AI interview services: nine compared (October 2026)

Conclusion first

What this article comes down to

Start here

What an AI interview is

An interview the candidate takes in conversation with an AI. They open a link and sit it without waiting on an interviewer’s calendar. The AI asks against criteria defined in advance, follows up on the answers, and leaves a score with the reasoning behind it. People decide who is hired.

Counting the services that appear across the comparison articles, more than twenty are available in Japan. They all line up much the same features, so the work of comparing grows with the number.

Background

Why AI interviews are spreading

The problems vendors name are much the same. There are two.

AI RECOMEN, SHaiN, TA Team and Z Career AI Interviewer all name these two as what they set out to solve.

There is a candidate side too. While they wait for an interviewer’s diary to open, another company’s process moves ahead.

The answer is much the same everywhere as well. Make the interview available 24 hours a day, 365 days a year and run any number of them in parallel. Nothing waits on a diary any more, so the load and the waiting fall together. Up to here, every service says the same thing.

What changes

What adoption actually changes

The more applicants you have, the more it pays. For a handful of hires a year, the design work may not earn its keep.

Caveats

Three things to settle before you start

The problem

Why comparing them settles nothing

The feature grids fill with the same rows, most vendors keep pricing off the site, and “you can tailor the evaluation criteria” appears on every one. There is more to compare and less to decide on.

One reason is the axis. Most comparison articles sort AI interview services by how much the AI takes over: conversational, one-way recorded, or support for recording and analysing interviews. That axis tells you what the AI does. It does not tell you who owns the evaluation criteria. When the owner changes, so does how much you can adjust, and what you are able to do a year in.

Classification

Nine services, four types

So we sorted them by what each service optimises for, which gives four types.

The four types, and the five layers behind them, are set out in how to choose an AI interview tool. Only what the calls rest on is repeated here. Type is what a service optimises for, and splits into the four above. On top of that we read three things: the questions, the criteria and the verification.

We do not rank them. Which type is right depends on what you want to optimise. Our own Proomi occupies one of the four, no more.

The table

Where nine services fall

Based on public information from each vendor, checked on 1 October 2026. We also looked at five more — Calaris Assess, harutaka, TG-WEB mee, NALYSYS AI Interview and AI RECOMEN — but each one repeats a call already made within the same type, so we left them out of the table. Their sources are listed at the end.

Choosing between them

What each one optimises for

01Shared-framework type

The vendor supplies the evaluation frame. PeopleX lets you author your own questions on top of its 150 presets, and evaluation changes the “weighting” of each item, with the items themselves supplied. SHaiN (Talent and Assessment) is built on its “strategic hiring method”: evaluation draws on 18 supplied items, while up to 20 free questions can be added. TA Team (VARIETAS) also supplies its 20 observation points and reweights them against the role profile; a client release states that VARIETAS “develops the interview design to match the items the company wants to evaluate”. Scope of customisation has to be read separately for questions and for criteria. Two services in this type verify. PeopleX’s insight feature “checks whether the selection was sound and suggests the questions and evaluation content for the next round”; TA Team runs as five AI agents, of which an audit agent “monitors the integrity of hiring continuously”, setting its practice out as a method that proposes “what to change in the next round”.

02Low-cost / flat-rate type

MiAI (Props) charges from 1,000 yen per interview, a level at which you need not hesitate over volume. Our AI Mensetsu (JetB) is a flat-rate service that states “monthly fee only, no usage charges”, so the bill does not move with the number of interviews.

03Sourcing / pool type

CaseMatch also offers a scouting product and uses interview-practice history to build the candidate pool. Z Career AI Interviewer (ROXX) began in October 2024 as a feature for companies using Z Career, a job platform for non-desk workers, and opened to standalone use in January 2025. Its release states it can be used “not only with Z Career but also for interviews arising from recommendations through recruitment agencies”. Running interviews on candidates the platform gathered is where it starts.

04Criteria-owned type

Here the evaluation criteria become yours. With both Track and Proomi, the criteria are built from your own requirements. What differs is who designs them and who corrects them.

Proomi (our own service) has the customer define the criteria, and the check on whether those criteria were right also runs on the customer’s data. The score distribution per criterion shows where the AI is separating candidates, matching against hire/no-hire results shows which criteria actually drove the decision, and following retention afterwards shows which ones held up. When the AI and the interviewer disagree, you change the definition yourself rather than waiting for a model update. Evaluation uses only what the candidate says, not facial expression, tone of voice, or speaking pace. Because the evidence sits in what was said, you can trace why a criterion scored as it did on your own data.

Track AI Interview (Givery) “designs the evaluation items, scoring criteria and question flow individually” and states that it “improves them periodically in operation, optimised to your criteria”. The criteria are described as yours, while the design work and the corrections after go-live sit with Givery.

When you look at this type, ask who does the work of writing the criteria and who does the work of correcting them once you are live. A service that assumes you will define and improve everything in-house and one where the vendor stays through design and operation mean very different workloads, even when both say “customisable”.

Even for high-volume graduate intakes  when you define the criteria yourself, the competency model and behavioural definitions you already hold go straight onto the AI interview. Scoring can be written as a rubric, so the yardstick is the one your human interviewers use.

The criteria used in selection then carry over to evaluation after joining. Put the same format to the people already inside and you can see from real data which criteria the strong performers score on. Set that against retention and you learn whether that year’s criteria were right, and next year’s can be rewritten in-house without waiting for the vendor to update a model. The larger the intake, the faster that loop turns.

What decides it

Three questions to ask in the meeting

“Can we adapt the evaluation criteria to our company?” no longer separates the types. All nine services claim customisation of some kind, and even one at 1,000 yen per interview writes that you may design everything yourself. The difference sits one level finer: whether you define the evaluation items themselves or reweight the ones supplied, and how far back you can check that the evaluation was right.

Verification is not a yes or no either. Seven of the nine say something about it, and they reach different distances: the distribution across candidates, the design of the next round, retention after joining. One of the nine writes that it reaches as far as retention. These three questions bring that out.

Whichever of the three you ask, when the answer is yes, ask one more thing. Is that feature running today? Some services note that their feature list includes items in development, and public information does not tell you what has shipped.

The remaining three questions are in how to choose an AI interview tool.

FAQ

Frequently asked questions

What is an AI interview

An interview the candidate takes in conversation with an AI. They open a link and sit it without waiting on an interviewer’s calendar. The AI asks against criteria defined in advance, follows up on the answers, and leaves a score with the reasoning behind it.

Does the AI decide who is hired

People do. Across vendors the stated arrangement is the same: the AI goes as far as the score and the reasoning. How much is handed to it varies, so it is worth asking during selection.

If it says “customisable”, is it the criteria-owned type

No. All nine claim customisation of some kind. What differs is who decides. For questions: whether the service designs them from your own criteria, or you write them yourself as free text. For criteria: whether you pick from supplied items, reweight supplied items, have the service design them from your own criteria, or define them yourself as free text. The table uses those words. What is left after that is whether you can check the criteria were right.

Are there services that sit across two types

Yes. Some services offer both introductions from a candidate pool and an AI interview designed on the customer’s criteria. Each call follows what the vendor puts forward publicly, and where a service sits across two types the table names both.

Can current employees sit the same evaluation as candidates

Across the nine services here, we found no statement that the people already inside can be measured in the same format. When the same format reaches them, you can see from real data which criteria the people who do well actually score on. This is where the depth of verification shows itself.

Can the same criteria run across roles and languages

It depends on whether you can hold separate criteria per role, and run English and Japanese without maintaining two standards. Where movement is limited to a standard frame or template, the operation gets heavier as the number of roles grows.

What if we cannot write the criteria ourselves?

Choosing the criteria-owned type does not require you to have them written. We can build them with you, from the definitions your performance reviews already use and from what your interviewers actually look at. What matters is that the finished criteria belong to you and that you can rewrite them afterwards. Who writes the first draft is not what separates the types.

In short

What this article establishes

About this article

Written by Prance Holdings, which makes an AI interview service called Proomi. We include ourselves in the comparison and do not rank anyone. The table rests on what each vendor publishes; we have not verified what is already shipped.

Sources

Public information used for the calls