Research · Privacy
Do transcription apps train AI on your audio? 10 policies compared
What Otter, Fireflies, Fathom, Granola, Plaud, Zoom, Rev, Apple, OpenAI and Deepgram each publish about training AI on your recordings, dated.
We may earn a commission if you buy through links on this page. It never changes what we recommend. How this works
It depends on the app, and the answer is in each company's own privacy policy or terms. As of November 2026, some say they train their own models on de-identified recordings (Otter, Fathom, Granola), some say they don't train on your meeting content at all (Fireflies, Zoom), one says it only does with your consent first (Plaud), and the developer services split: OpenAI's API doesn't train on your audio by default, and Deepgram does unless you opt out.
This is a research-based roundup: we read each company's published privacy policy, terms or trust page on November 14, 2026, and quote or closely paraphrase what each one says. We haven't tested or audited any of them, and policies change, so every row carries the date the company shows. Sources are at the bottom.
The table
| App or service | Trains AI on your recordings? (as published) | Your choice | Policy date |
|---|---|---|---|
| Otter | Yes: its own AI, on de-identified audio and on transcriptions | No training-specific opt-out described | Effective June 16, 2026 |
| Fireflies | No: "We do not use personal information for AI model training" | Not needed, per the policy | Updated Sept. 28, 2026 |
| Fathom | Yes: its in-house models, on de-identified meeting data | Opt out in account settings | Updated Aug. 16, 2026 |
| Granola | Yes: on "anonymized" data | Opt out in Settings; off by default for Enterprise | No date shown |
| Plaud | Only with your explicit, informed consent in advance | Opt in | Effective dates by account (see below) |
| Zoom | No: not audio, video, chat or similar customer content | Not needed, per the policy | Updated Sept. 17, 2026 |
| Rev | Yes for its speech-to-text models; no for generative AI | Not described in the Terms | Updated Oct. 14, 2026 |
| Apple Intelligence | No, unless you explicitly choose to help improve the models | Opt in | No date shown |
| OpenAI API | No, unless you explicitly opt in | Opt in | Since March 1, 2023 |
| Deepgram | Yes, by default | Opt out per API request | Updated Nov. 5, 2026 |
"Trains" here means the company uses your content to train or improve its AI models. Every company above still processes your audio to give you a transcript. That's the service, and it isn't the same thing as training.
App by app
Otter
Otter's privacy policy, effective June 16, 2026, lists among the ways it uses data: "training our proprietary AI technology on de-identified audio recordings and on transcriptions (which may contain Personal Information)." The policy doesn't describe a separate switch to turn training off. If you use Otter through an employer's Business or Enterprise account, Otter says your employer's policies apply too.
What it means for you: assume Otter uses de-identified versions of your recordings to improve its AI. Our Otter review covers its plans and features.
Fireflies
Fireflies' privacy policy, last updated September 28, 2026, says: "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." Content Fireflies processes for business customers is governed by those customers' agreements, which aren't public.
Fathom
Fathom's privacy policy, last updated August 16, 2026, says that, depending on how your account is set up, it may create de-identified data from your meeting content "to improve our Service by training, improving, and customizing our in-house artificial intelligence models," and adds: "You can opt out from this use of your data in your account settings." It also says it doesn't let third parties such as OpenAI, Anthropic or Google train their models on your meeting content.
Granola
Granola's security page says: "Granola trains on your anonymized data so we can keep making Granola better. You can opt out of this in your Settings. Enterprise users have model training turned off by default." The same page says third parties like OpenAI or Anthropic may not train on your data, and that Granola doesn't store the audio from meetings. It transcribes, then keeps the text. The page doesn't show an update date, and it doesn't define "anonymized."
Plaud
Plaud's privacy policy says it won't use your content to train, optimize or develop its own or any third-party AI or machine learning models "unless we have obtained your explicit and informed consent in advance." It also says its agreements with the AI model providers it uses require that your data isn't used to train their models. The policy shows effective dates rather than an update date: the day you accept it for accounts created on or after October 20, 2026, and November 21, 2026 for older accounts.
Zoom
Zoom's privacy statement, last updated September 17, 2026, says Zoom "does not use any of your audio, video, chat, screen sharing, attachments or other communications-like Customer Content" to train Zoom's or third-party AI models. That covers recordings and meeting content you make in Zoom, including what AI Companion works from.
Rev
Rev's Terms of Service, updated October 14, 2026, say that if you use Rev's speech-to-text services, your content "will be analyzed by our ASR models and other Rev artificial intelligence models and may be used for continuous training of those models." The same section says: "Customer Content will not be used for any generative AI model training." So Rev draws a line between its own speech-recognition models (may train) and generative AI (won't). The Terms don't describe an opt-out. Rev's help center has an article on AI training opt-out, but it didn't load for us, so we can't tell you what it says.
Apple
Apple's page on the data used to train Apple Intelligence says: "Apple does not use our users' private personal data or user interactions when training our foundation models unless users explicitly choose to help improve them." That page covers Apple Intelligence in general. Apple doesn't publish a training statement specific to Voice Memos or call recording transcripts that we could find, so we don't make one here.
OpenAI API (for developers)
If an app sends your audio to OpenAI's speech-to-text API, OpenAI's data controls page says: "As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)." This applies to the API that developers use. The ChatGPT app has its own settings.
Deepgram (for developers)
Deepgram's privacy notice, last updated November 5, 2026, says: "We may use your Content to train and improve our models." For its APIs, you can opt out "on a per-request basis by setting the applicable parameter in your request," and the opt-out applies only to the request where it's set. Content Deepgram processes on behalf of a business customer falls under that customer's agreement instead.
What "de-identified" and "anonymized" mean, and don't
When a policy says it trains on de-identified or anonymized data, the company is saying it removes details that point to you before training. None of the policies above explain how they do that for speech, and recordings are hard to scrub: people say names, addresses and account numbers out loud, and a voice can identify a person. We can't measure how well any company does this, so we don't rank them on it.
How to choose
- If training is a dealbreaker, pick a service whose policy says no, or one that only trains with your opt-in, and read the policy again when it changes. Zoom, Fireflies, Plaud, Apple Intelligence and the OpenAI API fit that today, as published.
- If you're fine with training but want a choice, Fathom and Granola describe an opt-out in settings. Turn it off when you sign up, before your first recording.
- If you use a work account, your employer's agreement usually decides. Ask your IT team, and don't assume the consumer policy applies.
- Want nothing to leave your computer? Transcribing on your own machine with open-source Whisper avoids the question. See Whisper vs. cloud transcription.
- Remember the other people in the room. Their voices are in your recordings too. Our post on keeping recordings private has more on that.
Where Daytapes stands
Daytapes is our own product, so weigh this accordingly. Daytapes doesn't train AI models on your recordings. It runs speech-to-text and summaries on Cloudflare Workers AI to build your recording pages, and it shares what you choose through links you control. It's invite-only right now, and its limits are real: on iPhone, the web app pauses recording when the screen locks.
Questions
Does Otter train AI on my recordings?
Otter's privacy policy (effective June 16, 2026) lists training its own AI on de-identified audio recordings and on transcriptions, which it says may contain personal information, as one way it uses data.
Which apps say they don't train on my meetings at all?
As of November 2026, Fireflies and Zoom state in their privacy policies that they don't use your meeting content to train AI models, and Plaud says it won't without your explicit consent in advance.
Can I opt out of training?
Sometimes. Fathom and Granola say you can opt out in your settings. Deepgram's API lets you opt out per request. Otter's policy doesn't describe a training-specific opt-out.
Is de-identified data still my data?
It's built from your recordings, but the company has removed details that identify you. How well that works for speech, which can include names and personal details, isn't something these policies measure.
Does Daytapes train AI on my recordings?
No. Daytapes runs speech-to-text and summaries on Cloudflare Workers AI to make your pages and doesn't train models on your recordings. Daytapes is our own product.
Sources
- Otter.ai Privacy Policy (effective June 16, 2026) (accessed November 14, 2026)
- Fireflies.ai Privacy Policy (last updated September 28, 2026) (accessed November 14, 2026)
- Fathom Privacy Policy (last updated August 16, 2026) (accessed November 14, 2026)
- Granola: Security at Granola (accessed November 14, 2026)
- Plaud Privacy Policy (accessed November 14, 2026)
- Zoom Privacy Statement (last updated September 17, 2026) (accessed November 14, 2026)
- Rev Terms of Service (updated October 14, 2026) (accessed November 14, 2026)
- Apple Legal: Training Data, datasets used for Apple Intelligence (accessed November 14, 2026)
- OpenAI: Data controls in the OpenAI platform (accessed November 14, 2026)
- Deepgram Privacy Notice (last updated November 5, 2026) (accessed November 14, 2026)
- Daytapes: Privacy (accessed November 14, 2026)