How should you answer personality questions in a job application?

You are forty questions into an application and it has started asking whether you enjoy being the centre of attention. There is a slider. You have a strong suspicion that one end of it gets you the job, and a weaker suspicion that they can tell if you reach for it.

Both suspicions are reasonable. The research on this is better than you’d expect, and it disagrees with itself in a way that is genuinely useful to know about.

The short answer

Answer as yourself at work, not as yourself in general. If a question could describe you differently at home and in a meeting, answer for the meeting. This is not a trick — it is the framing that performs better in the research. Contextualised personality measures (“at work, I…”) predict job performance at ρ = .25 against .19 for generic ones, per Sackett and colleagues’ 2022 re-analysis.

Don’t try to reverse-engineer the profile. People can distort their answers when told to — that much is settled. Whether real applicants do it enough to matter is genuinely contested, and the strategy has a specific failure mode: the scales designed to catch it are the ones that move most when you try.

Know that the questionnaire is not the main event. In the same 2022 revision, structured interviews are now the single most valid selection method at .42, ahead of cognitive ability at .31. Overall conscientiousness, the best of the Big Five predictors, sits at .19. Those revised numbers come from a re-analysis of earlier meta-analyses rather than new data — an argued methodological position, not settled fact — but they are the current best account.

Can people actually fake these tests?

Yes, and this part isn’t controversial.

Viswesvaran and Ones (1999) meta-analysed 51 studies in which people were explicitly instructed to distort their responses. They could. All five Big Five dimensions proved equally fakeable — there is no trait that resists it. Interestingly, “fake bad” instructions produced larger distortions than “fake good,” and the largest movements of all appeared on social desirability scales: the very measures built to detect impression management.

That last detail is the practically useful one. If a questionnaire includes those checks, the surest way to trip them is to present yourself as uniformly excellent — never irritated, never disorganised, never tired.

But note precisely what this establishes: capability under laboratory instruction. It is the ceiling, not the base rate. It does not tell you what a real person applying for a real job actually does.

Do real applicants fake, though?

Here the evidence splits, and both sides are well-designed studies with opposite conflicts of interest. This is worth seeing in full rather than resolved.

Griffith, Chmielowski and Yoshita (2007) had real applicants complete a personality inventory while applying, then retake it later under honest instructions. Depending on the confidence threshold used, between 30% and 50% had elevated their scores — and crucially, the distortion “resulted in significant rank ordering changes that impacted hiring decisions.”

Hogan, Barrett and Hogan (2007) ran something close to a natural experiment. Real applicants took a five-factor measure, were rejected, and six months later 5,266 of them reapplied for the same job and took the same measure again — a group with maximum motivation to improve their score. 5.2% or fewer improved on any scale, scores were as likely to fall as rise, and only 3 of 5,266 changed on all five scales beyond a 95% confidence threshold.

Read the interests honestly: Hogan and Hogan are the test publisher, and a finding that their instrument resists faking suits them. Griffith’s 30–50% range is wide precisely because it depends on where you set the threshold. Neither is a knockout.

Does it actually change who gets hired?

This is the question underneath, and the answer is a careful “sometimes, and less than the faking debate implies.”

Ones, Viswesvaran and Reiss (1996) titled their meta-analysis “the red herring” for a reason: social desirability scales did not predict job performance, and removing social desirability variance from the Big Five left their criterion-related validity intact. In aggregate, impression management does not appear to break the prediction.

Against that, Griffith’s rank-order finding is the practical harm. If everyone inflates slightly but unevenly, the average correlation between test and performance can survive perfectly well while the order of the shortlist changes — and hiring happens at the top of a ranked list, not in the aggregate. Those two findings are compatible, and together they are probably the fairest summary available.

It is also worth knowing that serious people think the whole practice is oversold. Morgeson and colleagues (2007) — six authors including former editors of Personnel Psychology and Journal of Applied Psychology — argued that faking on self-report measures cannot be avoided, that the deeper problem is the low validity of personality tests for predicting job performance, and that their use in selection should be reconsidered. That is a commentary rather than new data, but the 2022 numbers (.19 for overall conscientiousness) do not exactly contradict them.

Is it fair for them to ask me this at all?

Your discomfort is well documented and widely shared, which is worth saying plainly.

Hausknecht, Day and Thomas (2004) meta-analysed applicant reactions across 86 samples and 48,750 people. There is a clear hierarchy of what candidates consider fair: interviews and work samples rank highest, cognitive ability tests in the middle, and personality inventories near the bottom, alongside honesty tests and biodata. Face validity — whether a method looks related to the job — was a strong predictor of whether people judged the process just.

So the instinct that a slider about parties has little to do with the role is not you being awkward. It is the modal reaction, and it is one reason employers who use these well explain them.

What you are entitled to ask depends on where you are. It is reasonable to ask what the assessment is used for, whether it is one input or a filter, and whether you can see your results. In the UK and EU there are specific protections around automated decision-making and discrimination that go beyond fairness perceptions — but those are jurisdiction-specific legal questions, and this post is not the place to answer them.

So what should I actually do?

• Answer for the work self. Where a question could go either way, picture yourself at work. It is honest, and it is the version that predicts better (.25 vs .19), so a well-designed process wants it.
• Use the full range. Straight-line answering at the top of every scale is the pattern that trips detection, and it makes you look like nobody in particular.
• Don’t guess the target profile. You do not know the job’s criterion. The assumption that every employer wants maximum extraversion is wrong often enough to cost you — plenty of roles select for the opposite.
• Be consistent, because these tests repeat themselves. Most inventories ask the same construct several ways. Consistency is easy when you’re reporting and hard when you’re constructing.
• Don’t over-weight it. The questionnaire is, on the evidence, a weaker signal than the structured interview you will likely also face. Prepare for that instead.

One last thing worth holding onto: a trait score describes how you answered relative to a norm group. It is not a verdict on your worth, and a rejection that follows one is not evidence about you as a person — often it is evidence about a profile someone decided on in advance.