Three meeting recorders, three privacy policies, one question: when you record a client call, does that recording end up training somebody’s model? We read all three policies on 3 November 2026. The answer is different for each, and so is how much you can trust the answer to stay the same.
The short version
- Otter.ai trains on your recordings and transcripts, de-identified, and says so.
- Fireflies.ai says it does not, and extends that to its vendors by contract.
- tl;dv says it does not, and extends that to its third-party AI providers.
The interesting part is not the split. It is that one of those three policies has no date on it.
Otter.ai: yes, on de-identified data
Otter’s privacy policy, effective 16 June 2026, lists among its purposes: “Improve and monitor the Services, including training our proprietary AI technology on de-identified audio recordings and on transcriptions (which may contain Personal Information)”.
Read that clause slowly. The recordings are de-identified before training. The transcriptions are used too, and Otter notes in the same breath that those “may contain Personal Information”. De-identification applies to the audio; the transcript is acknowledged to carry personal information into the process.
The same policy also names “Data labeling service providers who provide annotation services and use the data we share to create training and evaluation data for Otter’s product features”. That means human annotation companies, not only machines.
None of this is hidden. It is written in plain language in a dated document, which is more than some competitors manage. It is simply a different answer from the other two.
Fireflies.ai: no, and the vendors are bound too
Fireflies’ privacy policy, dated 28 September 2026, states: “We do not use personal information for AI model training and we contractually prohibit our vendors from using this information for their own model training.”
It goes further on meeting content specifically, saying your meeting content and Fireflies Talk data is not “(1) stored by any third-party vendor after processing; (2) accessed by any third-party vendor once the service is completed; or (3) used for training internal or external AI models”.
And once more, flatly: “We do not collect, use, or sell personal data for the purpose of training large language models.”
Three separate statements, one dated document. That is a strong position.
tl;dv: no, and it is the clearest wording of the three
tl;dv gives the question its own numbered section. Section 9, “AI Model Training”, reads in full:
“Tldx Solutions GmbH does not use Customer Content, including meeting recordings, transcripts, notes, files, or other data processed through the Services, to train, fine-tune, or improve foundation models, large language models, or other generative AI models for the benefit of tldx Solutions GmbH or any third party. Where we use third-party AI service providers to deliver features of the Services, Customer Content is processed solely to provide the requested functionality and is not used by tldx Solutions GmbH or such providers to train or improve their general-purpose AI models.”
That is the most precise of the three. It names the content types, names the thing it will not do to them (train, fine-tune, improve), and closes the obvious gap by binding the providers it passes data to. tl;dv also names those providers elsewhere in the policy: Anthropic and Google Vertex for AI, AssemblyAI and ElevenLabs for transcription.
The problem with the best-worded promise
tl;dv’s privacy policy carries no effective date and no last-updated date. We looked for one. The only date anywhere on the page is 27 April 2016, which is the date of the EU GDPR regulation itself, not a revision date. Section 12 is titled “Revision of the Personal Data Protection Policy” and says the policy may be updated “according to the needs of the company and legal requirement”, without saying when it last was.
A promise with no date on it cannot be checked for change. If tl;dv softened that wording next month, nothing on the page would tell you, and you would have no version to point back to. Otter dates its policy. Fireflies dates its policy. tl;dv, which has the strongest wording, does not.
tl;dv does date its terms of service — the version published is 18 May 2026 — so this looks like an oversight on one document rather than a pattern. It is still the one thing we would want fixed before relying on section 9 contractually.
What this actually changes for you
If you record internal meetings about your own work, any of the three is defensible. If you record client calls, candidate interviews, user research or anything covered by an NDA, the training clause stops being a preference and becomes a term you may have to answer for.
- Recording a call usually requires consent from everyone on it. That is separate from, and prior to, any of this.
- “De-identified” is a process, not a guarantee. If your transcripts contain company names, product names or figures, ask what de-identification actually removes.
- A no-training promise in a privacy policy is a statement of practice. A no-training promise in a signed DPA or enterprise agreement is a contract. If it matters, get it in the second kind of document.
- Check the date on whatever page you are relying on, and keep a copy. Policies change and most vendors do not announce it.
Prices, for context
tl;dv publishes both billing rates: Pro at $18 per seat a month billed annually, or $29 billed monthly, and Business at $29 annually or $39 monthly. The page opens on the annual figure, so the monthly price is 61 per cent higher than the number you see first on Pro. Both are labelled, which is better than several competitors manage. Current prices for the other two are on their entries: Otter.ai and Fireflies.ai.
tl;dv’s free tier is unusual in that recording and transcription are genuinely unlimited; what it caps is the AI layer, at 10 meetings with AI notes and 10 Ask AI queries.
What we checked, and what we did not
We read the three published privacy policies on 3 November 2026 and quoted them directly. We did not test the products, and we did not audit whether any company does what its policy says. We also did not reproduce tl;dv’s own comparison table against Fireflies, which appears on its pricing page — those are a vendor’s claims about a competitor.
An earlier piece here compared two of these on the same question: Otter.ai vs Fireflies.ai: only one of them trains its AI on your meetings. This updates that to three, with the policies re-read today.