Which AI note takers don't train on your recordings? 8 policies compared

AI training policies, side by side

Wave AI

Best for: phone calls, in-person conversations and video meetings in one app, with no AI training on what you record there.

Wave AI never uses the recordings and transcripts in its apps to train AI models. Files uploaded to the free web tools on wave.co are the exception: they may be used to improve Wave's speech recognition. For a private file, import it in the Wave app instead; file import is part of Wave AI Pro.

Pros

• Records phone calls (both sides, from your own number), in-person meetings, Apple Watch and Wear OS recordings, and Zoom, Google Meet and Microsoft Teams meetings in one account.
• Encrypts data with TLS 1.2+ in transit and AES-256 at rest.
• Signs a HIPAA Business Associate Agreement (BAA) with organizations on a Wave Teams plan.

Cons

• Files uploaded to the free web tools on wave.co may be used to improve speech recognition.
• Wave AI's SOC 2 report is Type 1: it covers how controls were designed as of March 1, 2025, with no period of operation tested.
• The BAA needs Wave Teams, which has a 5-seat minimum.

Price: free for 30 minutes of recording a month. Pro is $15 a month, or $80 a year on the web ($100 a year in the App Store and Google Play). See Wave AI pricing.

Otter.ai

Best for: companies on Otter's Enterprise plan, which Otter's help center says is opted out of AI model training by default.

Otter.ai's privacy policy says it trains its own AI technology on de-identified audio recordings and transcriptions. Its help center says Enterprise workspaces are opted out of AI model training by default.

Pros

• Otter's help center says its third-party AI providers don't train on user data.
• Otter says it is SOC 2 Type II certified.

Cons

• Outside Enterprise workspaces, your recordings and transcripts can be used for training once de-identified.
• HIPAA compliance is an add-on on the Enterprise plan.

Price: the Basic plan is free with 300 transcription minutes a month. Pro is $16.99 a user a month, or $8.33 a month billed yearly. See Otter.ai's pricing page.

Fireflies.ai

Best for: online meetings, for companies that need a SOC 2 Type II report and no AI training on meeting data.

Fireflies.ai's security page says it doesn't use your meeting data to train AI models, and that its AI vendors don't keep that data after processing. Its privacy policy adds that it contractually bars its vendors from training their own models on your personal information.

Pros

• SOC 2 Type II certified and GDPR compliant.
• HIPAA-ready on the Enterprise plan, with Private Storage and a signed BAA.

Cons

• HIPAA needs the Enterprise plan, which is billed yearly only.
• Its Google Play listing describes in-person recording and a notetaker for Zoom, Google Meet and Microsoft Teams, with no mention of phone calls.

Price: free plan. Pro is $10 a seat a month billed yearly ($18 monthly), Business $19, and Enterprise $39, yearly only. See Fireflies.ai's pricing page.

Fathom

Best for: free, unlimited recording of video meetings, once you switch off model training.

Fathom's privacy policy says that, depending on your account settings, it may use de-identified data from your meetings to train its in-house AI models, and that you can opt out there. It says it doesn't let third parties such as OpenAI, Anthropic or Google train on your data.

Pros

• The free plan has unlimited recordings and transcriptions.
• HIPAA compliance with a standard BAA comes with the Team plan.

Cons

• Trains its in-house models on de-identified meeting data until you opt out.
• The Team plan, which carries the BAA, has a 2-user minimum.

Price: free plan. Premium is $20 a month, or $16 a month billed yearly. Team is $19 a user a month, or $15 billed yearly. See Fathom's pricing page.

Granola

Best for: notes from meetings on your own computer, with no bot in the call and no meeting audio kept.

Granola's privacy policy says it trains AI models only on de-identified data, which you can opt out of in your account settings, and that Enterprise workspaces are opted out by default. It doesn't let third parties such as OpenAI or Anthropic train on your data.

Pros

• Doesn't store meeting audio; it keeps only the transcript and your notes.
• On Enterprise, admins can opt out the whole team.
• SOC 2 Type 2 compliant.

Cons

• Outside Enterprise, training on de-identified data is on until you opt out.
• HIPAA-compliant workspaces are on the Enterprise plan only.

Price: the Basic plan is free. Business is $14 a user a month, and Enterprise $35. See Granola's pricing page.

Read AI

Best for: teams that want model training off by default and a SOC 2 Type 2 report.

Read AI's Trust Center says user data isn't used for model training by default. Its Customer Experience Program, which uses your meeting data and connected email and messages to improve the product, is opt-in only.

Pros

• Its Trust Center lists SOC 2 Type 2, HIPAA and GDPR.
• Raw audio and video are deleted within 24 hours.

Cons

• Its privacy policy says it infers demographic characteristics from audio and visual information and uses them to improve its models.
• HIPAA-compliant workspaces are on Enterprise+ only.

Price: free plan with 5 meeting transcripts a month. Pro is $15 a user a month billed yearly ($19.75 monthly), and Enterprise+ is $29.75 billed yearly. See Read AI's pricing page.

tl;dv

Best for: meeting recording hosted in Europe, with no AI training on your content.

tl;dv's privacy policy, published by its company tldx Solutions GmbH, says it doesn't use your meeting recordings, transcripts, notes or files to train generative AI models, and that its AI providers don't either.

Pros

• Its services are hosted in the European Economic Area.
• It says staff don't access your recordings unless you share them for support.

Cons

• Some AI processing can run outside Europe, depending on the hosting location you choose. tl;dv may use Anthropic's models through Google Cloud Vertex AI for written summaries.
• Privately hosted AI is an Enterprise feature.

Price: free plan. Pro is $18 a seat a month billed yearly, and Business $29 billed yearly. See tl;dv's pricing page.

Notta

Best for: transcribing on your own computer with Privacy Mode in Notta Desktop.

Notta's privacy policy says it may use the information it collects to evaluate and improve its service, and doesn't say whether that includes training AI. Its help center lists "No data training" as an Enterprise plan feature, and says recordings made in Privacy Mode aren't used for AI model training.

Pros

• Privacy Mode transcribes on your device and keeps those recordings off Notta's cloud (Pro, Business and Enterprise).
• SOC 2 Type II and ISO 27001.

Cons

• The privacy policy doesn't say whether your cloud recordings are used to train AI.
• Privacy Mode pauses or limits cloud transcription, cloud AI summaries, sync across devices and team collaboration.

Price: free plan with 120 transcription minutes a month. Pro is $8.17 a month billed yearly ($97.99 a year), and Business $16.67 a month billed yearly. See Notta's pricing page.

Prices checked on November 13, 2026.

How to check a note taker's AI training policy yourself

1. Open its privacy policy and its trust or security page. Search them for "train", "model", "machine learning", "de-identified" and "aggregate".
2. Find out whose models are trained. A company can promise that OpenAI or Anthropic won't train on your data and still train its own models. Fathom and Granola say both.
3. Look for the default. "You can opt out" means training is on until you change a setting, so switch it off before your next recording.
4. Check your plan. Otter.ai and Granola turn training off by default only for Enterprise workspaces, and Notta lists no data training only on Enterprise.
5. Look at uploads and feedback too. Files uploaded to Wave AI's free web tools may be used to improve speech recognition, and rating a transcript in Otter.ai can give Otter permission to access that conversation for training.
6. Save a copy of the policy with the date you read it. Policies change.
7. If it's still unclear, ask the company in writing and keep the reply.

What else matters besides AI training

• Ask which SOC 2 report a company has. Wave AI's is Type 1: it covers how controls were designed as of March 1, 2025, with no period of operation tested. Fireflies.ai, Granola, Notta, Otter.ai and Read AI say they have Type II (or Type 2) reports, which matters if your company requires one.
• If you record patient conversations, ask for a signed Business Associate Agreement (BAA) and check which plan includes it. Wave AI signs a BAA with organizations on a Wave Teams plan.
• Look for encryption both in transit and at rest. Wave AI uses TLS 1.2+ in transit and AES-256 at rest.
• Read the terms for who owns your content. With Wave AI, your recordings and transcripts belong to you.
• Check that you can delete a recording for good, how long backup copies last, and whether data already used for training can be taken back out. Granola's privacy policy says de-identified data already built into its AI models may stay there. Wave AI's deletion details are on its security page.

How we compiled this

We read each company's own privacy policy, terms, and trust or security pages on 13 November 2026, and quoted what they say about using your recordings to train AI. Where a company's own pages leave the answer open, we marked it unclear instead of guessing. Plan details come from each company's help center and pricing page. We didn't test the products or audit these claims. Policies change, so check the linked page before you decide. We make Wave; if a fact about another note taker has changed, tell us and we'll update the page.