What does AI look for in an interview? It looks for evidence in your words: a direct answer to the question you were asked, proof of the skill the employer is assessing, a clear structure, your own role, and specifics such as numbers, tools and results that connect to the job. Most AI scoring works from a transcript of what you said, compared with criteria the employer chose for the role, so content and clarity count for far more than polish.
This guide lists the eight signals, explains where the criteria come from, covers what vendors say they don't score and walks through one answer signal by signal. Vendor details were checked on their own sites in October 2026. Take the assessment yourself and follow the employer's rules — use CoPilot Interview to practice beforehand.
What does AI look for in an interview? The 8 signals
| Signal | What the scoring checks | How to show it |
|---|---|---|
| 1. An answer to the question asked | Does your answer respond to this question, not a nearby one? | Put your answer, example or decision in the first sentence. |
| 2. Competency evidence | Does the answer show the skill the question targets, such as problem solving or teamwork? | Pick a story where that skill drove the outcome. |
| 3. Structure | Can the situation, your actions and the result be told apart? | Use STAR, or simply "context, what I did, what happened." |
| 4. Your own role | Is it clear what you did, as opposed to the team? | Say "I" for your actions and "we" only for shared results. |
| 5. Specifics | Are there concrete details: numbers, tools, scale, timeframes? | Add one or two real numbers and name the tools you used. |
| 6. A result and what you learned | Did something change because of what you did? | End with the outcome and what you would repeat or change. |
| 7. Relevance to the job | Does the evidence match skills the job description names? | Use the posting's terms where they are true for you. |
| 8. Enough clear content | Is there enough audible content to evaluate at all? | Speak in full sentences for a minute or two and test your microphone. |
None of these signals rewards length for its own sake, a big vocabulary or a confident tone. They reward an answer a stranger could read once and summarize correctly. If you want the answer structure behind signals 2 to 6, our guide to the STAR method covers it.
Where the scoring criteria come from
The employer chooses what gets scored, usually from the job itself. A few vendors describe this in their own words, as of October 2026:
- HireVue AI Interviewer. The AI suggests questions and answer rubrics based on the job description, and the employer customizes and approves them. Qualification questions are scored as meets or does not meet; skill questions use a 1-to-5 rubric (HireVue AI Interviewer).
- HireVue video assessments. These evaluate what you know about performing the job plus "soft" competencies such as communication, conscientiousness, problem-solving, team orientation and initiative, based on research by organizational psychologists. HireVue says each model is developed for one type of job role and that there is no universal "perfect candidate" you are measured against (HireVue candidate FAQ).
- micro1. Its AI recruiter, Zara, uses open-ended technical questions for your chosen role and scenario-based questions to evaluate how you think, problem-solve and communicate (micro1 AI interview guide).
The practical takeaway: the job description is the best preview you will get of the rubric. Read it, list the skills it repeats, and prepare a real example for each. Our explainer on competency-based interviews shows how those skills turn into questions, and our guide to using job description keywords in answers shows how to map them to your stories.
What AI interview tools say they don't score
Vendors publish very different amounts of detail here. HireVue has published the most. Its 2025 Explainability Statement, still online as of October 2026, says its AI "relies only on what is said by the candidate" and does not assess facial expressions, body language, background and surroundings, or tone of voice (HireVue Explainability Statement). Its candidate FAQ adds that assessments never look at age, race or gender, or at résumé details such as your GPA or the college you attended, and that HireVue never uses your interview video to identify you.
Sapia.ai, which runs text-based chat interviews, says there is no video recording and no camera, that you are assessed only on your written answers, and that no sensitive characteristic such as gender, age or race is used in its AI (Sapia.ai candidate explainer; see our Sapia AI interview guide).
Other vendors may say less, so don't assume the same rules apply everywhere. Read the notice you get before the interview, and ask the recruiter if it doesn't say. Two things still matter even where delivery isn't scored:
- Clarity affects the transcript. If you mumble, talk over the interviewer or get cut off by the timer, the text the AI scores is thinner or wrong. See how AI evaluates interview transcripts.
- People may watch the recording. A recruiter or hiring manager can review your video later, and they notice delivery the way any person would.
For the delivery side, read whether AI interviews analyze tone of voice and whether AI can analyze body language.
Why keyword stuffing doesn't help
It's tempting to pack your answer with words from the job posting. HireVue's FAQ says its video assessment "considers the whole context of a response, not just certain words here or there," and its Explainability Statement says the model reads meaning regardless of the specific vocabulary used. Answers that sound recited or AI-written also tend to hurt: they lack the specific detail rubrics look for, and many scorecard tools flag them. Sapia, for example, says its Chat Interview checks for plagiarism and can detect answers that come from content generators.
Before (example, stuffed with posting words): "I'm a results-driven, data-driven team player with strong stakeholder management and cross-functional collaboration skills, and I always drive impact."
After (example, evidence instead of labels): "Our support and product teams disagreed about which bugs to fix first, so I set up a weekly 20-minute triage where support brought its top five customer issues with ticket counts. Within a month two of those five were fixed, and the product manager asked to keep the meeting."
The second answer never says "cross-functional collaboration," but it shows it, with a role, an action and a result. That is what a rubric row for collaboration describes.
Worked example: one answer, signal by signal
Question: "Tell me about a time you solved a problem with limited information."
Example answer (we wrote this; it is illustrative): "The clearest example is from my job as an operations coordinator at a regional food distributor. One Friday our routing software went down two hours before the trucks left, and the vendor couldn't give us a fix time. I pulled the previous week's routes from our email exports, matched them to that day's orders in a spreadsheet and flagged the eleven stops that were new. I called those customers myself to confirm delivery windows, then handed drivers printed routes. All 14 trucks left within 30 minutes of schedule, and the following Monday I wrote a one-page backup procedure that the team adopted."
| Signal | Where it shows up in the answer |
|---|---|
| Answers the question | The first sentence names the example; the second states the missing information. |
| Competency evidence | Problem solving with incomplete data: rebuilding routes from old exports. |
| Structure | Situation (software down), actions (pulled, matched, flagged, called), result (trucks left). |
| Your own role | "I" for every action; no vague "we handled it." |
| Specifics | Two hours, eleven stops, 14 trucks, 30 minutes. |
| Result and learning | On-time departures plus a backup procedure, which shows follow-through. |
| Relevance to the job | Operations, planning and customer communication, all likely rubric rows for the role. |
| Enough clear content | About 45 seconds spoken, in complete sentences. |
What AI looks for by question type
| Question type | What it's usually checking | What a strong answer includes |
|---|---|---|
| Behavioral ("Tell me about a time…") | Past evidence of a competency | One real story, your actions, a result |
| Situational ("What would you do if…") | Judgment and priorities | Your first step, why, the trade-off and how you would check it worked |
| Technical explanation | Accurate knowledge, explained clearly | A definition, your approach, one trade-off and a short example |
| Motivation ("Why this role?") | Fit with the job and the company | Specific facts about the role linked to your experience |
| Qualification (availability, work authorization, certifications) | Meets or does not meet | A direct, complete fact in one sentence |
How to practice so each signal shows up
- Pick five likely questions from the skills in the job description.
- Answer each one out loud and record it on your phone.
- Check the eight signals against your recording. Look first for two gaps: a missing result, and "we" where "I" belongs.
- Get an outside score. Our free AI interview assistant includes a Mock Interview mode that builds questions from your résumé and the job description, reads them aloud and scores each spoken answer out of 10 with what to fix and a model answer. The free plan gives you 15 minutes every day with no card and no signup.
- Re-answer the lowest score in your own words, then move on. Our guide to what a score out of 10 means explains how to act on each range.
To check your overall readiness first, take the free interview readiness quiz.
See which signals your answers are missing
Answer real interview questions out loud in your browser and get AI feedback on each one, with nothing to install: 145 real interview questions across 10 tracks.
Try the free AI interview demoFAQ
What does AI look for in an interview?
Evidence in what you say: a direct answer to the question, proof of the competency being assessed, a clear structure, your own role, specifics such as numbers and tools, a result, and relevance to the job. Most tools score a transcript against criteria the employer chose for the role.
Does AI look for keywords in interview answers?
Not in the simple sense. HireVue says its assessment considers the whole context of a response rather than certain words. Use the job posting's terms where they are true for you, but show the skill with a real example instead of listing labels.
Does AI judge my appearance or facial expressions?
HireVue says its AI-scored interviews rely only on what you say and do not assess facial expressions, body language, your surroundings or tone of voice. Other vendors publish less, so read the notice you are given or ask the recruiter.
How long should my answers be for AI scoring?
Long enough to show the situation, your actions and a result, which is often a minute or two for behavioral questions. Stay inside the time limit. HireVue says answers with too little content to score are flagged for human review.
Is it better to say I or we in an AI interview?
Use I for the actions you personally took and we only for results the team shared. Scoring looks for your contribution, and an answer told entirely as we makes your role hard to find.
Can I find out what the AI scored me on?
Sometimes. HireVue says employers can switch on a Candidate Feedback Report that gives you insight into how you performed on the assessment. Many other employers share nothing, so ask the recruiter whether feedback is available.