AI note takers that don't store audio: which keep a recording of the interview, and which discard it

Several AI note takers transcribe an interview and then discard the audio. As of October 2026, Granola says its desktop apps store "only the transcript," Jamie says meeting audio is "streamed for transcription and never stored," Fireflies' bot-free desktop mode saves no audio file, and Interview Signal does not retain audio. Most meeting bots, including Otter, Fireflies' notetaker, Fathom and tl;dv, keep a recording by default, and Otter and Fireflies let Enterprise admins switch audio storage off. Whether a bot joins the call is a separate question from whether audio is kept.

We make Interview Signal, one of the tools below; everything about the other products comes from their own sites, linked and dated. For bot versus bot-free capture, see AI note takers without a bot; for where each tool keeps what it does store, see where AI note takers store interview data.

Which AI note takers keep the audio, and which discard it?

As stated in each vendor's documentation, as of October 2026.

Sources, as of October 2026: Granola consent and audio guide and security FAQ; Jamie security; Fireflies desktop audio and recording modes; Otter Enterprise admin controls and desktop app; Shadow privacy and data; Fathom organization settings and pricing; tl;dv privacy policy; Zoom meeting summary and data governance; our security page. We could not confirm Krisp's current audio handling from its own help center, so it is not in the table.

Interview Signal's weak spots for this use: there is no audio file to replay, it is desktop only (Windows and macOS), phone screens must run through the computer, and it has no ATS integration, so you paste the scorecard.

Is "no bot" the same as "no recording"?

No, and security reviewers often conflate them. Capture method decides who can see the tool in the call; storage decides what exists afterwards. The table shows all four combinations:

• Bot, audio kept: Fathom and tl;dv by default, and most ATS notetakers. Greenhouse's Notetaker, for example, lets admins choose "Audio and video, or Audio only" per rule, with recordings kept up to 60 days (Greenhouse).
• Bot, audio discarded: Fireflies' Enterprise transcript-only mode.
• No bot, audio kept: Otter's desktop app by default, and Shadow, which keeps audio on your Mac unless you change it.
• No bot, audio discarded: Granola, Jamie, Fireflies' Take Notes and Interview Signal.

For a regulated employer, such as a bank, the strongest position for interviews is usually the last one: nothing visible joins the candidate's call, and the only thing that persists is text you can retain and delete on a schedule.

Where does the audio go, if it is not stored?

Almost always to a speech-to-text service during the call. Granola says audio is cached for transcription and then "deleted from our systems and any third-party services." Fireflies says its transcript-only mode captures audio temporarily "and automatically deletes it once processing is complete." Shadow is the exception in this list: its help page says raw audio "is not sent to an external transcription provider," although its optional AI features can send transcripts and notes out.

So "we don't store audio" is a statement about the vendor's own storage. Your security review still needs the subprocessor list and the transcription provider's retention terms; the AI note taker security questionnaire has the questions, and zero data retention for AI note takers explains what a provider's "zero retention" promise does and does not cover.

What does dropping the audio change for security reviews?

• Less to lose. A breach of a transcript store exposes words; a breach of a recording store exposes voices and, for video, faces and surroundings.
• Fewer biometric questions. Voiceprints count as biometric identifiers under Illinois law (740 ILCS 14/10). Dropping audio does not by itself stop voice profiling, though: Otter offers Enterprise admins a separate setting to disable speaker learning from stored voice data. Ask every vendor about voice profiles directly.
• Smaller retention schedule. One fewer data type to expire, export, hold and delete. Our comparison of auto-delete settings shows how uneven those controls are.

What does it change for candidate access requests?

A candidate who asks for their data under laws such as the GDPR or the California Consumer Privacy Act is asking for what you hold. If no recording exists, there is no recording to find, review or hand over. The transcript, summary, scorecard and any submittal are still personal data, and you still need to locate every copy. See candidate data access requests for the process. This is general information, not legal advice.

What do you lose by not keeping the recording?

The ability to check. If a transcript turns "fifteen" into "fifty," or gives the interviewer's words to the candidate, the audio is how you would normally settle it. Without audio:

1. Check key quotes while the call is fresh, before they go into a scorecard or submittal.
2. Prefer tools that keep the two sides apart. A desktop app that captures your microphone and the call's sound separately avoids most speaker mix-ups on a one-to-one call.
3. Keep the transcript for your full retention period, since it is now the only verbatim record.
4. Write down that you chose not to keep audio and why, so the decision is documented if a hiring decision is challenged.

What should you ask a vendor that says it doesn't store audio?

1. Is that true in every capture mode, or only in one? Fireflies, for example, keeps audio with its bot but not with its desktop Take Notes mode.
2. Is it true on every platform? Granola's iPhone app caches audio during the meeting before deleting it, unlike its desktop apps.
3. Which speech-to-text and AI providers receive the audio, and do their terms let them keep or train on it?
4. Is the setting enforced for the whole workspace, or can each user change it?
5. Does the change apply to past meetings? Otter's "Remove audio recordings" applies to historical meetings too; Fireflies' recording modes apply only to future ones.
6. Are voice profiles built for speaker identification, and can they be turned off?

Get the answers in writing, with the date, and keep them with your vendor review.

How do you choose?

• You want the least data possible: a bot-free tool that discards audio, with transcript auto-deletion set.
• You already pay for Otter or Fireflies Enterprise: switch on "Remove audio recordings" or the transcript-only mode before interviews start, and confirm it applies to all users.
• You need to rewatch or share clips: a recorder such as Fathom or tl;dv, with a short retention policy and explicit consent.
• You only need a summary of a Zoom interview you host: Zoom's meeting summary without cloud recording, with the summary's own auto-delete set.

Whatever you pick, consent does not change: Shadow's own help page notes that not keeping audio does not remove the need for recording consent. Our interview recording consent script has wording that works for transcription-only tools.