Choose an AI interview tool that can explain its scores

Why scoring

Can you explain the AI’s scores to your colleagues?

If you roll it out without being able to explain its scores, nobody is sure how to use the scores, and staff may end up re-interviewing candidates themselves.

Japan’s AI Guidelines for Business (version 1.2, March 2026), issued by the Ministry of Internal Affairs and Communications and the Ministry of Economy, Trade and Industry, ask organisations not to leave judgements to AI alone but to bring in human judgement at the right moments. They also call for the people affected to be told that AI is being used. The Personal Information Protection Commission likewise asks organisations that analyse a person’s behaviour or interests (so-called profiling) to state that analysis in their purpose of use. The same thinking is a useful reference when AI evaluates interview records.

Both assume that the organisation itself can explain what the AI is doing. Some AI interview vendors overseas publish documents explaining how their scoring works. It is fair to ask the same when choosing a service in Japan.

Sources: Ministry of Internal Affairs and Communications and Ministry of Economy, Trade and Industry, AI Guidelines for Business (version 1.2), March 2026 (in Japanese) https://www.soumu.go.jp/main_content/001064279.pdf; Personal Information Protection Commission, Q&A on the Guidelines for the Act on the Protection of Personal Information, Q2-1 (in Japanese) https://www.ppc.go.jp/personalinfo/faq/APPI_QA/

Question 1

What does it base its scores on?

The first question is what input the AI uses. Whether it is only what the candidate says, or also facial expressions and tone of voice, makes a large difference to whether the reason for a score can be put into words.

Proomi transcribes what the candidate says and scores from that transcript. It does not summarise the interview first; scoring works directly from the transcript. Facial expressions, tone of voice and speaking speed are not used for scoring.

Research has raised doubts about reading emotions from faces. In a 2019 paper drawing on a large body of studies, Barrett and colleagues showed that people express the same emotion with different faces, and that the same expression can mean different things in different situations. The EU AI Act prohibits AI that infers people’s emotions in the workplace, and guidelines the EU published in 2025 include candidates in a selection process within that workplace.

If only what was said is used, the reason for a score can be explained in the candidate’s own words.

Sources: Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements. Psychological Science in the Public Interest, 20(1), 1–68. https://doi.org/10.1177/1529100619832930; Regulation (EU) 2024/1689 (AI Act), Article 5(1)(f) https://eur-lex.europa.eu/eli/reg/2024/1689/oj; European Commission, Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689 (AI Act), February 2025, paragraph 254 https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-prohibited-artificial-intelligence-ai-practices-defined-ai-act

Question 2

What standard does it score against?

Next is the standard behind the scores. If the service supplies the standard, ask the provider to explain what it contains and how scores are assigned; that makes it easier to explain internally.

In Proomi, you write the evaluation criteria in your own words rather than choosing them from a list. Beyond the name of each criterion, you can define what an answer at each score level looks like.

Level definitions are easier for both the AI and human interviewers to apply when they contrast what a strong candidate says and does with what a weak one says and does. Here is an example.

Criteria written as impressions, such as “shows initiative”, push both the AI and interviewers towards safe middle scores. In another article, we re-scored the same persona interviews after adding level definitions to the criteria, and the gaps between candidates widened. See “Rewrite your criteria, then test them on past interviews”.

Question 3

What happens when there is no evidence?

Next, check how it handles a criterion the candidate never spoke to. Guessing a score there leaves nobody sure what the score means.

For each criterion, Proomi writes a score and a reason and quotes the statements it relied on. A criterion with no supporting statement can be recorded as “no evidence”.

A criterion marked “no evidence” suggests that the interview may not have included enough questions to test it. If a criterion is a concern for a particular candidate, we expect it to be checked in a follow-up interview, by the AI or by an interviewer. If the same criterion keeps coming back with no evidence, that is the cue to add questions or rewrite the criterion. For how to add questions, see “Use your performance review criteria in interviews, word for word”; for how to interpret skewed scores, see “More criteria do not mean more evidence”.

Tying the evidence to statements also helps when explaining things to candidates. Proomi can send candidates feedback written for them; by default it omits the pass/fail result and scores, and instead points to where their strengths showed in what they said. The company decides whether to send feedback and what it contains.

Question 4

Does it present scores in a form people can act on?

Last is whether the AI’s scores arrive in a form people can use straight away. If only a number comes back, hiring staff have to check from scratch whether that number is enough to move someone forward.

Proomi gives an overall score and a score for each criterion, each with its reason and quotes of the statements behind it, and the recording and transcript sit on the same screen. Whether you move high-scoring candidates straight to the next interview or reread the records only for those you are unsure about, you can show the reason for the decision in the candidate’s own words.

Hiring staff record the pass/fail decision, and each company sets its own pass mark for moving candidates forward.

Using the records

With records kept, you can check your standard later

Because the scoring is visible, the records can feed directly into improving your criteria.

When you rewrite your criteria, you can re-score past interviews against the new criteria and see how the scores change. We can do this work with you.

When you compare the records with pass/fail decisions and post-hire retention, you can also see which criteria were linked to real decisions and performance. For the approach, see “Was your interview evaluation right? You find out after the hire”.

What to compare

What to ask when choosing a service

A service that can answer these four questions lets you explain to colleagues what the scores mean once you adopt it.

Q&A

Frequently asked questions

Do you use facial expressions or tone of voice?

No. Scoring uses the transcript of what the candidate said. The recording is kept, so hiring staff can watch it again later.

How much of the criteria and level definitions can we write ourselves?

Both the names of the criteria and the definition of each level can be written freely, and you can use the criteria and definitions from your performance reviews word for word. If you have not settled them yet, we can build them with you from your current evaluation sheet.

Can we use the AI’s scores to decide who moves to the next stage?

Yes. Each overall and per-criterion score comes with its reason and the statements behind it, so you can decide from the scores who moves to the next interview. Hiring staff record the pass/fail decision.

Related pages

How it applies, by situation

Once you have checked how the scoring works, these pages cover choosing a service and putting it to use. If you are at the stage of comparing services, the comparison of the leading services, sorted by what each one optimises for, may also help.