To score your interview answers with AI, give the chatbot a rubric instead of asking "how did I do?", make it quote the words that earned or lost each point, and score each answer more than once, because AI scores drift from run to run. Use the scores to find patterns across many answers. What the AI can't judge is whether your story is true and whether it's the best one you have; that stays with you.
This page gives you a rubric to paste, a calibration step that keeps the scores honest, five prompts, a scored example and a simple tracking sheet. For what an AI score measures in the first place, see how AI interview assistants score answers.
What an AI score can and can't tell you
| What you want to know | Can an AI score tell you? | Better check |
|---|---|---|
| Did I answer the question asked? | Yes, mostly | Read your first sentence next to the question |
| Did I give specific evidence? | Yes, if you make it quote the evidence | Count your numbers and named outcomes |
| Is this the right story for the job? | Partly, by comparing it with the job description | A mentor or someone in the field |
| Is my technical claim correct? | Sometimes; it can be confidently wrong | Documentation or a worked example |
| How did I sound? | No, not from a transcript | Your recording, or a friend |
| Will I get the job? | No | Nothing can predict that from one answer |
A rubric you can paste into any chatbot
We wrote this rubric for behavioral and general questions. Each criterion scores 0, 1 or 2, for a total out of 10.
| Criterion | 0 | 1 | 2 |
|---|---|---|---|
| Direct answer | The question is never answered | Answered late, after setup | The first sentence answers it |
| Personal action | Only "we" | Some "I", mostly team | Clear actions you took and decisions you made |
| Evidence | Claims only ("I'm organised") | A general example | A specific example with a number or named outcome |
| Result and reflection | No outcome | An outcome without a lesson | A clear outcome, plus what you'd repeat or change |
| Role fit | No link to this job | An implied link | A link to a named requirement in the job description |
For technical answers, swap "Role fit" for "Correctness" and check that line yourself. Length isn't in the rubric because a chatbot can't time you; use a stopwatch.
The scoring workflow
- Score spoken answers. Use a transcript of what you said, not a polished written version you'd never say in the room.
- Paste the rubric and ask for quotes (prompt 1). A score without quoted evidence is an opinion.
- Calibrate first (prompt 2). Give it one deliberately weak and one strong answer to the same question. If they land within a couple of points of each other, the rubric is too loose or the AI is being polite; tighten the wording.
- Score three times. Run the same answer in three fresh chats, or ask for three independent scorings, and keep the middle one. Treat a one-point change as noise.
- Take one fix, tied to the lowest criterion.
- Re-answer out loud and re-score, then record the result in your tracking sheet.
- Every five answers, look for patterns (prompt 5).
Five scoring prompts
1. Rubric scorer
Act as a strict interviewer for a [job title] role. Score my answer to "[question]" with this rubric: [paste rubric]. For each criterion give 0, 1 or 2 and quote the exact words from my answer that justify it; if nothing does, write "no evidence". Then give the total out of 10 and one fix tied to the lowest criterion. Don't rewrite my answer. Job description: [paste]. My answer (transcript): [paste]
Why it works: the quote requirement makes vague praise impossible and shows you exactly which words earned the point.
2. Calibration
Before scoring my answers, score these two answers to the same question with the same rubric. Answer A: [deliberately weak version]. Answer B: [strong version]. Explain the gap. If the gap is less than four points, tell me which rubric lines are too vague.
Why it works: if the grader can't separate a bad answer from a good one, its scores on yours mean little.
3. Blind comparison
Here are two versions of my answer to "[question]", labelled X and Y in random order. Score both with the rubric and tell me which one an interviewer would remember a day later, and why.
Why it works: hiding which version you prefer removes the pull towards agreeing with you.
4. Evidence audit
List every claim I make in this answer about myself. For each, say whether I backed it with a specific example, a number, a named outcome or nothing. Answer: [paste]
Why it works: unsupported claims are the most common reason a reasonable-sounding answer scores low.
5. Patterns across answers
Here are my last [number] scored answers with their criterion scores: [paste]. Which criterion is lowest on average, which question types score lowest, and what one habit would raise the most scores at once?
Why it works: one habit, such as stating the result, usually lifts several answers at the same time.
Worked example: a 2 that became an 8
Question: "Tell me about a time you disagreed with your manager." The candidate is applying for a QA engineer role whose job description stresses release quality. An example we wrote.
First answer (transcript)
"I think disagreements are healthy and I'm always open about my opinions. At my last job my manager wanted to launch a feature before it was fully tested and I didn't agree. We talked about it and eventually came to a compromise. I learned that communication is really important and that you have to respect your manager's decisions even if you don't agree. It worked out fine in the end."
The AI's scores (prompt 1, middle of three runs)
"Direct answer 1: the disagreement arrives in sentence two. Personal action 0: only 'we talked about it'. Evidence 0: nothing about what you said or did. Result and reflection 1: 'worked out fine' is vague and the lesson is generic. Role fit 0. Total 2/10. Fix: say what you did to make your case."
Re-take (transcript)
"My manager wanted to ship a billing feature a week early, and I disagreed because our tests didn't cover annual plans yet. Instead of just objecting, I listed the three missing test cases, estimated two days to cover them and suggested releasing to monthly customers first. She agreed to the split. In those two days we found a proration bug in annual plans and fixed it before any annual customer saw it. Since then I bring a smaller option with any objection."
Re-scored: direct answer 2, personal action 2, evidence 2, result and reflection 2, role fit 0, because the answer never ties back to release quality in so many words. Total 8/10. The re-take isn't the AI's writing: the candidate answered the fix from memory, out loud, and the new transcript was scored.
Track scores without chasing them
| Date | Question | Middle score of 3 | Lowest criterion | Fix for next time |
|---|---|---|---|---|
| Mon | Disagreed with manager | 2 | Personal action | Say what I did to make the case |
| Wed | Disagreed with manager | 8 | Role fit | Name release quality from the job description |
| Wed | Biggest failure | 5 | Result and reflection | Add what changed afterwards |
A rising score on the same question means the fix stuck. A flat score after three re-takes means change the story, not the wording. Don't game the rubric: stuffing an answer with numbers raises "evidence" and lowers believability. For what a 5, a 7 and a 9 tend to look like, see what the mock interview score out of ten means.
Check the AI's scores
- Generous grading. If everything scores 8 or more, ask it to grade as a strict interviewer and to say what a 2 would need on each line.
- Quotes that aren't there. Check every quote appears in your answer. A misquote means that score was made up.
- Scoring the AI's own rewrite. Never score a version the AI wrote. You'd be grading the AI.
- Technical judgements. A confident "correct" on a technical point isn't proof. Verify it.
- Privacy. Remove employer secrets and other people's names before pasting answers.
Before a big interview, add one human opinion to the AI's scores; our guide on how to get feedback on interview answers lists where to find one. For where scoring fits in a full prep routine, see how to use AI for interview prep.
The practice loop: answer, score, re-take
Scores are only useful if they change what you say next. A daily loop of 15 minutes: answer two questions out loud with a stopwatch, transcribe or paste them, score each with prompt 1, re-take the lower one with its fix and log both. Employers use their own rubrics, which you won't see, so the aim isn't to match a hidden scorecard; it's to make every answer direct, personal and backed by evidence. Our page on how AI evaluates interview transcripts explains why those three qualities matter even when a person, not a model, is doing the scoring.
Where CoPilot Interview fits
CoPilot Interview is an AI interview assistant that does the scoring loop for you. Its AI mock interview reads each question aloud and scores each spoken answer out of 10, and the free practice demo scores typed answers out of 10 with strengths, areas to improve and a follow-up question, free and with no signup. Running the same answer through your own rubric and through the demo is a quick way to see whether the two agree. The free plan gives 15 minutes a day of live help in sessions of up to 15 minutes, with AI mock interviews included. Paid plans are Standard at $24.99 a month and Pro at $39.99 a month.
The desktop app, available as a desktop app for Windows 10 and 11 and macOS 12 or later, a free iPhone and iPad companion on the App Store (iOS 16 or later) and an Android companion in beta, also offers live help in real interviews. If you use it, follow your employer's rules, and practise instead for online assessments, proctored tests and AI-led rounds, which typically forbid outside help.
Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.
Frequently asked questions
Can ChatGPT grade my interview answers?
Yes, reasonably well for structure and evidence, if you give it a rubric, ask it to quote the words behind each score and score each answer more than once. It can't judge your delivery from text or tell you whether a specific employer will like the answer.
Why does AI give different scores for the same answer?
AI grading isn't perfectly consistent: the same answer can land a point or two apart on different runs. Score three times, keep the middle score and react to patterns across several answers rather than to one number.
What is a good score for an interview answer?
On the 10-point rubric above, a 7 or 8 means a direct answer, clear personal actions and a specific result. That's a strong answer; chasing a 10 on every question isn't worth the time.
Do employers score interviews the same way?
Not exactly. Many employers rate answers against their own scorecards and standards, which candidates don't see. A self-scoring rubric is a practice tool that rewards the same basics: answering the question, showing what you did and backing it up.
Try a practice question
The browser demo has 145 real interview questions across 10 tracks. Type your answer and get a score out of 10, strengths, areas to improve and a follow-up question. Free, no signup.
Try a practice question