How AI Scores a CV: A Look at the Ranking Breakdown

AI scores a CV by breaking the match into separate factors - skills, experience relevance, seniority, and education - scoring each against the job description, then combining them into a single weighted number out of 100. A good tool shows you that breakdown and the reasoning behind it, so the score is something you can check rather than just trust.

Here's what goes into the score.

The factors behind the score

Most AI screeners weight a few core dimensions. In SwiftShortlist, the composite score is built from:

• Skills match (35%) - how well the candidate's skills line up with the must-haves in the job description.
• Experience relevance (30%) - not just years, but whether the experience is relevant to this role.
• Seniority (15%) - is the candidate at the right level - not under- or over-qualified?
• AI specialisation (15%) - depth in any AI/ML requirements the role calls for (neutral if the role doesn't need it).
• Education (5%) - relevant qualifications, weighted lightly because experience usually matters more.

Each sub-score is calculated independently from 0–100, then combined by those weights into the final score.

Why a breakdown matters

A single number tells you what but not why. The breakdown lets you see, for example, that a candidate scored 82 because of an excellent skills match (95) despite slightly light experience (70). That's a very different candidate from one who scored 82 on the strength of seniority alone - and the breakdown makes the difference visible.

Beyond the number

The score is a ranking signal, not a verdict. Alongside it, a good tool surfaces:

• Strengths - what makes this candidate compelling.
• Gaps - where they fall short of the requirements.
• Risk factors - things to probe, like employment gaps or job-hopping.
• Interview-fit verdict - a holistic read beyond the raw score.

Together these turn a score into a brief you can act on.

How to use it well

Read the reasoning before trusting the number. If a score looks surprising, the breakdown usually explains it - and if it doesn't match your judgement, that's a signal to sharpen your job description. The score gets you to the right candidates faster; the decision stays yours.

A worked example

Take a mid-level backend developer role whose must-haves are Python, PostgreSQL and three or more years building APIs. Two candidates both land at 78.

Factor (weight)	Candidate A	Candidate B
Skills match (35%)	94	72
Experience relevance (30%)	70	90
Seniority (15%)	76	80
AI specialisation (15%)	60	60
Education (5%)	74	96
Weighted score	78	78

Same number, different people. Candidate A has every skill on the list but less directly relevant experience: a strong learner who may need ramp-up time. Candidate B has done almost exactly this job before but is missing one of the named skills. Which one you interview first depends on whether this role needs someone productive in week one or someone you can grow. The breakdown is what lets you make that call; the single score never would.

What the score does not measure

• Things the CV does not say. Communication, judgement and how someone works in a team mostly show up in interviews, not CVs.
• Potential. Career changers and people early in their careers can score lower on experience relevance while being excellent hires. Read below the top of the list for roles where trainability matters.
• Your context. The model scores against the job description. If something matters to you but is not written down, it is not in the score.

Why scores can differ from your gut feeling

When a score surprises you, the cause is usually one of three things: a must-have that is missing from the job description, a must-have written so vaguely that it matches too many CVs, or experience described in words you did not expect. Fixing the job description and re-running fixes most of these. Our guide on writing a job description for AI screening shows how.

Frequently asked questions

Is a score of 70 good? Scores only mean something relative to the other applicants for the same role. On a role with strong applicants, 70 might be mid-table; on a hard-to-fill role it might be your best candidate. Look at the order and the reasoning, not a fixed pass mark.

Can a candidate game the score with keywords? Much less than with a keyword filter. The model reads for meaning and evidence, so a list of buzzwords with no experience behind them scores poorly on experience relevance.

Should I reject everyone below a certain score? No. Use the score to decide reading order. A person should make every rejection, and UK GDPR gives candidates rights around decisions made solely by automated means. See the AI CV screening GDPR guide.

See the complete guide to AI CV screening for the bigger picture. Related reading: the JD side of the equation - how to write a job description for AI screening - and the vendor-selection lens - how to choose AI recruiting tools. Or try SwiftShortlist free and see a real breakdown on your own CVs.