Does Meetbook Train AI on Your Meetings?
The short answer: No. Meetbook does not train AI on your meetings — not on your recordings, transcripts, summaries, or notes, and not on any plan, free or paid. Meetbook's help centre states that your meetings, transcripts, and reports "are processed to generate your results and are not fed back into any model's training set." The third-party AI providers Meetbook uses for transcription and report generation are contractually barred from training on your data. There is exactly one opt-in exception — an off-by-default voiceprint feature for recognising the same speaker across meetings — and it stores biometric voice embeddings, not meeting content. Most rival notetakers do train on your data in some de-identified form, which makes a plain "no" rarer than you would expect.
Why "does it train on my data?" is the question that matters
When a notetaker says it "trains on your data," it means your meeting audio and transcripts become part of the dataset that teaches its speech-to-text or summarisation models to do better next time. For a team, that is a meaningful line: a discovery call contains a prospect's budget and objections, a board review contains strategy, a customer call contains names and numbers. "De-identified" training — the standard compromise most vendors offer — removes obvious identifiers but still uses the substance of your conversations, and once something is baked into model weights it generally cannot be extracted or deleted later. That is why the question is worth asking before you press record, not after. The honest answer differs by vendor, and only a few of them publish it in plain language.
At a glance: do these notetakers train on your meetings?
| Notetaker | Trains on your data? | What actually happens | Where it says so |
|---|---|---|---|
| Meetbook | No | Recordings, transcripts, and notes are never fed into a training set. AI providers (OpenAI, AssemblyAI, Deepgram) are contractually barred. Voiceprint is opt-in and off by default. | Meetbook help |
| Fireflies | No | "No AI Training. Ever." A zero-data-retention policy that extends to OpenAI and Anthropic. | Fireflies guide |
| Otter | Yes | Trains its speech models on your recordings and transcripts after de-identifying them, automatically. Its own AI providers do not train. | Otter privacy & security |
| Granola | Yes (opt-out) | Trains on de-identified, aggregated data by default; you can opt out in settings. Enterprise is off by default. Third parties are barred. | Granola privacy policy |
| Fathom | Yes (opt-out) | Trains in-house models on de-identified meeting content; you can opt out in settings. Sub-processors are barred. | Fathom privacy policy |
Does Meetbook train AI on your meetings?
No. Meetbook's own answer to the question "Is my data used to train AI models?" is a single word followed by a commitment: your meetings, transcripts, and reports are processed to produce your results and are not fed back into any model's training set. That holds across the Free, Solo, Business, and self-hosted Enterprise plans — there is no paid tier where training quietly switches on, and no setting you have to find to opt out of something that never happens.
The distinction matters because several notetakers frame their answer as "your data is de-identified before training." Meetbook's position is different in kind: there is no training step to de-identify for, because customer data never enters a training pipeline at all.
Where is Meetbook meeting data stored?
Meeting data is hosted on Microsoft Azure in the EU West Europe region, according to Meetbook's subprocessors page. The AI providers sit further down the pipeline: OpenAI handles report generation and embeddings, while AssemblyAI and Deepgram provide speech-to-text transcription. Meetbook also lists its payment processor Polar.sh, Crisp for support chat, and Mixpanel for analytics. Each subprocessor is named on that page with its purpose, the data it touches, and its processing location.
Meetbook is certified under the EU-U.S. Data Privacy Framework, plus the UK Extension and the Swiss-U.S. Framework, and it offers a Data Processing Agreement on request through [email protected]. One honest gap, stated plainly: Meetbook does not hold a SOC 2 report today, so teams that require a SOC 2 attestation as a contractual condition should raise that before buying.
Do Meetbook's AI providers train on your data?
No. Meetbook says its AI providers are used "under agreements that exclude your data from being used to train their models." OpenAI is contractually prohibited from training on your transcripts, and the same bar applies to the transcription providers. This is the part most people actually mean when they ask about AI training — not "does the vendor fine-tune a model on my voice," but "does the foundation-model company behind the button get to keep my meetings." On Meetbook, the answer is no on both counts.
What is the voiceprint exception?
The one place Meetbook stores something derived from your meetings for reuse is speaker recognition. Cross-meeting voiceprint matching — recognising that the same person spoke in two different meetings — is a separate, opt-in, off-by-default feature. If a workspace admin turns it on, voice embeddings are stored specifically to match the same speaker across meetings, and Meetbook treats them as biometric data under stricter rules than everything else described here. It is not on unless someone with admin rights deliberately switches it on, and it stores an embedding, not a recording or transcript. If you are unsure whether it is enabled for your organisation, the answer is in your workspace admin settings.
Can you delete your data from Meetbook?
Yes. You can delete individual meeting recordings, transcripts, and other content at any time through the product. If you delete your whole account, Meetbook permanently deletes your personal information within 30 days, except where it must keep data to meet a legal obligation, resolve a dispute, or enforce an agreement — the details are in the privacy policy. Aggregated and anonymised data may be retained for analytics, which is the standard carve-out most vendors make, and it sits well apart from "training on your meetings."
Does Otter train on your data?
Yes. Otter's privacy and security page says: "Otter uses a proprietary method to de-identify user data before training our models so that an individual user cannot be identified," and that "this training method is automatic." In plain terms, Otter trains its own speech models on your recordings and transcripts by default, after stripping identifying information, with no manual review. Its AI service providers, in contrast, are told not to train on customer data, and imported documents from Google Workspace are not used for training. If your team uses Otter, assume your meeting content contributes to Otter's model training unless you are on a plan that excludes it. See Otter vs Meetbook and the alternatives for the botless comparison.
Does Fireflies train on your data?
No. Fireflies is the clearest "no" in the category alongside Meetbook. Its policy page is blunt: "No AI Training. Ever. Your meeting content — including audio, video, transcripts, and summaries — is never used to train any AI models." Fireflies pairs that with a zero-data-retention policy, meaning vendors do not store or keep access to your data after processing, and the commitment extends to its transcription providers and to OpenAI and Anthropic. On the AI-training question specifically, Fireflies and Meetbook give the same answer; the products differ on other axes like bots, platforms, and pricing.
Does Granola train on your data?
Yes, with an opt-out. Granola's privacy policy says it uses "de-identified, aggregated data, not Personal Data, for our AI model training," and that you can opt out of this within your account settings. Enterprise users have model training turned off by default and must opt in. Granola is also explicit that OpenAI and Anthropic are not allowed to train on your personal data. So Granola is not a "no" — it is a "Granola trains on anonymised aggregates unless you opt out," which is meaningfully different from Meetbook's no-training-at-all position. If you are weighing the two, read Granola vs the alternatives.
Does Fathom train on your data?
Yes, with an opt-out. Fathom's privacy policy allows it to "use and create de-identified data generated from Meeting Content Information to improve our Service by training, improving, and customizing our in-house artificial intelligence models," with an opt-out in account settings. Fathom's AI sub-processors — Anthropic, OpenAI, and Google — are contractually barred from training on your data. Like Granola, Fathom's position is "de-identified and opt-out," not "never."
Which notetakers don't train on your meetings?
Two of the well-known names give an unqualified no: Meetbook and Fireflies. Otter, Granola, and Fathom all train on meeting data in some de-identified form, with Granola and Fathom offering an opt-out. If "my meetings never touch a training set" is a hard requirement for you, that short list is your starting point — and Meetbook is the one on it that records from your browser tab with no bot joining the call. The Chrome extension captures the meeting tab directly, and the same no-training policy applies whether the recording comes from the extension, the calendar notetaker, or the Android app. If you are still deciding how to choose a notetaker at all, the buyer's guide walks through capture method, platforms, and privacy.
How to check any notetaker's training policy yourself
You do not need to trust a summary — the answer is almost always in two places on the vendor's site. First, open the privacy policy and search for the word "train." If the only sentence you find is about third-party providers, look for whether the vendor itself trains on de-identified or aggregated data. Second, check whether there is an opt-out, and whether it is on or off by default. Three questions to ask of any vendor, in order:
- Does the vendor train on my recordings or transcripts at all?
- Are its AI providers — OpenAI, Anthropic, Google, and the rest — allowed to train on my data?
- Is training opt-out, opt-in, or not happening, and which plans does that apply to?
If a vendor will not answer these three in writing, treat that silence as a "yes." Meetbook's answers are no, no, and "not happening, with one opt-in voiceprint exception," and they are written down in its help centre.
FAQ
Does Meetbook train AI on my meetings?
No. Recordings, transcripts, summaries, and notes are never fed into a model's training set, on any plan.
Does Meetbook sell my meeting data?
No. Meetbook does not train on or sell meeting content; its AI providers are contractually barred from training on it, and the full list of who touches your data is published on the subprocessors page.
Is there anything Meetbook does opt me into automatically?
No. The only optional data use is the voiceprint speaker-recognition feature, which is off by default and switched on by a workspace admin, not by you silently.
Is Meetbook SOC 2 compliant?
No. Meetbook does not hold a SOC 2 report today. It is certified under the EU-U.S. Data Privacy Framework and offers a DPA on request, but teams that require SOC 2 should confirm that gap before purchasing.
Does Otter train on your data?
Yes. Otter trains its speech models on de-identified recordings and transcripts automatically, while its AI providers do not train on customer data.
Does Fireflies train on your data?
No. Fireflies states "No AI Training. Ever" and applies a zero-data-retention policy across its vendors.
Does Granola train on your data?
Yes, unless you opt out. Granola trains on de-identified, aggregated data by default; you can opt out in settings, and Enterprise is off by default.
Does Fathom train on your data?
Yes, unless you opt out. Fathom trains in-house models on de-identified meeting content with an opt-out in settings, and its sub-processors are barred from training.