Does an AI note taker train on your recordings? It depends on the vendor, and each vendor states its answer in its own privacy, trust or help pages. On Nov 12, 2026, Fireflies states that it does not use meeting content for AI training, and Plaud states that it does not train on user data unless you opt in. Otter and Fathom state that they use de-identified data, and Granola states that it may use anonymized data, to improve their own models and services. Comu does not use recordings to train AI.
Comu makes Comu Note Pro and Comu Action Pro. If you only want to know what Comu does with a recording, skip to what Comu AI does with your recordings.
What does "training on your recordings" mean for an AI note taker?
Training means a vendor uses your audio or transcripts as examples to build or improve a machine learning model. Processing is different. Processing is the work the tool does for you: turning audio into a transcript and a summary. Every AI note taker processes your recordings. Training is a separate use of the same data, and vendors state it separately.
Four questions separate the cases:
- Who trains? The note taker company can train its own models. The AI providers it pays to run transcription and summaries, called sub-processors, can train theirs. Otter, Fireflies, Fathom, Plaud and Granola each state both.
- On what data? Raw audio and transcripts, or de-identified and anonymized versions with names and identifying details removed.
- Is it on by default? A vendor can train unless you opt out, train only if you opt in, or not train at all.
- Does a person listen? Automated training is different from a human reviewing a recording. Otter states that its training is automatic and that audio recordings and transcripts are not manually reviewed by a human. Otter also states that it requires explicit consent from a customer before Otter staff or third parties review a recording manually.
Applied to five vendors and Comu, the vendors' statements fall into four groups:
- State that they do not train: Fireflies and Comu.
- State that they train only after you opt in: Plaud.
- State that they train on de-identified or anonymized data, with an opt-out setting: Fathom and Granola.
- State that they train on de-identified data, with Enterprise workspaces opted out by default: Otter.
AI note taker data privacy: what should you check on a vendor's privacy page?
Eight checks cover a vendor page. Run them in this order.
- Find the training sentence. Search the privacy policy, trust center or help center for "train". A clear statement names the data (audio, transcripts), the default, and who is covered. Plaud's Trust Center states: "Plaud will not train on your data by default." It adds that explicit opt-in is required.
- Find the opt-out setting. A setting has a name and a path. Granola's toggle is "Use my data to improve models for everyone", under Settings > Preferences > Data & sharing.
- Check whether the control depends on the plan. Otter states that Enterprise workspaces are opted out of AI model training by default. Granola states that Enterprise workspaces are opted out by default, while on Granola Business each user opts out individually. Granola's HIPAA support covers Enterprise customers only.
- Read the sub-processor terms. Fathom names Anthropic, OpenAI and Google as sub-processors and states none of them is contractually permitted to use user data to train AI models. Fireflies states that vendors do not store meeting content after processing, do not access it once the service is complete, and do not use it for training.
- Find the retention rule. Granola states that it deletes audio once transcription is complete and keeps transcripts indefinitely unless an auto-deletion period is set. Otter states that a deleted conversation stays in Trash for 30 days.
- Find the deletion path. Look for a delete action in the app and a stated timeline for account closure. Fireflies states that it deletes personal information related to a closed account within 30 days. Fathom states that it removes recording data and metadata when you delete your account and clears backups after 7 more days.
- Check voice data. Speaker identification can build voice profiles. Fireflies states that its service providers may use voice recordings to extract voice characteristics, which some jurisdictions treat as biometric identifiers. Otter lets Enterprise admins turn off speaker learning, and states that it does not build or update voice profiles in that workspace.
- Check certifications and the policy date. Otter, Fathom, Plaud and Granola each state SOC 2 Type II and GDPR compliance, and Comu is GDPR, HIPAA and SOC 2 compliant. Note the "last updated" date too. Fireflies' privacy policy is dated Sept 28, 2026.
Which AI note takers train on your recordings? What five vendors state
| Training on recordings and transcripts | Default | Where to limit it | |
|---|---|---|---|
| Comu | Does not use recordings to train AI | No training | Delete a recording, a contact or the whole account in the app |
| Otter | De-identifies user data before training its own models; AI service providers do not train on customer data | Enterprise workspaces are opted out | Enterprise: opted out by default |
| Fireflies | Does not use personal information or meeting content for AI model training; vendors are contractually prohibited from training | No training | States no training; delete information in account settings or through live chat |
| Fathom | Uses de-identified customer data to improve its own models; sub-processors are not permitted to train | Uses de-identified customer data; you can opt out at any time | My Settings; Team Edition admins use Organization Settings |
| Plaud | Does not train on user data by default | Off | Opt in manually in settings to allow training |
| Granola | May use anonymized data to improve its services | On for Free and Business, off for Enterprise | Settings > Preferences > Data & sharing |
Checked Nov 12, 2026.
Does Otter train on your recordings?
Otter states that it uses a proprietary method to de-identify user data before training its models, and that its training data is encrypted. Its AI service providers receive customer data through an API and do not store it on their platform. Otter states that no customer data is used to train or improve those providers' models.
Otter states that Enterprise workspaces are opted out of AI model training by default and that data in those workspaces is not used to improve Otter's global models or any third-party large language model. Otter states that its HIPAA-aligned configuration also restricts using customer data for AI model training.
Does Fireflies train on your meetings?
Fireflies' privacy policy states that it does not use personal information for AI model training and that it contractually prohibits vendors from using this information for their own model training. A separate Zero Data Retention rule covers meeting content, meaning audio, video, transcripts and summaries. That content is not stored by a third-party vendor after processing, is not accessed by a third-party vendor once the service is completed, and is not used for training internal or external AI models.
Does Fathom train on your meetings?
Fathom states that none of its AI sub-processors, Anthropic, OpenAI or Google, is contractually permitted to use user data to train AI models. Fathom itself uses de-identified customer data to improve its proprietary AI models. You can opt out at any time in My Settings. Organizations on Team Edition can opt out all users from Organization Settings.
Fathom states that it does not and will never sell your data to third parties, and that all Fathom data is stored in the United States.
Does Plaud train on your recordings?
Plaud states that it does not train on user data by default unless users opt in manually in their settings. Plaud states that the sub-processors behind its language models have a zero data retention policy and that contracts and Data Processing Agreements prohibit them from using customer data for model training. Plaud states that it uses several enterprise-grade models for transcription, including proprietary fine-tuned models, and that user data is not used to train them unless the user explicitly opts in.
Does Granola train on your data?
Granola states that by default it may use anonymized data to improve its services and that this data is never sent to third parties. Granola also states that enterprise agreements prevent its providers from using your data in their model training. You can opt out at any time. Enterprise customers are opted out by default.
AI note taker opt out of training: what are the steps?
Granola and Fathom offer an opt-out switch. Otter opts Enterprise workspaces out by default. Plaud's setting works the other way round: training stays off unless you opt in. Fireflies and Comu state no training, and Comu lets you delete a recording, a contact or the whole account in the app.
Granola, desktop app:
- Click your workspace name in the bottom-left corner and select Settings.
- Open Preferences, then Data & sharing.
- Find the setting "Use my data to improve models for everyone".
- Turn the toggle off.
The setting applies to your whole account, so you change it once. The toggle is not in the Android app yet, so change it from desktop or iOS. On the Business plan each user opts out individually. On Enterprise the toggle is labeled "Use my workspace data to improve models for everyone"; Granola asks Enterprise customers to email sales@granola.so to change it.
Fathom: open My Settings and opt out. On Team Edition, an organization admin can opt out every user from Organization Settings.
Otter: Enterprise workspaces are opted out of AI model training by default.
Plaud: training stays off unless you opt in manually in settings.
What does an opt-out cover?
Granola's opt-out limits future use. Granola states that once you opt out, none of your data is used for any future model training, and also states that it cannot guarantee anonymized data was not used before you changed the setting. Change the setting on the day you sign up, and write down the date you changed it.
Deletion is a separate action, covered next.
Do AI notetakers store recordings? What each vendor states about retention and deletion
| Audio and transcripts | How to delete | Retention rule | |
|---|---|---|---|
| Comu | Files are encrypted | Delete one recording, a contact or the whole account in the app; the text linked to a recording is deleted with it | Deleting the account clears all data |
| Otter | AWS S3 storage with server-side 256-bit AES encryption | A deleted conversation moves to Trash; clearing the Trash deletes it at once | Trash is emptied automatically after 30 days |
| Fireflies | Third-party vendors do not retain meeting content after processing | Delete information in account settings or through live chat; an attendee without an account asks the meeting host to delete the recording | Personal information of a closed account is deleted within 30 days |
| Fathom | Stored in the United States | Delete your account to remove all recording data and metadata | Backups are cleared 7 days after deletion |
| Plaud | Audio and transcription data are not stored in the cloud unless Cloud Sync is on | With Cloud Sync off, data is deleted from the cloud once it is transferred back to the phone | Kept while the account is active; deleted or anonymized when the account is deactivated |
| Granola | Audio is cached temporarily during the meeting and deleted after transcription | Settings > Profile > Delete Account | Transcripts are kept indefinitely unless you set an auto-deletion period |
Checked Nov 12, 2026.
Several of these controls sit with administrators. Otter Enterprise admins can remove audio from conversations, set a custom data retention policy and disable speech deletion. On Granola non-Enterprise plans, you set your own transcript auto-deletion period under Settings > Preferences > Data & sharing, with a one-week cooldown before it takes effect.
For where a recorder keeps audio and what you can remove, read where an AI voice recorder stores audio and what you can delete.
Which certifications and sub-processor terms should an AI note taker have?
| Certifications and compliance stated | Data location or storage | |
|---|---|---|
| Comu | GDPR, HIPAA and SOC 2 compliant; CCPA compliant | Files encrypted |
| Otter | SOC 2 Type 2 report, GDPR, CCPA, HIPAA | AWS S3 storage |
| Fireflies | EU-U.S. and Swiss-U.S. Data Privacy Frameworks, with the UK Extension (stated) | United States and other countries |
| Fathom | HIPAA, SOC 2 Type II and GDPR compliant | United States |
| Plaud | GDPR, HIPAA, SOC 2 Type II | US, Frankfurt, Singapore and Japan data centers |
| Granola | SOC 2 Type II; GDPR and UK GDPR; HIPAA for Enterprise customers | AWS in the United States |
Checked Nov 12, 2026.
Read these terms in three layers.
- Audits and laws. SOC 2 Type II is an audit report on a company's controls. GDPR and CCPA are laws. HIPAA is a U.S. health-privacy law, and Otter and Granola tie their HIPAA support to a signed Business Associate Agreement (BAA): Otter describes its HIPAA configuration as working together with a BAA, and Granola signs a BAA for Enterprise customers.
- Sub-processor contracts. Plaud states that all language models in Plaud have zero training and zero data retention policies enforced through Data Processing Agreements. Granola states that enterprise agreements prevent its providers from training on your data. Fireflies applies a zero data retention rule to vendors that process meeting content.
- Where the data sits. Plaud has data centers in the US, Frankfurt, Singapore and Japan, and users in Europe get EU-hosted sub-processors. Fathom and Granola state that data is stored in the United States.
For a side-by-side of AI voice recorders on GDPR and SOC 2, read which AI voice recorders are GDPR and SOC 2 compliant.
What does Comu AI do with your recordings?
Comu does not use your recordings to train AI and does not sell data to third parties. Other protections for Comu recordings, as of Nov 12, 2026:
- Encryption. Files are encrypted.
- Deletion in the app. You can delete a single recording, and the text linked to it is deleted too. You can delete a contact. You can delete the whole account, which clears all data.
- Access and control. You can access, edit and delete your data at any time. To get a copy of your data, contact Comu and request it under the data portability right in the Comu privacy policy.
- Compliance. Comu is GDPR, HIPAA and SOC 2 compliant, and CCPA compliant. Independent security experts review Comu's systems regularly.
Comu Note Pro and Comu Action Pro use the same Comu app, so these controls are the same on both devices.
Can an AI note taker change its training policy after you sign up?
A vendor can update its policy after you sign up. In February 2024, the U.S. Federal Trade Commission's Office of Technology wrote that a company may act unfairly or deceptively when it adopts more permissive data practices and tells consumers only through a surreptitious, retroactive change to its terms of service or privacy policy. It named using customer data for AI training as an example.
Two habits keep you ahead of a change:
- Save the date of the policy you read. Fireflies states that it notifies users of changes by revising the date at the top of its privacy policy, and gives additional notice for material changes.
- Reread the training paragraph whenever a vendor emails you about updated terms.
For recording consent rules, read is an AI note taker legal.
Which AI note taker privacy setup fits your situation?
| Your situation | Check first | What the vendors state |
|---|---|---|
| You want no training and no setting to manage | A vendor whose privacy policy states no training | Fireflies and Comu state no training. Plaud states that it does not train unless you opt in. |
| You use a free or individual plan on a vendor that trains on de-identified data | The opt-out toggle | Fathom: My Settings. Granola: Settings > Preferences > Data & sharing. |
| Your employer has an Enterprise agreement | The workspace training setting, through your admin | Otter and Granola state that Enterprise workspaces are opted out by default. Granola admins can enforce the setting for the whole organization. |
| You do not want audio kept | Audio retention | Granola states that it deletes audio after transcription. Plaud states that it keeps audio in the cloud only with Cloud Sync on. Otter Enterprise admins can remove audio from conversations. Comu lets you delete any recording in the app. |
| You handle health information | HIPAA terms and a BAA | Otter's HIPAA-aligned configuration works with a BAA. Granola signs a BAA for Enterprise. Fathom, Plaud and Comu state HIPAA compliance. |
| You want to limit voice profiles | Speaker learning | Otter Enterprise admins can disable speaker learning. Fireflies states that its service providers may use voice recordings to extract voice characteristics. |
For the choice between a table recorder and a call bot, read in-room recorder vs meeting bot: when each is allowed at work for that choice, and what an AI note taker is, its types and how to choose for the product categories.
Frequently Asked Questions
Does Otter train on my recordings?
Yes, on de-identified data. Otter states that it de-identifies user data before training its own models and that the training is automatic, with no manual human review of audio or transcripts. Enterprise workspaces are opted out of AI model training by default.
Does Fireflies train AI on my meeting data?
Fireflies states that it does not. Its privacy policy states that it does not use personal information for AI model training, and its Zero Data Retention rule states that meeting content is not used for training internal or external AI models.
Can I opt out of Fathom's AI training?
Yes. Open My Settings and opt out at any time. Team Edition organizations can opt out all users from Organization Settings. Fathom states that none of its AI sub-processors is contractually permitted to train on user data.
Does Plaud use my recordings to train AI?
Not by default. Plaud states that it does not train on user data unless users opt in manually in their settings, and that its sub-processors are contractually prohibited from training on customer data.
Does Granola store my audio?
Only temporarily. Granola states that audio is cached during the meeting for transcription and is deleted once transcription is complete. Transcripts and notes are kept indefinitely unless you set an auto-deletion period.
Does Comu train AI on my recordings?
No. Comu does not use recordings to train AI and does not sell data to third parties. You can delete a recording, a contact or the whole account in the app.
Are the AI providers behind a note taker allowed to train on my recordings?
Otter, Fireflies, Fathom, Plaud and Granola each state that their AI providers are not. Fathom names Anthropic, OpenAI and Google and states none is contractually permitted to train on user data. Plaud, Granola and Fireflies state that contracts prohibit their providers from training on customer data.
Related reading
- Which AI voice recorders are GDPR and SOC 2 compliant: certifications compared for AI recorders.
- Where does an AI voice recorder store audio, and what can you delete: local and cloud storage and deletion options.
- Is an AI note taker legal: the consent rules that apply before you record.
- AI note taker: what it is, types and how to choose: the categories behind the vendors above.
- In-room recorder vs meeting bot: when each is allowed at work: how the two setups differ at work.
Sources
Checked Nov 12, 2026.
- Otter Privacy & Security and Otter Help Center: Enterprise Admin Controls Overview
- Fireflies Privacy Policy, last updated Sept 28, 2026
- Fathom Help Center: Is Fathom secure?
- Plaud Trust Center
- Granola Docs: Models & training and Security, Privacy & Data FAQs
- FTC Office of Technology: AI (and other) Companies: Quietly Changing Your Terms of Service Could Be Unfair or Deceptive (Feb 2024)
