the way they do when the sample is drawn properly is covered on the average IQ by profession page and the average IQ by education page.

9 What the Report Contains, Including Its Best Feature

The per item record BRGHT shows after a test is the most transparent artifact in the consumer IQ market, and it is worth taking seriously. The published example result shows a table with one row per question, listing the section it came from, whether the answer was correct, the difficulty level of the item, and the time the taker spent on it, with each question number linking to a worked solution. Nothing a clinical publisher sells offers a test taker that.

The rest of the report is thinner. The example shows an IQ figure, a distribution curve, three section labels for Numerical, Logical, and Spatial Reasoning placed on a five point band from Low to High, and a single percentile statement against BRGHT finishers. The sharing format additionally reports per section percentiles and the score interval discussed earlier. There is no age reference, no composite structure, no statement of what the section scores are built from, and no indication of how many items contributed to each section.

That last gap has consequences. Forty items divided across three areas leaves roughly a dozen items per area at best, and in an adaptive sequence the split is not fixed, so one section may rest on far fewer. A section level percentile built on a handful of items carries a very large error, and reporting it beside a total on the same visual scale invites a reader to compare them as if they were equally solid. For contrast, Pearson's published WAIS-IV sample score report pairs every composite with a percentile rank, a 95 percent confidence interval, and a qualitative description.

Two features sit at the boundary between product and privacy. BRGHT publishes profile pages that carry a person's name, occupation, employer, country, and IQ on an indexable page, and its sharing feature produces public result links. Separately, its test others feature lets one person generate a test link for another and receive that person's results by email and in their account. The privacy policy lists the personal data collected, including first name, last name, age group, country, highest education level, most relevant educational institution, field of study, professional experience, most recent employer, email, and IP address, along with a question ID, correct or wrong, and time spent for every item answered. It enumerates the standard European data subject rights and commits to anonymizing before any research sharing.

10 Price and Billing, as of September 2026

BRGHT advertises a price on its homepage and documents none of the billing anywhere else, which is the clearest commercial gap in the product. The homepage, read on September 5, 2026, states that BRGHT offers unlimited testing with its comprehensive, data-backed results starting at only 19.95 dollars. The phrase starting at is doing work, and nothing on the site says what it starts at or what the higher tiers are.

The terms and conditions do not fill the gap. As noted earlier, that document contains no price, no billing cycle, no renewal clause, no cancellation mechanism, and no refund policy, and it was last updated in May 2023. Its definitions section refers to a Paid Plan you subscribe to and to an Initial Term specified in the Sign-up Form, so the contract contemplates a recurring arrangement without describing one. The checkout itself is a Stripe hosted session, which means the authoritative figure appears only at the moment of payment.

Users report a different shape from the advertised one. BRGHT's Trustpilot page showed a TrustScore of 2.8 across 164 reviews on September 5, 2026, with 15 percent of them at one star. A United States reviewer on October 8, 2025 described a one dollar trial followed by a 24.99 dollar charge. A reviewer in Great Britain on September 27, 2025 described one euro becoming a 49.99 euro charge days later. An Australian reviewer on September 25, 2025 described 49.99 dollars leaving their card, and another United States reviewer on September 24, 2025 described expecting one small fee and receiving recurring monthly charges.

Because Assessly BV is a Dutch company operating under Dutch law, the relevant consumer framework is Directive 2011/83/EU, whose Article 9 grants a fourteen day withdrawal right and whose Article 16(m) removes that right for digital content once delivery has begun with the buyer's prior express consent. A test report delivered on screen is digital content whose performance has begun. Practical guidance on reading a checkout screen before confirming, and on what a card dispute can and cannot recover, is on the page on paid IQ test billing models.

11 Strengths and Limitations Side by Side

A review that finds nothing good in a competitor is not a review, so here is the ledger, restricted to what BRGHT publishes about itself. Read the left column before the right one, because several of the strengths are genuinely unusual in this market and would take a well funded competitor real effort to match.

The pattern across the rows is coherent. BRGHT invested in the parts of test construction that are intellectually interesting, namely adaptive delivery, item calibration, an item pool large enough to resist exposure, and an open data posture. It did not invest in the parts that are tedious and expensive, namely an external reference sample, reliability estimation, validity evidence, and a consumer contract. Those are different kinds of work, and the split is visible in every row.

12 Who It Fits, Who It Does Not, and Where ACIS Sits

Take BRGHT when you want a fast, well built reasoning challenge and you read the number as a rank among its own takers, which is what it is. On that use it is a reasonable product and better engineered than most things at its price. The adaptive delivery is real, the per item feedback is genuinely educational, and the site tells you enough about its own method that you can evaluate the result rather than trusting it. That is more than most of the market permits.

Do not take it if you need a number that means something outside BRGHT. It is not evidence for a diagnosis, a gifted placement, an accommodation request, a job application, or a high IQ society submission, because those bodies define their own accepted instruments and an unproctored web test is essentially never among them. It is also not the right instrument if you want to know where a specific ability sits, because three content areas measured across 40 shared items cannot separate reasoning from speed, and it measures no verbal or working memory construct at all. For a comparison of what an unsupervised score can and cannot support, see the page on online test accuracy, and for the proctored route see the page on where to take a test.

ACIS competes with BRGHT and has its own limits, which are worth stating in the same breath. ACIS is a paid, self-administered online assessment for adults aged 16 to 90, running 20 subtests across six Cattell-Horn-Carroll domains and reporting a Full Scale IQ with the six primary indices, in three tiers at 15, 30, and 50 dollars charged once with no subscription. It is not a clinical instrument, not diagnostic, and not appropriate for hiring decisions. Its adult reference frame is documented in the technical manual.

Where we think the difference is real, we can point at published numbers rather than adjectives. The ACIS technical manual reports omega .9885 with a standard error of measurement of 1.61 points and a g loading of .954 for the fifteen subtest General Ability Index, omega .9745 with a standard error of 2.40 for Verbal Comprehension, and omega .9727 with 2.48 for Fluid Reasoning. It also publishes its weakest composite, the two subtest Processing Speed Index, at omega .8790 with a standard error of 5.22. BRGHT publishes no equivalent figure at all, and that gap, rather than any claim about which test is better, is the concrete difference a reader can check.

Where BRGHT is ahead of us, we would rather say it than not. Its open data offer to researchers is a stronger transparency commitment than ours, its per item feedback after a session is better than what we return, and its item pool is larger. Our reasons for a longer battery are on the comprehensive assessment page, and the criteria we use to compare instruments across the market are on the accuracy guide.

13 Sources Behind This Page

Everything attributed to BRGHT below was read from its own public pages on September 5, 2026 and may change. Where a claim rests on an inference or on third party reports, the section that makes it says so.

• BRGHT homepage. States the 2PL item response theory claim, a pool of over 2 million participants, unlimited testing starting at 19.95 dollars, and a report covering logical, numerical, and spatial reasoning.
• BRGHT, Mission, Vision, and Commitments. The scoring method: 40 questions in 30 minutes, 20 difficulty levels from Rasch and 2PL models fitted to over 500,000 test-takers, a level 1 start with plus one and minus one steps, points equal to item level, and a raw score compared to the entire population of test-takers at a median of 100 with a standard deviation of 15.
• BRGHT start page. Forty questions in 30 minutes, a pool of over 5000 questions, 45 seconds granted for the first question with 45 more added per question, sequential answering with no skipping, and the instruction to answer all questions for the highest score.
• BRGHT, Data Request for Psychometric Research. An item pool of approximately 3,000 questions in 10 categories answered more than 75,000 times per day by around 2,500 unique daily users, plus the item, user, and response fields offered to researchers and the conditions attached.
• BRGHT terms and conditions, last updated May 2023. Names Assessly BV, KVK 76121224, Kerkstraat 310, 1017 HC Amsterdam, under Dutch law. Contains no price, renewal, cancellation, or refund clause, and states that testing is a statistical process with inherent randomness whose results may not accurately reflect reality.
• BRGHT privacy policy. Lists the personal data collected including name, age group, country, education, institution, field of study, professional experience, employer, email, and IP address, plus per item outcome and time, and describes anonymised sharing with university researchers.
• BRGHT country ranking. South Korea 117.61, Togo 116.06, French Southern Territories 115.71, Taiwan 114.41, China 112.99, Falkland Islands 111.80, Tokelau 111.34, Holy See 107.57, Tuvalu 107.07, with the global figure of 1,485,900 completions and no per country sample size.
• BRGHT example result page. The report layout: an IQ figure, a distribution curve, three section bands, a percentile against BRGHT finishers, and a per question table listing section, outcome, difficulty level, and time used.
• BRGHT test others page. The feature that generates a personal test link for another person and returns their results by email and in the sender's account.
• Terres australes et antarctiques françaises, the official administration. States that the territory has no permanent population and no elected representatives and is administered by a prefect.
• Trustpilot, brght.org. TrustScore 2.8 across 164 reviews with 15 percent at one star on September 5, 2026, including dated user reports of a one dollar or one euro entry charge followed by larger charges. User reports, not independently verified.
• Pearson (2008). WAIS-IV sample score report. The reference point for what a report contains: every composite paired with a percentile rank, a 95 percent confidence interval, and a qualitative description.
• Buros Center for Testing. Publisher of the Mental Measurements Yearbook, whose 2010 review of the WAIS-IV by Gary L. Canivez documents the standardization sample of 2,200 stratified to the October 2005 United States Census across 13 age groups.
• Directive 2011/83/EU on consumer rights. Article 9 grants a fourteen day withdrawal right and Article 16(m) removes it for digital content once performance begins with prior express consent, which is the framework governing a Dutch seller of an instantly delivered report.

The yardstick applied throughout is the Standards for Educational and Psychological Testing (2014), published jointly by the American Educational Research Association, the American Psychological Association, and the National Council on Measurement in Education. The Standards require a test to document its norms and the composition of its reference sample, to report reliability evidence for each score it produces, to provide validity evidence for each intended use, and to present scores with their measurement error. Measured against that list, BRGHT documents its norm frame and its scoring rule, and does not document its reliability or its validity. The APA standards on test use place the corresponding duty on whoever reports a score: state what it supports, state what it does not, and do not let a design choice stand in for evidence that was never gathered.

14 Frequently Asked Questions