A useful AI interview summary has seven parts: a verdict for a human to confirm or overturn, section scores against the competencies, two or three evidence quotes per section, red and green flags phrased as hypotheses, the questions that went unanswered, logistics, and the next steps. Anything else is padding, and a summary that is mostly padding is a summary nobody reads. This page sets out that anatomy, the ways AI summaries go wrong, and a five-minute review routine before a summary reaches a hiring committee.
The seven parts
| Part | What it contains | Why it is there |
|---|---|---|
| Verdict | Hire / no-hire / more information, with one reason | The committee needs the claim before the evidence |
| Section scores | A score per competency on the agreed scale | Comparable across candidates |
| Evidence quotes | Two or three per competency, verbatim | Lets readers judge the answer, not the summary |
| Flags | Red and green, as hypotheses to validate | Directs the next round |
| Unanswered questions | Competencies not covered or answers that dodged | Prevents false confidence |
| Logistics | Availability, compensation, authorization | Kept separate from evaluation |
| Next steps | Who probes what, by when | Turns the summary into action |
Interviewer Mode's report follows this shape: section scores, red and green flags, notable quotes and a verdict to debate. The committee-facing version of the same document is covered in how to write an interview scorecard.
How AI summaries go wrong
Most failures share one root cause: the summary was generated from a transcript nobody opened again. A summary that cites the transcript line by line is harder to get wrong and much faster to check, which is why the quotes matter more than the prose.
- Invented specifics. A number or a technology the candidate never mentioned. Quotes guard against this; summaries without quotes cannot be checked.
- Score without evidence. A 7 with no quote is an opinion the tool had.
- Flattened nuance. "Strong communicator" for a candidate who was clear on one topic and lost on another.
- Transcription errors propagated. "Cash" for "cache" becomes a flag about finance.
- Logistics scored. A notice period or a salary ask treated as a weakness.
The five-minute review
- Read the verdict, then check that at least one quote per section supports the score.
- Delete any specific you cannot find in the transcript.
- Reword flags into questions for the next interviewer.
- Move anything about health, family or protected characteristics out of the document entirely.
- Add your own one-line impression, labeled as yours.
The last step matters for candidates too: the summary is often the basis of the feedback they receive, and good candidate feedback needs evidence, not adjectives.
Retention and candidate access
Treat summaries as hiring records: keep them for your standard retention period, restrict access to the hiring team, and be ready for access requests, which in several jurisdictions give candidates a right to see the evaluation. A summary written from quotes and competencies reads fine under that scrutiny; one written from impressions does not. Candidates increasingly know what these documents contain; see AI interview scorecards explained.
Where CoPilot Interview fits
CoPilot Interview is the AI interview assistant candidates use to practice; Interviewer Mode is the same desktop app on the hiring side, producing a report with section scores, red and green flags, notable quotes and a verdict the minute the call ends, free on Windows and macOS with no signup. Details are at Interviewer Mode. Review every summary before it travels, and follow your employer's rules on retention and candidate access.
Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.
Frequently asked questions
What should an AI interview summary contain?
A verdict, section scores per competency, evidence quotes, flags as hypotheses, unanswered questions, logistics kept separate, and next steps.
How do I know if an AI summary invented something?
Check every specific against a quote or the transcript. Summaries without verbatim quotes cannot be verified and should not reach a committee.
Should the AI's verdict decide the outcome?
No. The verdict is a claim for the committee to confirm or overturn against the evidence; the hiring decision is human.
How long should I keep interview summaries?
As long as your other hiring records, with access limited to the hiring team, then delete them. Several jurisdictions give candidates a right to see evaluations.
How long does reviewing a summary take?
About five minutes: verify quotes, delete unsupported specifics, turn flags into questions, remove sensitive information, add your own line.
Try Interviewer Mode on your next screen
Role-specific questions, live rubric scoring with evidence notes and a structured report at the end of the call. Free in the desktop app for Windows and macOS, no signup.
Get the desktop app