Granola

What is Granola?

Granola describes itself as an AI notepad rather than an autonomous meeting bot. It runs on the user's computer, captures microphone and system audio, builds a transcript, and enhances notes that the user writes during the conversation. The result is a human-guided document: the participant decides what matters, while the transcript supplies missing context.

This model is especially attractive for product interviews, discovery calls, hiring conversations, and internal decisions where a fully automatic chronological summary is less useful than a selective record. It also changes the risk. No bot appears in the participant list, so transparent consent depends on the user and configured notification.

Human-in-the-loop note workflow

A user opens Granola, selects or creates a note, and starts capture manually. The app listens to microphone and system audio without joining the meeting. During the discussion, the user can jot terse points. Granola later expands those notes using the transcript and a chosen template. Original notes and AI-added text remain visually distinguishable.

Templates can structure output for one-to-ones, customer interviews, sales discovery, project reviews, and other recurring calls. Granola Chat can answer questions within and across meetings the user can access. Shared folders, integrations, API access, and MCP extend the knowledge into other tools.

Human guidance is not automatic verification. The model can still attach context to the wrong note, mishear a speaker, or produce a confident action item. Review quotations, names, technical terms, dates, amounts, commitments, and owners. Preserve the transcript until the note is approved when exact wording matters.

Consent and capture visibility

Granola's consent guidance says users are responsible for obtaining consent and recommends always informing participants. On macOS, an optional automatic consent message can post a customizable disclosure at the start of Zoom and Google Meet sessions.

That feature does not cover every platform, device, in-person meeting, or participant state. A chat message can be missed, and bot-free capture looks like a normal call. Add notice to the invitation, state it aloud before starting, record any required affirmative response, and stop Granola when someone declines. Enterprise can add an organization-wide notification pilot, but legal and policy review still belongs to the customer.

Capture is manual rather than completely automatic, which lowers accidental recording risk. However, the app can continue transcribing if the meeting runs over or the user forgets to end it. Make “click End and confirm the note title” part of the meeting closeout routine.

Audio, transcript, sharing, and deletion

Granola's security and data FAQ says it does not retain audio recordings. Audio is temporarily cached during transcription and deleted from Granola and third-party services after processing. This reduces the raw-voice-data surface, but the transcript and generated notes still contain the substance of the conversation.

Notes are private by default, including on the free tier. Users can share with named people, company members, folders, or anyone with a link depending on settings. Administrators cannot open an individual's private notes merely because they administer the workspace, although Enterprise can cap maximum sharing and API access.

Transcript auto-deletion is available to individual users and can be enforced workspace-wide by Enterprise. Choices range from one day to one year or Off. Individual changes have a one-week cooldown; an Enterprise policy can immediately remove existing transcripts older than its period. Deletion is permanent.

Notes remain after transcript deletion. This is useful for minimization, but it creates important side effects: notes cannot be regenerated properly, and chat answers are limited to retained note content. Decide when a note is final, then delete the transcript according to policy. Test folders, shared links, integrations, API copies, and MCP clients separately.

Model training and security controls

Granola says OpenAI and Anthropic are contractually prohibited from training on customer data. On Basic and Business, anonymized data may be used to improve Granola's own models by default. Every user can turn off “Use my data to improve models for everyone.” The provider cannot guarantee that historical anonymized data was not used before opt-out, but future use stops after the change.

Enterprise can apply the opt-out across the organization and adds SSO, sharing restrictions, transcript retention, and usage administration. Authentication on other plans is delegated to Google or Microsoft rather than a built-in password and MFA system, so enforce MFA at the identity provider.

Plans and price

The official Granola pricing page lists Basic at no cost, Business at $14 per user per month, and Enterprise at $35 per user per month at the review date. Basic provides AI notes, limited history, chat, shared folders, templates, languages, and an individual training opt-out. Business adds unlimited history, advanced models, integrations, centralized billing, MCP, and API access. Enterprise adds organization controls, SSO, support, analytics, auto-deletion, sharing controls, and an organization-wide training opt-out.

Verify history limits, platform availability, integration entitlements, and billing immediately before purchase. A free evaluation should include the same consent and data settings planned for production.

Who should choose it?

Granola is a strong fit for people who actively take notes and want AI to improve them without changing the meeting's participant list. It is less suited to autonomous capture, stored recordings, or organization-wide controls on a low-priced plan.

Compare Fathom for automatic capture and a generous individual plan, Notion AI when the output belongs in a broader workspace, and Otter.ai for live transcription and playback. See the best AI meeting assistants guide for the full framework.