Learn
What is AI CV screening?
AI CV screening is the use of a language model to read every application for a role and rank it against that role's stated requirements. Unlike keyword search, it reads the CV as prose — so "led the migration off Oracle" can count as database experience even though the word never appears in a skills list. A well-built screener returns an ordered list with the evidence behind each position, not a verdict.
1.What it actually does
A screener takes two inputs: the requirements for a role, and a pile of CVs. It reads each CV against each requirement and produces a score, a band, and — this is the part that matters — the specific lines of the CV that justify it.
The output is an ordering. Candidate 14 sits above candidate 3 because of something the model can point at. That is the whole product. Everything else built on top of it, including everything HireOS builds on top of it, depends on that ordering being explainable.
2.How it differs from keyword search
An applicant tracking system searches for strings. Type "Kubernetes" and you get the CVs containing the word "Kubernetes". This fails in both directions: it misses the platform engineer who wrote "k8s", and it promotes the candidate who listed it once in a training course from 2019.
A language model reads for meaning. It can recognise that eight years running production infrastructure is stronger evidence than a certification, and that a candidate describing the same job three ways is describing one job. It can also be wrong about all of this, which is why the evidence has to be visible.
3.Why the evidence matters more than the score
A score of 82 tells you nothing. A score of 82 with three quoted lines from the CV tells you whether the model understood the role, and lets you disagree with it in about four seconds.
It also matters legally. Under UK GDPR a rejected candidate can ask why they were rejected, and "the software scored them low" is not an answer. A screening decision needs a reason that existed when the decision was made.
4.Where it fails
It fails when the requirements are vague. "Strong communicator" cannot be evidenced from a CV, and a model asked to score it will produce a confident number built on nothing.
It fails on career shape. A model reads what a CV says, not what it implies — a two-year gap, a sideways move, a return from a career break. Those are the parts a recruiter is genuinely better at, which is why screening should end at the shortlist rather than at a decision.
Where this gets it wrong
- A screener is only as good as the requirements it is given. Vague criteria produce confident nonsense.
- It reads what is written. A candidate who undersells themselves on paper will rank below one who does not.
- It should never auto-reject. Ranking is a suggestion; rejection is a decision, and decisions belong to people.
How HireOS does it
Everything above is true whether or not you use anything we built. These are the modules where it shows up in HireOS.
Common questions
AI CV screening, in short.
Read next
What is a screening rubric?
A screening rubric is the written standard a role is judged against, set before the first application is read. What goes in one, and why writing it first changes the outcome.
What is an audit trail in hiring?
An audit trail records what was decided, by whom, on what evidence, and when. What UK GDPR requires from a screening process, and what a defensible record looks like.
The offer
See it on a role you are working today.
Twenty minutes, your own job description, your own CVs. You judge it on the shortlist it produces.
- 1Twenty minutesYou bring one live role. We look at it together — no deck, no discovery questionnaire.
- 2Set up on the callYour job description becomes a rubric while you watch, and your own CVs go through it.
- 3Then 7 days aloneYou run it on real work and judge it on the shortlist, not on anything we said.