How online reasoning tests work

The short version

• Online tests are widely used as an early filter for graduate and early-career jobs, and the largest providers each report tens of millions of assessments a year across dozens of countries.
• The best single predictor of job performance in the largest recent review is the structured interview, not a test. Even that predicts only a modest share of how well people do at work.
• Reasoning tests are a weaker signal than once believed. Personality questionnaires are weaker still, and applicants can and do adjust their answers.
• Scores rise with familiarity. That means a test partly measures whether you have seen the format before, which is something you can fix.

Who runs these tests

Some of the largest providers describe their reach like this, in their own words:

• SHL: “45 million SHL assessments are taken every year”, with support in 150 countries and products translated into up to 39 languages (shl.com).
• Aon: “Each year, we administer more than 30 million assessments in over 40 languages across 90 countries” (aon.com).
• Alva Labs, founded in Stockholm in 2017: 650+ organisations use it and 1.5M+ candidates have completed an assessment; it offers a logic test of general mental ability and a Big Five personality test (alvalabs.io/assessments, alvalabs.io/trust/science).

The point is scale, not any one provider: the same kinds of test, and often the same products, are used by employers in many countries at once. How you perform on an unfamiliar format can decide whether a person ever reads your CV.

What the tests measure

• Numerical reasoning: reading tables and charts, percentages, ratios and rates under time pressure.
• Verbal reasoning: deciding whether a statement is true, false or cannot be determined from a short passage, using only what the passage says.
• Logical and abstract reasoning: finding the rule in a sequence or matrix of shapes and applying it.
• Situational judgement: choosing the most and least effective responses to a workplace scenario.
• Personality questionnaires: self-report statements, often scored on the Big Five traits. These have no right answers in the usual sense, and Wamberry does not coach them.

Reasoning tests are usually timed, and some are adaptive: the next question gets harder or easier depending on your answers.

How well do they predict job performance?

The largest recent re-analysis of selection research (Sackett, Zhang, Berry & Lievens, 2022, Journal of Applied Psychology) estimated these average correlations with job performance:

• Structured interviews: .42
• Cognitive ability tests: .31
• Unstructured interviews: .19
• Conscientiousness (the strongest Big Five trait): .19

Squaring a correlation gives a rough sense of how much of the variation in performance it accounts for: about 18% for structured interviews, about 10% for cognitive tests and under 4% for conscientiousness. The same review concluded that earlier, much higher estimates for cognitive tests had been inflated by the way they were corrected. These are averages across many studies; validity varies by job and by how well each method is designed.

So the best signal available is a modest one, and the tools commonly used to filter applicants before anyone speaks to them are weaker still. In our view, leaning on test and personality scores at the first stage, while the better method is left until later or skipped, is poor hiring practice. It is efficient for the employer; it is not well supported by the evidence on who will do the job well.

That does not make reasoning tests worthless. A correlation of .31 is real. The problem is how heavily a single score is trusted, and whether it is used as a hard cut-off rather than as one piece of evidence.

Personality questionnaires can be gamed

A meta-analysis comparing job applicants with non-applicants found applicants score noticeably higher on desirable traits such as conscientiousness and emotional stability, consistent with people presenting themselves in the best light when a job is at stake (Birkeland et al., 2006, International Journal of Selection and Assessment). A tool that the people being assessed can steer is a weak basis for rejecting them.

Familiarity changes scores

Across 107 samples and 134,436 people, retaking a cognitive ability test raised scores on average, roughly the difference between the 50th and the 60th percentile (Hausknecht et al., 2007, Journal of Applied Psychology). Part of a first-time score is therefore unfamiliarity with the format, not ability. That is unfair on candidates who have never seen these tests, and it is also the part you can change.

Your rights

• See your data. Under the GDPR you can ask an employer or test provider for a copy of your personal data, which can include your scores. In the UK see the ICO subject access request service; in Sweden see IMY on your rights.
• How to ask for your results. You can make a data access request to the organisation that handled your application. Ask for the personal information it holds about your assessment, such as your scores, rankings and notes about you. Its privacy notice should tell you how to contact it. A request is normally free, and the organisation generally has one month to respond, though it can extend this for complex requests. You may not receive every record you ask for. Find out more from IMY in Sweden or the ICO in the UK.
  A ready-made request email

  Mikhail Girshovich has published a template email for asking employers what they hold about your application. It is his work, shared on GitHub; we link to it rather than copy it.

  Open the template by Mikhail Girshovich on GitHub

• Ask for a human. In the EU, including Sweden, Article 22 of the GDPR gives you the right to human involvement and to contest a decision with significant effects made solely by automated means. In the UK the Data (Use and Access) Act 2025 changed these rules, but the organisation must still let you make representations, obtain human intervention and contest the decision (Article 22C UK GDPR, as inserted by the 2025 Act).
• Ask for adjustments. If you are disabled, for example dyslexic, you can ask for adjustments such as extra time. In the UK see gov.uk on reasonable adjustments; in Sweden see the Equality Ombudsman (DO), which treats inadequate accessibility as a form of discrimination.
• Ask how long a result is kept and what it is used for. A score from one application may be stored and reused. You are entitled to know.

How to prepare

• Learn each format before the real thing. Reading the instructions for the first time under time pressure costs marks.
• For verbal tests, answer from the passage alone. “Cannot say” is correct whenever the passage does not settle the statement, even if you know the answer from elsewhere.
• For numerical tests, estimate first, then calculate. If two answer options are close, check units and which column you read.
• For shape puzzles, test one rule at a time: position, rotation, count, colour, size.
• Practise timed, then review every miss. The review is where the improvement comes from.
• Do the real test when you are rested, in a quiet place, with a calculator and paper if they are allowed.

Wamberry is built for exactly this: free practice in numerical, verbal, logical and abstract reasoning, with a full worked explanation every time you miss a question. Start practising.

A note on who wrote this

Wamberry sells test preparation, so we have an interest in this subject. That is why every figure above links to its source, so you can check it yourself.

Last updated: 8 November 2026