How the CV match score works

The score is advisory. It does not represent a decision by an employer and does not guarantee that any applicant tracking system will accept or rank an application.

When you paste a job advert, it is read once to extract the requirements it actually names — in the advert's own words, not paraphrased into a generic skills taxonomy. That extraction also decides how much each of four areas matters for this particular advert.

Your CV is then measured against those requirements. The result is a number out of 100 that answers one narrow question: how well does this CV evidence what this advert asked for?

It is not a measure of how good your CV is, how qualified you are, or how any employer will react to it.

Every score is built from four weighted areas. The weights are set per advert within fixed bands, and always total 100:

• Hard skills — the concrete tools, technologies and disciplines named. Between 35 and 55 of the 100, typically 45.
• Context and recency — not a requirement the advert can name, but a measure of how current and substantial your evidence for the others is. Between 20 and 35, typically 30.
• Seniority match — whether the scope and level of your experience line up with what the advert asks for. Between 10 and 25, typically 15.
• Soft skills and domain — the interpersonal and sector-specific requirements. Between 5 and 20, typically 10.

The weights move because adverts differ. A role that never mentions leadership should not penalise a CV for not evidencing it, so an advert that says nothing about an area gives that area a low weight — and for hard skills, seniority and soft skills, an area the advert never raised scores as satisfied rather than as failed.

Context and recency works differently, and it is the one exception worth understanding. No requirement is ever filed under it: instead of asking whether you match something the advert named, it measures how current and how substantial your answers to the other three are. So it is not governed by the unasked-question rule — it scores 100 only when the advert named no requirements at all, and 0 when the advert asked and your CV evidenced none of it. It carries between 20 and 35 of the 100 either way.

This is the single most important thing to understand about the number: because each advert sets its own weights, two scores from two different adverts are not comparable. A 78 against a demanding, specific advert can represent a far stronger application than an 88 against a vague one. Never rank your applications by this number.

Not every mention of a skill counts the same. Where a requirement appears on your CV determines how much it is worth:

• A bullet point with a concrete result or figure — full credit.
• A plain bullet point describing the work — 80% of full credit.
• An entry in a skills list, with nothing behind it — 45%.
• A claim made only in your summary or tagline — zero.

That last rung is deliberate and it is the one people find surprising. Anyone can assert a skill in a headline; a summary is where unearned claims go. Credit for a requirement has to come from somewhere that shows the work.

The practical consequence is worth stating plainly: moving a skill from your skills list into a bullet that says what you did with it is the single most effective change you can make to a score — because it is also the change that makes the CV more convincing to a human.

Each piece of evidence is then adjusted for how substantial and how recent it is.

Depth reflects the seniority and duration of the roles a requirement appears in — the same skill evidenced across a five-year senior role counts for more than a single mention in a three-month placement.

Recency applies a discount to evidence last used more than 5 years ago. Skills decay, and an advert asking for current expertise is asking about now.

Undated content is not penalised. A CV that omits dates on a project is being unhelpful, not dishonest, and guessing against it would punish formatting rather than substance.

Three deductions can reduce a score. All three measure concentration rather than volume — a long, genuinely detailed CV is never penalised for being long:

• Keyword stuffing — one term repeated through a large share of a single section rather than appearing where it is relevant.
• Phrase flooding — a single bullet packed with wording lifted from the advert. This is the pattern behind the old "white-font keywords" trick.
• An employment gap of more than a year between roles, in recent history.

The first two exist because this score can be gamed, and a scoring system that rewards gaming produces CVs that fail with humans. Their exact thresholds are not published, for the same reason — knowing precisely where the line sits is knowing how to stuff a CV right up to it. The gap window is stated above because the product already tells you when it fires, and a rule you are measured against should not be a secret.

Where AI-composed content contributed to a deduction, that is reported separately, so you can see when a penalty came from something the assistant wrote rather than from your own CV.

This distinction matters for whether you can trust the number to be stable.

AI is used exactly once per advert: to read the advert and extract the requirement terms, the weights and the alternative wordings that should count as the same thing. It never returns a score.

The score itself is calculated by ordinary deterministic code, with no model involved, and recalculated every time you save. That means the same CV against the same advert always produces the same number — a score that drifted when nothing changed would not be a measurement.

It also means checking your CV repeatedly costs nothing and cannot quietly change the target.

Automatic extraction gets things wrong. Two controls let you correct it, and both feed straight back into the number:

• Highlight any phrase in the advert to track it as a requirement the analysis missed.
• Dismiss a requirement that does not genuinely apply to the role.

Both change the score, the requirement checklist and the advert's colour-coding together, because all three are views of one list. There is no hidden second opinion.

The advert text itself is tinted so you can see the state of every requirement at a glance: shown in the advert's own words, shown in different words, claimed only in a summary, or absent from the CV entirely.

Being direct about the limits, because a confident number invites more trust than it has earned:

• It does not predict what any employer's applicant tracking system will do. Employers run many different systems, configured differently, and almost none publish how they rank.
• It cannot see how your CV is parsed by a specific ATS. Parsing failures caused by layout are a real problem and are not what this measures.
• It does not assess whether you would be good at the job.
• It does not compare you to other applicants.
• It cannot be compared across adverts, because each advert sets its own weights.

What it does tell you is narrower and genuinely useful: which of the things this advert asked for are visible on your CV, which are only asserted, and which are missing. That is a drafting aid, and treating it as anything more is a mistake the number will not protect you from.

No. The score measures how well your CV evidences the requirements one advert named. It does not model any specific employer's applicant tracking system, and no tool that lacks access to those systems can honestly claim otherwise.

It should not. The calculation is deterministic and involves no AI, so the same CV against the same advert produces the same number.

A score can move if you added or dismissed a requirement, if you edited content the CV shows, or if you changed which entries the CV displays — but not on its own.

No, and this is the most common misreading. Each advert sets its own weights, so the two numbers are measured on different scales. Use the score to improve a CV against one advert, not to decide which advert to apply for.

Because a summary is where unearned claims live. Anyone can write "expert in Kubernetes" in a tagline, so crediting it would make the score trivially inflatable — and would reward exactly the CVs that disappoint a human reader.

Move the claim into a bullet that says what you did, and it counts.

The job advert is analysed by a third-party AI provider, and CV content is sent for the tailoring features. The score calculation itself runs locally in ordinary code. The privacy notice is the authoritative document on what is sent where.

Somewhat, and less than you would expect. Repeating a term through a section or packing a bullet with the advert's wording both trigger deductions, precisely because that behaviour produces CVs that fail with human readers.

The reliable way to raise a score is to evidence a requirement you genuinely meet in a place that shows the work.