Why generic skills tests are broken, and what CV-personalised testing fixes

Generic skills tests have a structural problem, and it is not the questions. It is the fact that every candidate sits the same ones.

When a test is drawn from a shared question bank, three things follow. The questions leak, because thousands of candidates see them and some of them write them down. Practice platforms index them, because there is a market for passing the exact test your next employer uses. And the test stops measuring the skill, because it now measures familiarity with the bank.

This is not a hypothetical decay. Search for the name of any major assessment vendor plus the word "answers" and you will find communities, paid prep courses and full walkthroughs of specific tests. The vendors rotate questions, but rotation is a treadmill: each new question is fresh for weeks, then it is in the prep ecosystem too. The half-life of a banked question is short, and it only ever gets shorter.

What a generic test actually measures

Be precise about what a banked test score tells you. It tells you the candidate performed well on a set of questions that were written for everyone and seen by many. It does not tell you whether the candidate can do the specific things their CV says they have done.

That gap matters more than it used to, for two reasons.

First, CVs themselves have changed. Most CVs that reach a pipeline in 2026 have been drafted or polished with AI assistance. The claims are better written, better targeted at the job description, and harder to separate on paper. The CV has become a weaker signal exactly when hiring teams need a stronger one.

Second, the candidates who practise the most are not necessarily the ones who can do the job best. A banked test rewards preparation time. That systematically favours people who can spend evenings on prep platforms, and it tells you nothing about the claim you actually need to verify: can this person do what their CV says, at the level it says?

The interview was always the patch

Hiring teams have known this for years, which is why the real verification has always happened in interviews. A good technical interviewer reads the CV, picks the claims that matter for the role, and probes them. "You say you led the migration to the new platform. What broke? What would you do differently?"

That works. It is also expensive, inconsistent and late. It costs senior staff an hour per candidate, it depends entirely on which interviewer the candidate gets, and it happens after screening, which means the screening itself was done on the weakest evidence in the whole process.

The question worth asking is: what if the probing a good interviewer does could happen at the assessment stage, for every candidate, consistently?

What CV-personalised testing changes

A CV-personalised assessment starts from the candidate's own CV rather than from a bank. The platform reads the CV, identifies the specific skills and experience it claims, and generates a unique set of questions that probe those claims at the seniority the CV asserts.

That single change repairs the broken parts of the generic model:

• There is nothing to practise. No two candidates sit the same test, because no two CVs make the same claims. Prep platforms cannot index a test that did not exist until the CV arrived.
• The score means something specific. Instead of "performed well on general questions", you learn "the assessment evidence supports, or does not support, what this CV claims". KimonRecruit expresses this as a CV Confidence Score: a measure of how well assessment performance backs the CV.
• Seniority is part of the question, not an afterthought. A claim of eight years leading infrastructure work gets probed differently from a claim of two years contributing to it. Generic banks calibrate difficulty per question; personalised generation calibrates it per claim.
• Inflated claims surface naturally. When a CV claims a skill and the candidate cannot engage with a question generated from that exact claim, the gap is visible and specific. You know which claim did not hold, not just that a score was low.

What it does not change

A personalised assessment is still evidence, not a verdict. This matters legally as much as ethically. Under the EU AI Act and UK equality law, the safe and defensible posture is that assessment output is decision support for a human reviewer, never an automated gate that takes candidates out of a pipeline on its own.

KimonRecruit is built on that posture by construction. Scores are presented to recruiters alongside the CV and structured scorecards, every score is replayable from the prompt and model version that produced it, and a human makes every progression call. Adverse-impact monitoring runs across the pipeline so that any group-level skew in outcomes is visible early, not at litigation time.

Personalised testing also does not remove the need for good role definitions. A test generated from a CV probes what the CV claims; the hiring team still has to decide which claims matter for the role. The platform makes the evidence better. It does not make the judgement for you.

The practical difference for an SME

If you are hiring five to a hundred roles a year without a dedicated talent team, the practical difference looks like this. With a generic tool, you pay per candidate or per test for scores that prep-savvy candidates can inflate, and you still pay for an ATS on top. With CV-personalised testing inside the ATS, every applicant who reaches the assessment stage produces evidence tied to their own claims, and the review happens in the same pipeline where the rest of the hiring lives.

The CV told you what the candidate says they can do. The assessment should tell you whether that specific story holds. Generic tests were never built to answer that question. Personalised ones are built from it.