How AI interview feedback is scored: the five categories we use
The rubric behind an AI interview report: what Communication, Structure, Depth, Impact and Confidence each measure, how the overall score and pass rate are derived, and what the score does not mean.
- Five categories are scored 0–100: Communication, Structure, Depth, Impact, Confidence.
- The overall score is the average of the category scores; the pass rate is derived from it, not from a separate judgement.
- Every per-answer entry contains the question, your answer, a score, one or two sentences of feedback, and a stronger rewrite.
- A score is coaching feedback on a transcript. It is not an employer's decision and not a prediction of one.
What each category measures
- Communication — can a listener follow it once, without re-reading: pacing, filler, clarity of the headline.
- Structure — does the answer have a beginning, a decision and an end, or does it wander.
- Depth — is there reasoning and a trade-off, or only a description of what happened.
- Impact — is the result stated and quantified; does the answer say why it mattered.
- Confidence — ownership language (“I decided”) versus hedging (“we sort of tried”).
How the numbers are produced
Each answer is scored against the transcript, then the categories are aggregated. Numbers come from the model's reading of your words, which is why two answers with the same content but different specificity can score ten points apart.
Because it is transcript-based, delivery is only visible where it shows up in text: filler words, abandoned sentences, and over-long answers. Eye contact and presence are outside what the score can see.
How to read a low score
Look at which category is lowest, then at the per-answer feedback for the answers that scored worst. The useful question is not “why is my score low” but “which single part is missing from most of my answers” — usually the quantified result, or the decision you made.
Questions people also ask
Why did I score zero?
A zero means the transcript contained no assessable answer — greetings, one-word replies, or answers that never engaged with the question. It is a reading of the transcript, not an error in the scoring.
Is a score of 80 good?
Eighty means the answer has structure, a decision and a quantified result, with little hedging. It is not a hiring prediction: the same answer can still fail a specific company's bar for reasons a transcript cannot show.
Can the score be gamed?
Only in the sense that adding numbers and ownership language improves it — which is also what improves the real answer. Padding with invented metrics is the one thing that backfires: the follow-up question exposes it.
Answer this in the coach, and get it scored.
Novue runs a live mock interview, scores five categories, and rewrites each of your answers in your own words. Free during the beta.
Start practising