Quick answer: Explainable AI resume screening gives every candidate a score and the reasons behind it: strengths, gaps against the job description, and a short summary. The AI does the reading and the sorting. The recruiter makes the decision. A low score should not end the review on its own, because the reasons tell the recruiter whether the score is right, and a recruiter can still choose to move a low-scoring candidate forward.
Why screening takes so long
Screening is where recruiters lose the most time. In a UK study by Totaljobs (fieldwork June to July 2025), recruiters said CV screening takes 3.6 hours per vacancy. 77% said screening many irrelevant applications slows down hiring.
That is the work AI should take off their desk. Reading every CV in full, one after another, is slow, and it gets less accurate as the pile grows.
What an explainable score shows
Each candidate should come with more than a number. Look for:
- A match verdict, such as excellent, good, fair or weak match
- A score, rounded to the nearest 10
- Three strengths found in the CV
- Three gaps, measured against this job description and not against the person
- A short summary of the CV
The gaps matter most. They tell you what the tool thinks is missing for this role. If the gap is wrong, you can see that in seconds and move on.
Why a low score still needs reading
A score is built from many inputs, and some of them have nothing to do with the candidate’s ability. A thin job description can drag scores down. A career break, a non-traditional background or a CV written in a second language can also score lower than it should.
Take a candidate with a two-year gap after a layoff. The gap alone can pull the score below the line, even when the CV shows the exact tools your team uses every day. Without an explanation, that candidate looks weak. With one, the recruiter sees the strong match on tools and the gap the score flagged, and can decide the gap is fine.
That is why the reasons matter more than the number. If a candidate scores weak but the strengths match what you need, you can see the mismatch and decide to talk to them anyway. Without the reasons, you would never know whether to trust the score.
Explainable tools are weakest in exactly these cases, so read the CV yourself before you rule anyone out.
What the AI should do, and what it should leave alone
The AI should do the manual work: reading, sorting, pulling out the facts and writing a clear summary. It should make that work faster and more consistent from one CV to the next.
The decision should stay with the recruiter. Canvider’s product documentation says it plainly: “Never reject on score alone. It is a ranking aid, and a human must make the decision.”
If a tool quietly moves candidates out of the pipeline, you lose control of who gets a look. Ask for a tool that proposes actions and waits for a person to confirm them.
What the law asks for
If you use an automated tool to screen candidates in New York City, Local Law 144 requires a bias audit done by an independent auditor within the past year. A summary of the results must be public, and candidates must get notice that the tool is used. The city’s rules apply to employers and employment agencies.
The EU AI Act treats AI that filters job applications and evaluates candidates as high risk. Reports say the EU has moved the deadlines for those employment rules to December 2027, so check the final text before you plan around a date.
You do not need to be a lawyer to ask the right questions. Ask any vendor for its bias audit and what the audit covered.
Explainability helps here too. If a tool shows its strengths and gaps for every candidate, you have a record of why each one was or was not moved forward. That record is what you need if a candidate asks why they were not shortlisted, and it is what an auditor will look for.
A checklist for any screening tool
- What is the score built from: the job description, the skills list, or both?
- What does each candidate’s explanation show?
- Can a candidate be rejected automatically, or does a person always decide?
- Which file types and languages does it handle, and where is it weaker?
- Has an independent bias audit been done, and what did it cover?
If the vendor can’t answer all five in plain words, keep looking.
How to start without losing control
Start with one open role, and set up its job description and skills before you run anything. The first-day setup guide covers the company profile and job preferences you need in place. Then pick ten CVs you have already read, and run the analysis on them.
Compare the strengths and gaps with what you saw when you read the CVs yourself. If they match, widen the test to the next role. If they don’t match, fix the job description first, because a thin job description is the most common reason scores drift.
Once the explanations match your own reading, you can trust the tool with more of the sorting. The decision stays with you.
Try Canvider and run the analysis on ten CVs you have already read. Read more about how CriteriaMatch scoring rubrics set what the score measures.
Frequently asked questions
What is explainable AI resume screening?
It is AI screening that shows why each candidate got their score, not just the number. A good tool lists strengths and gaps against the job description, so the recruiter can check them against the CV.
Should an AI score reject candidates automatically?
No. The score is a ranking aid. A person should read the CV before any negative decision, including when the score is low.
Can New York City employers use AI to screen resumes?
Yes, but under Local Law 144 the tool needs a bias audit done within the past year, a public summary of the results, and notice to candidates.