Yeokgeom (역검): AI competency assessment video interviews, and how to prepare

역검 is a two-syllable contraction of AI 역량검사 — 역 from 역량, competency, and 검 from 검사, assessment. Korean job seekers type it romanized, as yeokgeom, when they search in English, and English-language pages that actually define the word are thin on the ground. What sits behind the word is an assessment the employer licenses and sends out at a screening stage.

The gap with an AI interview is scope. 'AI interview' is a loose English label for any hiring round software conducts or scores, whichever vendor an employer happens to use. Yeokgeom names one Korean assessment family, and the video answers are only one of its parts; an invitation can also carry a personality section and strategy games, with the employer's configuration deciding which of them you sit. Practice tools sit on the other side of that line: you pick them, repeat them, and stop when you want, while the 역검 sitting that counts comes with a deadline and a result that goes to the company.

How it is scored stays with the vendor and the employer — within the published guidance there is no weighting table and no pass mark. The part that answers to preparation is the video interview, and that is what the rest of this page is about: which experiences you bring, and the order you tell them in.

The usual format is one prompt on screen, a short window to think, and a fixed window to answer out loud to the camera. How many prompts you get, how long each window is, and whether re-recording is allowed vary by employer and by assessment. The rules in your invitation email and the pre-check screen are the authoritative version, so read those before you build a practice routine around any assumption. MIDAS Group, which built the assessment many Korean employers run, said in December 2025 that more than 1,200 companies and institutions now use it — at that scale, the details differ.

The prompts are mostly about experience. Some describe a situation and ask what you would do; others ask you to recall something that actually happened. The decisive difference from a human interview is that no one asks a follow-up, so if you spend the whole window setting up the situation, the answer that remains has no account of what you actually did.

Scoring is not published. This guide does not know it either, and material claiming otherwise is worth discounting. What a candidate controls is what they say and in what order — and that part reliably improves with practice.

Four things repay preparation: the set of experiences you will draw on, the three-beat shape you pour them into, a feel for how much fits into a minute, and delivery in front of a camera — pace, filler words, where you look. All four get reused no matter which prompt appears, which makes them the highest-return place to spend practice time.

Hunting for the right answer to game tasks or personality items is a different matter, and not one this guide takes up. Playing a character you are not tends to produce answers that contradict each other, and it also drifts away from the real experiences you describe in the video section. The vendor says the same: when MIDAS IN opened its official tutorial, it explained that the assessment is not a test of correct answers but a read on how someone judges and acts while working toward one.

Nor is complete preparation a thing that exists. Assessment formats change and the prompts are not published. What you can do is arrive with reusable material and a reusable shape, so that even facing an unfamiliar prompt you can choose an experience and order it within the thinking window. That is usually where the visible difference comes from.

Each card has five fields: the situation in one line, what you personally did, the result (with a number when you have one), how you felt at the time, and what you learned. Choose experiences along different axes so they do not overlap — a disagreement, a failure, a deadline crunch, something you did for the first time, a time you persuaded someone, a time the standard approach did not work and you changed it. That set covers most prompts.

One experience can serve several prompts. The same project can be told as a conflict story or a failure story, so tag each card with the angles it can be pulled from. If choosing a card takes only a few seconds, the rest of your thinking window goes into sharpening the opening sentence.

Do not leave the feelings field blank. Video prompts often ask directly how you felt, and without that field the answer sounds like a status report. It does not have to be a flattering feeling either — anxious, aggrieved, self-conscious: those words usually explain the action that followed better than anything else you could say.

For a one-minute prompt, roughly ten seconds of situation, fifteen on how you felt, twenty on what you did differently, and fifteen on the result and the lesson works well. Open with the conclusion. A single line like 'The time our team split on a decision and I proposed we agree on criteria first' lets the listener follow whatever comes next. HireVue, whose platform runs many recorded interview rounds, states the constraint plainly in its candidate guidance: “Most on-demand interviews ask you to give your answer in just a few minutes. It's helpful to think about how you will illustrate your skills clearly in that timeframe.”

The beat candidates most often leave empty is what they did differently. Working hard does not fill it. 'Most people were heading toward A; I chose B, because…' gives the judgment something to be measured against. This is also where you separate what the team did from what you did yourself.

Do not end on a number — add one sentence about the lesson. 'I did X and the result was Y, so since then I check Z first' is short and lands hard. A lesson that shows up in later behavior is what makes an experience read as growth, and when time runs short, keeping that sentence and trimming the setup is the better trade.

Recording yourself on a phone and watching it back is the fastest method. Three to five passes on the same prompt is usually enough for the sentences to settle. Avoid memorizing a script, though — the moment you reach for a memorized line your eyes drift upward and the delivery turns into reading. Memorize the three keywords and rebuild the sentences each time.

When you review, check only three things: whether the first sentence is the conclusion, how often fillers appear, and whether the pace is comfortable to your own ear. Look at the lens rather than at your own face on the screen — that is what reads as eye contact on the other side.

Sorting out the environment removes a whole class of accidents. Lighting that does not shadow your face, a quiet room, the most stable network you can get, and a permissions check for browser, microphone, and camera. HireVue's candidate checklist is blunt about where the light goes: “Make sure it's in front of you—lighting from behind makes it hard to see you.” A single rehearsal under the real conditions takes most of the surprise out of the day itself.

For sharpening the answers themselves, yes. In Preterview, three AI interviewers who have read your resume, portfolio, and the job posting ask questions by voice, and when you answer out loud they follow up on what you just said. Because the flow drills into one experience through feeling, differentiation, and result, you come out with material you can reuse directly in a video interview.

Each session ends with a report: scores across five competencies, delivery analysis covering pace and filler words, and per-answer notes on what was thin and how it could have been said. The thing hardest to judge when you record yourself alone is whether an answer was specific enough, and that is precisely what the report addresses answer by answer.

One boundary worth stating plainly. Preterview does not reproduce any particular employer's assessment and does not predict its score, and it does not touch game tasks or personality items. What it trains is the content, structure, and delivery of the answers you will speak. A portfolio review is free once a day, so you can check the quality of the output first, while a full mock interview runs on purchased passes.