Checked against Otter's privacy policy today: yes. It states Otter uses account information to "improve and monitor the Services, including training our proprietary AI technology on de-identified audio recordings and on transcriptions" — and specifically notes those transcriptions "may contain Personal Information," even where the audio itself has been de-identified.
What "de-identified" covers, and what it doesn't
De-identifying the audio addresses one part of the data — the voice recording itself — but Otter's own language draws a line between that and the transcriptions produced from it, which can retain names, companies, and anything else said in the meeting. Training on "de-identified audio" is a narrower claim than "we don't train on anything that could identify you," and the policy doesn't claim the broader one.
What opting out actually looks like
The policy doesn't describe a setting made specifically for AI-training consent. What it offers instead are the general privacy rights most services provide: a "Do Not Sell or Share My Personal Information" link in the site footer, and the ability to request access, correction, deletion or an export of personal information by contacting Otter directly. Whether either mechanism actually stops a specific recording from being used in a future training run isn't spelled out in the policy — it's a data-rights process, not a training toggle.
Why this is a live question, not a settled one
Otter is currently facing a consolidated federal class action alleging, among other things, that it used recorded conversations to train its models without adequate consent — a federal judge allowed several of the underlying claims to proceed past a motion to dismiss in August 2026. None of that changes what the privacy policy currently says, but it's the reason to read the policy's own wording rather than a summary of it, and to treat "de-identified" as the specific, limited claim it is rather than a broader guarantee.
What this means for a call with someone else on it
The training question isn't only about the account holder. Everyone on a call Otter records has their voice and words potentially feeding the same training process, and none of them sees or agrees to Otter's privacy policy — only the person who started the recording does. That's true of most cloud notetakers to some degree, but it's a sharper problem when the policy states outright, as Otter's does, that transcriptions used for training may still carry personal information rather than promising they've been stripped of it.
Where Coii AudioNotes differs
There's nothing to train on, because nothing leaves the Mac. Coii AudioNotes doesn't send audio, transcripts or summaries anywhere — training on them would require having them, and it never does. That's a structural difference from an opt-out setting: a setting can default the wrong way or change in a future policy update, and this can't, because there's no upload step for a policy to govern in the first place. Meeting apps that don't train on your data and Coii AudioNotes against Otter go further into both sides of that comparison, and Otter alternatives covers the field beyond just this one question. What an AI notetaker does with a recording more broadly, training included, is in what is an AI notetaker.