AI interview summaries are employment records, not neutral notes
Recording and summarisation can make an interview searchable and persistent. Employers need consent, correction, access, retention and deletion rules before routine use.
What happened
Challenger, Gray & Christmas warned that AI interview notes can omit context, reproduce transcription errors and create records candidates cannot inspect or correct.
Why it matters
A generated summary can influence hiring, accommodation and dispute decisions long after the conversation ends.
Challenger, Gray & Christmas argues that recording, transcription and AI summarisation are becoming normal in interviews while governance lags. Its review warns that summaries can highlight or omit details, transcription can lose nuance, and personality or emotion inference remains contested. It also says employers should disclose recording and AI use, explain access and retention, and offer alternatives or accommodations.
HR Dive framed the same problem as a durable paper trail. The operational consequence is simple: once a summary informs a score, shortlist or rejection, it is not casual meeting assistance. It is part of the decision record.
Separate transcript, summary and decision
Store the audio or transcript, generated summary and human decision as distinct objects. Record the model and prompt used, edits made by the interviewer and the fields copied into the applicant tracking system. Do not allow a summary to silently become the source of truth. A candidate or reviewer should be able to trace a material statement back to the underlying conversation.
Consent must be meaningful: say what is recorded, which AI functions run, who can access the result, how long it is retained and how to request correction or deletion. Provide a non-recorded path where law, disability accommodation or candidate preference requires one. Disable emotion, personality and protected-trait inference rather than trying to explain it after the fact.
Test for asymmetric error
Run paired quality checks across accents, audio conditions, languages and accommodation scenarios. Track omissions and meaning-changing errors, not just word accuracy. Require a recruiter to confirm every summary before it affects disposition, and make correction visible to downstream reviewers.
The counterargument is that consistent AI notes may reduce interviewer memory bias. That is plausible, but consistency is not accuracy or fairness. A controlled pilot should compare human-only notes, transcripts and AI summaries against the same review standard. Until the controls pass, the governance guidance supports stopping automated summaries from entering hiring decisions.