
AI Meeting Notes: Granola and Similar Tools, Without Forgetting Consent
AI meeting notes transcribe and summarize meetings. Granola records without a bot. How to choose one, what it keeps, what it costs and what consent to ask for.
What do these tools do?
Three pieces, in almost all of them:
- Audio capture, in two ways: with a bot that joins the call, or with the device’s own audio.
- Transcription, which turns audio into text, sometimes identifying who is speaking.
- Summary and actions, which groups what was said into topics, decisions and tasks with an owner.
What changes is the first piece, because it defines the relationship with participants: a bot shows up in the attendee list; device capture does not.
Why does it matter to a designer?
Because research sessions and product meetings produce decisions that get lost. A good note lets you return to a user interview without rereading an hour of audio, and share what was said with the team without rewriting it.
This fits with the transcription and tagging workflow and with interview synthesis. The note does not replace your judgment: the summary is a proposal you must check against what was actually said.
Which tools exist and what do they cost?
According to the official pages checked on November 6, 2026, in US dollars:
| Tool | How it captures audio | Free plan | Paid plans |
|---|---|---|---|
| Granola | Computer audio, no bot; also on mobile | Basic: 30 days of history | Business 14 per user per month; Enterprise 35 |
| Otter | A bot that joins the meeting | Basic: 300 minutes per month per user and up to 30 minutes per meeting | Pro, Business and Enterprise (see its page) |
| Fathom | A bot, or bot-free capture in beta on Mac only | Unlimited recordings and transcriptions, with limited AI use | Premium 20 per month (16 billed annually); Team 19 (15), minimum two users |
| Fireflies | A bot | Free plan | Pro, Business and Enterprise; billed per seat (see its page) |
Granola states that it deletes audio after transcription and that it is SOC 2 Type II certified. The option to keep your data from training models is turned off per user on Basic and Business, and on Enterprise it is managed for the whole workspace. Its help center specifies that on macOS and Windows it does not store meeting audio; on iPhone it keeps it temporarily and deletes it once transcription finishes. Prices change, so check them before you buy.
Which differences matter when choosing?
| Aspect | What to look at |
|---|---|
| How it captures audio | A bot visible on the call, or device capture |
| Free plan | Limits on history, minutes and available AI models |
| Retention | How long the transcript is kept and who manages it |
| Training | Whether you can stop your data from training models, and whether that also covers people who are not users |
| Compliance | Certifications such as SOC 2 and data processing agreements |
What does the law say about consent?
This is the most important point, and the tool does not solve it. In the United States two rules coexist. Federal law (18 U.S.C. § 2511(2)(d)) allows a person to intercept a communication if they are a party to it or if one of the parties gave prior consent, unless it is done to commit a crime or a tort. But some states require the consent of all parties: section 632 of the California Penal Code prohibits recording a confidential communication without everyone’s consent.
That is why capturing without a bot does not remove the duty to disclose. Granola’s own help center recommends making sure participants know when you use an AI notetaking tool, checking with your admin or legal team which meetings may be transcribed, and it offers built-in notices: a watermark on your camera and an automatic message in the meeting chat. Other countries have different laws; check the rules where the participants are.
What has been sued over
Two cases show why it matters. In both, these are allegations, not proven facts.
- Chamberlain v. Granola, Inc. (case 3:26-cv-07926, US District Court for the Northern District of California), filed on July 30, 2026 as a proposed class action. It alleges that Granola is designed so that the other participants do not know they are being recorded and that, by default, it uses the content to train its AI systems, an option only Granola users can turn off and not the people taking part in the meeting. It invokes the federal Electronic Communications Privacy Act (ECPA) and California privacy law, among others. According to the law firm Barnes & Thornburg, Granola had not responded on the merits and no court had ruled.
- Brewer v. Otter.ai, filed in August 2025 in a federal court in California. It alleges that Otter asked permission only from hosts, sometimes not even from them, and that it used recordings to train its models.
Whatever the outcome, they work as a checklist: who knows they are being recorded, who can object, and what the content is used for.
When should you NOT use it?
- In user interviews where you do not have explicit, documented consent.
- For meetings with medical, financial or minors’ information, unless the provider and your company allow it.
- When the automatic summary would replace the team’s decision: a summary has no political context.
- When the team has no retention policy: accumulated transcripts are a risk.
Example
Illustrative example: a designer has a 45-minute interview with a user. Before starting, she announces that the session will be recorded with a notetaking tool, asks for permission and records that consent on audio. Afterward, she reviews the summary, corrects one sentence where the AI misheard a word and deletes the transcript according to her team’s policy. The notes she shares with the team are the ones she validated. The summary saved her time, but the validation was hers.
What should you say when announcing the recording?
The notice does not need to be long, but it must cover three things: what is recorded, why and who can see it. A sentence like “We are going to record this session with a notes tool to capture what you say; only the design team will use it and we will delete it when the study ends. Is that okay?” covers all three. If someone says no, nothing is recorded and the session continues with manual notes.
What should you do with the note after the meeting?
The note is a draft. A workflow that works:
- Check the summary against the transcript at the moments where a decision was made.
- Split three lists: decisions made, open questions and tasks with an owner. If something has no owner, it is not a task.
- Fix the attribution of each decision: who proposed it and who accepted it.
- Share only the validated version, with a note that a person reviewed it.
- Apply your team’s retention policy: delete the transcript when it is no longer needed, or keep it somewhere with controlled access.
If the meeting included an important decision, the responsible person should confirm it in writing, even if the summary is correct.
Common mistakes
- Recording without telling anyone because “there is no bot”.
- Sharing summaries without checking them against the transcript.
- Keeping client transcripts indefinitely.
- Choosing by price without looking at the training and retention policy.
- Using the tool in meetings with sensitive data without reviewing the data processing agreement.
Sources
- Granola: pricing
- Granola: home page
- Granola Help: getting consent
- Otter: pricing
- Fathom: pricing
- Fireflies: pricing
- Cornell LII: 18 U.S.C. § 2511
- Complaint, Chamberlain v. Granola, Inc. (CourtListener)
- Barnes & Thornburg: what the Granola class action means for companies
- Workplace Privacy Report: lessons from the Otter.ai class action complaint