Granola AI Notetaker Review for Lawyers
Granola has become one of the most talked-about meeting notetakers in the general productivity market. It records what you say, transcribes it, and turns your own shorthand notes into a polished summary. It is genuinely good software. The question this review answers is a narrower one: does the way Granola handles audio, transcripts, and summaries fit the confidentiality obligations of a practicing lawyer?
Disclosure: Basil, which publishes this review library, competes with Granola in the meeting-notetaker category. I've tried to keep this fair — Granola does a lot of things well, and I'll say so. If you want to compare architectures side by side, our own approach is documented at Basil for Law.
What Granola Actually Does
Granola is a desktop app (macOS and Windows, with iOS available) that sits in the background during a meeting. Unlike bot-based notetakers, it does not join the call as a separate participant — it captures system audio and your microphone locally on your machine. You take rough notes during the meeting; when the call ends, Granola sends the transcript and your notes to a large language model, which produces a structured summary in a template you choose. See the vendor's own description on the Granola homepage and the Granola security page.
The interaction model is the differentiator. Most notetakers dump a wall of transcript on you. Granola treats your own bullet points as the spine of the meeting and uses the transcript as raw material to flesh them out. Lawyers who take contemporaneous notes during client calls, depositions prep, or witness interviews will find the workflow intuitive. The templates system lets you build custom output shapes — an intake memo, an issue list, a follow-up email draft — and Granola will consistently produce that shape.
Search across past meetings works well. The app integrates with Google Calendar, and there is a folder concept that groups meetings. For a solo or small-firm lawyer running a lot of intakes and status calls, the UX is a real productivity gain.
The Confidentiality Lens
This is the section that matters most for the intended audience. Granola's architecture, per its own security documentation and privacy policy, works roughly like this:
- Audio capture is local. Granola records audio on your device. That is genuinely better than a Zoom/Meet bot that streams audio to a vendor server for transcription.
- Transcription is cloud-based. Audio is sent to a third-party speech-to-text provider for processing. The audio is not, by default, retained by Granola after transcription (per the security page), but it does leave your machine.
- Summarization uses third-party LLMs. Granola sends transcripts and your notes to model providers (OpenAI and Anthropic are named in the security page) to generate the summary. Granola states it has enterprise agreements with those providers under which prompts are not used to train models.
- Transcripts and notes are stored in Granola's cloud so they are searchable across devices.
What this means practically:
A subpoena served on Granola could reach transcripts, generated summaries, and metadata about your meetings. A subpoena served on the downstream model providers could, in theory, reach whatever logs those providers retain under their enterprise terms — typically short-window abuse-monitoring logs, but non-zero. This is not unique to Granola; it is true of nearly every cloud notetaker. It is qualitatively different from a tool that processes everything on your device.
The US v. Heppner ruling (S.D.N.Y. Feb 2026, Judge Rakoff) is worth reading in this context: the court held that a litigant's chats with a public AI platform were not privileged because a third party — the platform — had access to the content. The reasoning in Heppner concerned a consumer chatbot, not a vendor with an enterprise agreement, and the doctrinal reach to cloud notetakers is not yet settled. But the direction of travel is clear: courts are willing to treat AI vendors as third parties for privilege analysis, and lawyers should assume any content that lives on a vendor server is potentially discoverable through the vendor.
Granola offers a Business plan with a Data Processing Addendum. I have not been able to confirm from public sources that Granola offers a BAA for HIPAA-covered workflows — if you handle health information, ask them directly before uploading anything. Check the current security page for the latest on DPA and enterprise controls.
On training: Granola's privacy policy states it does not use customer content to train its own models. That is an important commitment and it is stated plainly. The dependency on external model providers is governed by those providers' enterprise agreements rather than by Granola directly.
Ethics-Opinion Fit (ABA 512)
ABA Formal Opinion 512 (July 2024) is the starting point. It does not ban cloud AI tools, but it requires lawyers to make reasonable efforts under Model Rule 1.6 to prevent unauthorized disclosure — which the opinion frames as understanding how the tool handles inputs, whether they are used to train models, how they are stored, and who at the vendor can access them.
Granola gives you enough documentation to do that analysis. The security page names the sub-processors, describes the retention posture, and states the training commitment. That is more transparent than many competitors. A lawyer choosing Granola after reading that documentation is making an informed choice, which is what 512 calls for.
The harder question is client consent. NYC Bar Formal Opinion 2025-6 (December 2025) and the trajectory of state opinions increasingly suggest that using an AI notetaker on client communications should be disclosed to the client, and in some contexts specifically consented to. If your engagement letter does not currently address AI-assisted note-taking, that gap matters more than the choice of vendor.
The Brewer v. Otter.ai litigation (2025) is a useful cautionary tale about notetakers that join calls as participants and record other attendees without clear consent — Granola's non-bot architecture reduces that specific risk because it does not silently appear in a call as a third party. That's a real design advantage.
The June 2026 Incident
I want to be careful here. There has been discussion in the productivity-tools community about a confidentiality incident involving Granola in June 2026. As of the time of writing, I have not been able to independently verify the details from primary sources I would trust to cite. Rather than repeat unconfirmed claims, I'll say only this: before rolling any cloud notetaker out across a practice, check the vendor's security page and incident-history disclosures directly, and ask your account rep for a written summary of any material incidents in the last 24 months. That is a reasonable diligence step for any tool in this category, not just Granola.
Pricing and Who It's For
Granola publishes current pricing on its pricing page, including individual, Business, and Enterprise tiers. The Business tier is the relevant one for firms that need a DPA and administrative controls. I'd rather point you at the live page than quote a number that may drift.
Granola is a strong fit for:
- Lawyers whose meetings are largely internal — team standups, business development calls, vendor meetings — where the confidentiality stakes are lower.
- Transactional or advisory practitioners whose client meetings involve information the client is comfortable having processed by a cloud AI vendor under a DPA.
- Lawyers who already use AI heavily and have updated their engagement letters and client disclosures accordingly.
It is a weaker fit for:
- Criminal defense, matrimonial, or other practices where client conversations routinely contain content the client would not want on any third-party server.
- In-house or regulated-industry lawyers whose employer information-security policies prohibit sending call content to external SaaS.
- Lawyers whose clients have explicitly declined AI processing.
For those groups, a device-local architecture — where audio, transcript, and summary never leave the laptop — is the more conservative choice. That is the design our own product, Basil, is built around, and I'll disclose that bias openly.
Verdict
Granola is genuinely one of the best meeting notetakers in the general market. The note-plus-transcript interaction model is smart, the templates are useful, and the documentation of its data handling is more transparent than most competitors'. For lawyers, the question is not whether Granola is well-built — it is — but whether cloud processing of client conversations fits your matter mix, your engagement letters, and your view of ABA 512 and the emerging state opinions.
| Pros | Cons |
|---|---|
| Excellent notes-first UX and template system | Transcripts and summaries stored in vendor cloud (subpoena surface) |
| Does not join calls as a bot participant | Depends on third-party LLM providers for summarization |
| Clear "no training on customer content" policy | BAA availability for HIPAA-covered workflows not clearly documented publicly |
| Named sub-processors and reasonable transparency | Cloud architecture is a poor fit for the most sensitive client conversations |
| Strong search and cross-meeting knowledge | Requires client disclosure/consent hygiene most engagement letters don't yet have |
This review is for information only and is not legal advice.
Frequently asked questions
Is Granola safe to use for client meetings?
It depends on the matter and the client. Granola stores transcripts and summaries in its cloud and uses third-party LLM providers for summarization, which means client content leaves your device. That is a defensible choice for many advisory and transactional contexts under a DPA and with client disclosure, but a poor fit for highly sensitive matters where a device-local tool is more conservative.
Does Granola train AI models on my meetings?
Granola's privacy policy states it does not use customer content to train its own models, and it says its enterprise agreements with LLM providers prohibit training on prompts. Confirm the current terms on Granola's privacy and security pages before relying on this for a specific matter.
Could a subpoena served on Granola reach my client transcripts?
Yes, in principle. Any content stored on a vendor's servers is potentially reachable through legal process directed at the vendor. This is a general feature of cloud SaaS, not unique to Granola, but it is a material consideration under ABA Model Rule 1.6 and Formal Opinion 512.
How does Granola compare to bot-based notetakers like Otter or Fireflies?
Granola's key architectural difference is that it does not join the call as a separate participant. It captures audio locally from your device. That reduces the risk of surprising other attendees with a recording bot, which was central to the Brewer v. Otter.ai litigation in 2025. The cloud-processing question still applies.
Do I need to tell clients I am using Granola?
Increasingly, yes. NYC Bar Formal Opinion 2025-6 and the general trajectory of state ethics guidance point toward disclosing AI-assisted note-taking to clients, and in some contexts obtaining consent. Update your engagement letter before deploying any notetaker across a practice.
What is the on-device alternative if Granola is not the right fit?
If the confidentiality profile of your matters requires that audio and transcripts never leave your machine, you want a fully on-device notetaker. Basil, the tool that publishes this review library, is built for that use case; there are also open-source local transcription workflows worth exploring.
Meeting notes with no server to subpoena
Basil transcribes and summarizes entirely on-device — privilege-safe by architecture. See Basil for Law →
This review is for information only and is not legal advice.