The tool everyone adopted before reading the rules

An AI notetaker that joins a video interview, transcribes it, and hands the panel a clean summary and a list of who-said-what is a genuine improvement over a hiring manager half-listening while typing. Interviewers can actually pay attention. The scorecard gets written from a record instead of a fading memory. The problem is that the moment a bot starts recording a conversation, you have stepped from "taking notes" into "recording a person" — and recording people is governed by consent law that most hiring teams adopted the tool without ever checking.

This is practical HR guidance, not legal advice. Recording-consent rules are set by state, the candidate may be sitting in a different state than the interviewer, and the rules shift; confirm the specifics for the states you and your candidates are actually in before you standardize a process.

One-party versus all-party consent — and why the candidate's state matters

US wiretap and eavesdropping laws split into two camps:

  • One-party consent — only one person in the conversation has to know it's being recorded. In these states, the interviewer's own awareness can be enough.
  • All-party (two-party) consent — every participant must consent to the recording. A handful of states require this, and California is the one that drives most policies because of its size and its appetite for litigation.

Here's the trap that catches remote-first teams: in a video interview your candidate may be physically in an all-party-consent state even if your company isn't. The conservative, sane operating rule is to treat every interview as if all-party consent applies — get explicit, affirmative consent from the candidate every time, and you never have to litigate which state's law controls. It costs you one sentence and a click.

How to get consent that actually counts

"The bot was visibly in the meeting" is not the same as consent, and a buried line in your privacy policy is weaker still. Make it explicit and make it ahead of time:

  • Disclose in the invite. Say in the calendar invite or confirmation email that the interview will be recorded and transcribed by an AI assistant, and why (accurate notes, fair scoring).
  • Ask at the top of the call. Open with a plain-language ask: "We use an AI assistant to transcribe so I can focus on our conversation — are you OK with that?" Wait for a yes.
  • Offer a real opt-out. A candidate who declines recording must not be penalized — fall back to manual notes. If declining quietly tanks their evaluation, your "consent" was coercion. This is the same evenhandedness that keeps knockout questions and structured interviews defensible.
  • Log that you asked. Capture that consent was given (and by whom, and when) the same way you capture any other hiring step.

The transcript is now a hiring record

This is the part teams forget. Once the conversation is transcribed, that transcript is part of the candidate's hiring record — discoverable in litigation, subject to the same retention rules as the rest of the file, and a place where careless words live forever. Two consequences:

  • Retention applies. Keep interview transcripts on the same clock as your other recruiting records — see recruiting records retention — and purge on schedule rather than letting a bottomless archive of recorded interviews accumulate.
  • Watch what gets captured. A transcript that memorializes an interviewer asking about a candidate's age, family plans, health, or national origin is a documented liability. The fix isn't to stop recording — it's to run a real interviewer training program so the questions stay job-related. The recording just makes good interviewing legible and bad interviewing undeniable.

Don't let the summary make the decision

AI notetakers increasingly do more than transcribe — they score sentiment, rank "engagement," or summarize a candidate as "strong" or "weak." Treat any such output as the same kind of decision-support signal as an AI match score, not a verdict:

  • A summary is a compression of the conversation, and compressions drop nuance and can amplify bias. Read it against the actual transcript, don't substitute it for the transcript.
  • Never let an automated "fit" or sentiment score gate a candidate without a human reading the underlying evidence. A hard cutoff on a machine-generated interview score is exactly the auto-reject pattern regulators have warned about, and the legal frame is in AI, bias, and the EEOC.
  • If you operate where automated-decision-tool disclosure or bias-audit rules apply (NYC's Local Law 144 is the bellwether), an interview-scoring AI may fall squarely inside them.

Vendor and data questions to ask before you turn it on

Before a third-party notetaker touches a single interview, get clear answers:

  • Where does the audio and transcript live, and for how long? Can you set retention and deletion?
  • Is your interview data used to train the vendor's models? For candidate data, the answer you want is no — and you want it in writing.
  • Who at the vendor can access recordings? Sub-processors? Offshore support?
  • Can you delete a specific candidate's data on request? You may be legally obligated to.

A workable policy in five lines

You don't need to ban these tools. You need a repeatable rule: (1) assume all-party consent and ask every candidate, every time; (2) offer a penalty-free opt-out with manual notes as the fallback; (3) treat the transcript as a retained hiring record and purge it on schedule; (4) use summaries and scores to support a human decision, never to auto-reject; (5) vet the vendor on storage, training, and deletion before you enable it.

In Hosting HR, interview notes and scorecards attach to the candidate inside the hiring pipeline on the same records-retention clock as the rest of the file, so whatever you capture — typed by a human or transcribed by a bot — lives where it can be governed, surfaced, and purged on schedule rather than scattered across a vendor's cloud. The AI can take the notes. You still own the record.