Cognitive ability tests in hiring: how to assess reasoning and soft skills without bias, and what the law requires (2026)

The approaches that let a team assess cognitive ability and soft skills with the least risk of bias are job-related, validated measures used together rather than one test alone: a work sample or job knowledge test, a structured interview scored against written anchors, a situational judgment test for interpersonal judgment, and a cognitive ability test only where your job analysis shows the job needs it and the validation evidence covers your job. Around them you need process controls: pass rates by group checked at each step, an accommodation route, no score adjustments by group, and records. Under Title VII, a test with disparate impact must be job related and consistent with business necessity; that statute is unchanged as of November 2026, even though federal agencies have stepped back from the Uniform Guidelines.

This page is about tests of reasoning, numerical and verbal ability and learning, and about how to measure soft skills alongside them. Personality tests and the ADA's medical-exam limit are covered in pre-employment testing laws, interview questions for this ability in interview questions for technical aptitude, and the impact arithmetic in the four-fifths rule.

Why are cognitive ability tests the classic bias case?

Because they predict well and screen unevenly. The US Office of Personnel Management says tests of general cognitive ability "are good predictors of job performance and training success for a wide variety of jobs," and that "the more complex the job or training demands, the better these tests work." The same page warns that they "typically produce racial and ethnic differences larger than other valid predictors," and recommends using other methods with them to lower potential adverse impact (OPM, cognitive ability tests).

The first disparate impact case was about exactly this kind of test. In Griggs v. Duke Power Co. (decided March 8, 1971), the company required new hires for every department except Labor to have a high school education and satisfactory scores on two aptitude tests, the Wonderlic Personnel Test, which "purports to measure general intelligence," and the Bennett Mechanical Comprehension Test. Neither requirement had been shown to relate to performance in the jobs. The Supreme Court held that Title VII bars practices "fair in form, but discriminatory in operation," and that tests "must measure the person for the job and not the person in the abstract."

Four years later, in Albemarle Paper Co. v. Moody (June 25, 1975), the employer had commissioned a validation study of its two general ability tests, and the Court found it materially defective. Its four flaws are still the right questions to put to any test provider:

• It did not validate the tests for all the jobs they were used for.
• It compared scores with supervisors' subjective rankings, with no way to know what job-performance criteria the supervisors used.
• It studied employees near the top of each line of progression, though the tests screened new workers for entry jobs.
• It studied only job-experienced white workers, though the applicants were younger, largely inexperienced and in many cases not white.

Which tools assess cognitive ability and soft skills, and how do they compare on bias?

The table summarizes OPM's assessment guidance for each method, as of November 2026. It describes types of assessment, not products; for how to compare providers, see how to evaluate pre-employment assessment vendors.

For soft skills, the safest route is to define each skill as behavior the job requires, from the job analysis, and measure it at least twice in different ways: a situational judgment test or role play, plus behavioral questions in a structured interview, each scored against written anchors. "Culture fit" or "presence" without a written definition is where bias gets in.

What does Title VII require when a test screens out one group more often?

• Tests are allowed. An employer may "give and to act upon the results of any professionally developed ability test" that is not "designed, intended or used to discriminate" (42 U.S.C. 2000e-2(h)).
• Impact shifts the burden. Disparate impact is established if a candidate shows a practice causes a disparate impact on the basis of race, color, religion, sex or national origin and the employer fails to show it is "job related for the position in question and consistent with business necessity," or if the candidate shows an alternative practice with less impact that the employer refuses to adopt (2000e-2(k)(1)(A)). Each particular practice is analyzed on its own, so a test can be challenged even when the overall hiring numbers look balanced.
• No group norms. It is unlawful "to adjust the scores of, use different cutoff scores for, or otherwise alter the results of" employment-related tests on the basis of race, color, religion, sex or national origin (2000e-2(l)). If a test shows impact, change the test or how you use it, not the scores.

What do the Uniform Guidelines say, and where do they stand in November 2026?

The Uniform Guidelines on Employee Selection Procedures (29 CFR part 1607) set out how to validate a test and measure its impact. Five provisions matter most for cognitive and soft-skill tests:

• Validation: a procedure with adverse impact "will be considered to be discriminatory" unless validated (1607.3(A)), by a criterion-related, content or construct validity study (1607.5(A)).
• Content validity is not enough for these tests: a content strategy "is not appropriate" for procedures that "purport to measure traits or constructs, such as intelligence, aptitude, personality, commonsense, judgment, leadership, and spatial ability" (1607.14(C)(1)). A provider should show that scores relate to performance in jobs like yours, not just that the items look job-like.
• Alternatives: where two procedures are substantially equally valid, use the one with lesser adverse impact (1607.3(B)).
• Four-fifths rule: a group's selection rate below four-fifths of the highest group's rate is "generally" regarded as evidence of adverse impact, with caveats for small numbers (1607.4(D)). It is a rule of thumb, not a legal threshold.
• Cutoff scores: set them "so as to be reasonable and consistent with normal expectations of acceptable proficiency within the work force" (1607.5(H)).

Their federal standing has changed, though the statute above has not:

Even if the EEOC removes its copy, the Guidelines' methods remain a well-understood way to document that a test is job related, which is what the statute asks of you. How the federal changes fit together is in Title VII disparate impact and AI hiring tools.

How do you accommodate candidates on cognitive and timed tests?

The ADA treats tests in two ways. Tests that "screen out or tend to screen out" people with disabilities must be job related and consistent with business necessity (42 U.S.C. 12112(b)(6)). And employers must select and administer tests so that results "accurately reflect the skills, aptitude, or whatever other factor" the test measures, rather than a candidate's impaired sensory, manual or speaking skills, except where those skills are what the test measures (12112(b)(7); 29 CFR 1630.11).

• Say in every test invitation how to ask for an adjustment and who handles it, outside the hiring team.
• Common adjustments: extra time, a screen-reader-compatible format, breaks, a human-read version, or another way to show the same ability.
• Keep a time limit only if speed is part of the job. A reasoning test timed for convenience measures speed as well as reasoning.
• In California, the Civil Rights Council's rules say automated-decision system assessments, "including tests, questions, or puzzle games," that elicit information about a disability may be an unlawful medical inquiry (California Civil Rights Department).

The same approach for interviews is in ADA accommodations in interviews.

What rules apply when the test is automated or AI-scored?

Federal law applies to every test, however it is scored. A few cities and states add their own rules for automated tools. The main ones, as of November 2026:

Other states and their dates are in AI hiring laws by state.

How do you put a fair cognitive or soft-skills assessment in place?

1. Write the job analysis first. List the tasks and the abilities they need, and where reasoning speed, numerical work or interpersonal judgment actually matter.
2. Choose the measures from the analysis. Prefer a work sample or job knowledge test for what can be observed directly; add a cognitive test only for abilities the job needs and other measures miss.
3. Get validation evidence that fits your job. Ask for criterion-related or construct evidence for jobs like yours, with the performance measures used and who was in the sample, using Albemarle's four flaws as your checklist.
4. Set the cutoff from the job, at the level of acceptable proficiency, and write down why. Do not raise it to thin a large applicant pool without considering the impact.
5. Pilot and check pass rates by group at each step, using the four-fifths rule as a screen and statistics where numbers allow.
6. Look for a less discriminatory alternative when a step shows impact: a different test, a lower but job-based cutoff, a combination with a structured interview, or a work sample in its place.
7. Offer accommodations in every invitation, handled outside the hiring team.
8. Never norm or adjust scores by group.
9. Keep the records. Federal rules require employers to keep hiring records, including test results, for one year (29 CFR 1602.14), longer if a charge is filed; California requires four years for automated-decision data.
10. Review yearly and whenever the job, the test version or the applicant pool changes.

Questions to ask a test provider

ASSESSMENT PROVIDER QUESTIONS   Role: [ ]   Date: [ ]

1. What does the test measure, and which of our job's tasks does that map to?
2. What validation evidence do you have for jobs like ours: criterion-related
   or construct? Which performance measures, sample size and sample makeup?
3. What score differences by race, sex and ethnicity have you found, and on
   which populations?
4. Do your reports use separate norms or cutoffs by any group? (They must not.)
5. How do you recommend setting a cutoff, and what impact does it have?
6. Is the test timed? Why, and what accommodations and formats are available?
7. Is any part scored by AI or an algorithm? Has it had a bias audit that
   meets New York City Local Law 144, and can we see the summary?
8. Was any instrument designed to identify a mental health condition?
9. What candidate data do you keep, for how long, and who can see it?

Summary

Not legal advice. Statutes, regulations and agency pages are quoted as of November 2026; check each linked source before you rely on it.