Granola
Provenance unknown Report an errorPanel rating · 6 judges · How to read the stars
Category median
Sovereignty: 3 of 6 dimensions proven
0–5 in half steps. 5 means the rubric's top anchor is met on the evidence.
by Granola Labs Ltd · www.granola.ai
Report an error on this page Is this your product? →
Link to this page
Add ?theme=dark to the image address for dark backgrounds.
The badge shows the category and the month of the latest evaluation, and EU-Made where
that is the product’s provenance. It never shows a score.
Granola at a glance
- Vendor
- Granola Labs Ltd
- Headquarters
- Not stated on the vendor’s pages as of 10 Nov 2026
- Data hosting
- Outside the EU by default source, captured 10 Nov 2026
- Free plan
- Not stated on the vendor’s pages as of 10 Nov 2026
- Panel verdict
- The AI panel rates Granola 3/5, the mean of 6 judges. Granola is a botless AI meeting notetaker on three public tiers — Basic at 0 USD, Business at 14 USD per user per month, Enterprise at 35 USD per user per month. how scores are made
- As of
- Vendor pages captured 10 Nov 2026; evaluated by the panel 11 Nov 2026. all sources
Read this page as one judge. Each weighs the same scores by what they care about.
The panel's verdict
Granola is a botless AI meeting notetaker on three public tiers — Basic at 0 USD, Business at 14 USD per user per month, Enterprise at 35 USD per user per month. Its best scores land on capture and transcription and data control and administration at 6-7: desktop capture takes device audio for any meeting platform without a bot, transcripts auto-delete from one day to one year with workspace policies taking priority over individuals, and an optional camera watermark and automated chat message give visible notice, with start always manual. Sovereignty is the outlier at 3-4: notes live in a US-hosted AWS Virtual Private Cloud with no EU or private-cloud option today, the named providers — Deepgram, Assembly, OpenAI and Anthropic — are listed without processing regions, and the captured pages give different names for the contracting entity, Granola Labs Ltd versus Granola, Inc. Pricing transparency and consent and codetermination trail at 5-6 on unpublished seat definitions, VAT and annual terms and no public information on guest consent before recording starts. Score spreads never top one point; no genuine split emerges.
Speaks for it
- Botless desktop capture works with any meeting app on macOS and Windows, with speaker tags by display name on Google Meet, Zoom and Microsoft Teams
- Capture starts only manually — nothing auto-joins — and an optional camera watermark plus an automated chat message, which workspace admins can require, make transcription visible to everyone in the meeting
- Transcript auto-deletion spans one day to one year with permanent, irreversible deletion, and workspace policies take priority over individual settings
- Audio is never stored on desktop, mobile cached audio is deleted after transcription, and model training is off by default for Enterprise with third parties contractually barred from training
- The developer layer — a documented REST API with rate limits and scoped keys, signature-verified webhooks and an MCP server off by default for enterprises — scored a flat 6 on integrations and workflow
Held against it
- Notes are stored in a US-hosted AWS Virtual Private Cloud with no EU or private-cloud option today, the transcription and AI providers — Deepgram, Assembly, OpenAI and Anthropic — are named without processing regions, and nothing in the captured pages evidences a European processing chain
- We found no public information on supported languages, German included, on phone or dialer calls, or on transcription accuracy per language
- We found no public information on CRM integrations, on Zapier, Make or n8n, or on export to Notion, Confluence or SharePoint, leaving the API as the only bridge
- We found no public information on consent from external participants before recording starts, on a participant opt-out, on works-agreement material or DPIA support, or on an EU AI Act classification
- We found no public information on what counts as a seat, on annual billing or VAT treatment, or on SCIM provisioning, and single sign-on setup routes through sales
Best for
- You want botless capture that works with any meeting app and shows everyone in the meeting visible notice — a camera watermark or automated chat message — while it transcribes.
- Your workspace needs governed retention with stated security basics: transcript auto-deletion from one day to one year with policy override, SOC 2 Type 2, single sign-on via Google Workspace, Microsoft or Okta, and full export of notes, transcripts and account data.
- You have developers who can build your own bridges through the documented REST API, webhooks and MCP server.
- You can start on the free Basic tier at 0 USD, which allows unlimited notes with the last 30 days viewable.
Avoid if
- You require EU data residency or a European processing path — notes sit in a US-hosted AWS Virtual Private Cloud, and the vendor says private cloud is 'not at the moment'.
- You need to import pre-recorded audio files — ask the vendor: the public pages we read do not show it
- You cannot accept training on your anonymized data — it applies unless you opt out, and is off by default only on Enterprise.
- You need all meeting records to expire automatically — auto-deletion covers transcripts, while notes are stored indefinitely by default.
The scores
| Criterion | Panel median |
|---|---|
| Recording, transcription & languages | 3.5 of 5 |
| Summaries, action items & meeting knowledge | 3 of 5 |
| Integrations, API & workflow | 3 of 5 |
| Consent, participant rights & works council | 3 of 5 |
| Retention, AI training & administration | 3.5 of 5 |
| European sovereignty | 2 of 5 |
| Pricing transparency | 2.5 of 5 |
Recording, transcription & languages
Show reasoningHide reasoning
How this is scored
How the conversation gets in and how faithfully it is written down: a bot that joins Zoom, Microsoft Teams, Google Meet and Webex, botless capture from a desktop or mobile app, in-person meetings and phone calls, file upload, live transcription, speaker identification and names, the languages supported with German stated, meetings that switch language, custom vocabulary for names and jargon, timestamps linked to the recording, and editing of the transcript.
0 — Does not record or transcribe meetings; a note template or agenda tool without capture.
3 — A bot that joins one or two conferencing tools and produces a transcript; languages, speaker names, in-person capture and accuracy are basic or not described.
5 — A bot for Zoom, Teams and Google Meet, a desktop or mobile app for in-person meetings, file upload, speaker identification, German among the supported languages, timestamps linked to the recording, and an editable transcript.
8 — Botless capture documented next to the bot, Webex and phone or dialer calls, live transcription, mixed-language meetings handled, custom vocabulary, speaker names learned across meetings, and the transcription engine or its accuracy per language stated.
10 — Capture a regulated team relies on: every channel above with accuracy published per language and method, dialects and specialist vocabulary documented, on-device or customer-hosted transcription as an option, and how errors are corrected and fed back stated.
The Management Consultant
Botless desktop capture that works the same way for Zoom, Google Meet, Teams and any other meeting app, an iPhone and Android app built for in-person meetings, real-time transcription, speaker tags, personal and workspace jargon lists, and editable transcript chunks — the workshop-room capture story is genuinely there. But the pages never state which languages are supported, so I cannot confirm German transcription at all, pre-recorded file upload is stated as unsupported, and we found no public information on phone capture or transcription accuracy. 2 4 5
The Sales Team Lead
The desktop app captures audio straight from the machine with no bot, works identically for Zoom, Google Meet, Microsoft Teams and any other meeting app, transcribes live, and tags speakers by display name on the big three; the mobile apps handle in-person meetings with their own speaker recognition, and personal and workspace jargon lists tune transcription for company terms. The help pages also name the transcription and AI providers. What holds me back is that the pages never state which languages are supported, so I cannot confirm German for my discovery calls, phone or dialer capture is not described, and the pages state that importing pre-recorded audio files is not supported. 4 5 2
The Works Council Chair
Botless capture of the microphone and system audio works with any meeting app on desktop and the phone apps are built for in-person meetings, with speaker tags for Zoom, Google Meet and Teams, personal and workspace jargon lists, and the transcription engine named. Pre-recorded file upload is explicitly not supported, and we found no public information on the supported languages with German stated, on phone or dialer calls, or on accuracy figures per language. 2 4 5 10
The Data Protection Officer
Capture is botless: the desktop app takes device audio for Zoom, Google Meet, Teams and any other meeting app, transcribes in real time, never stores audio, labels speakers by display name with a Me/Them fallback, and supports personal and workspace jargon lists; mobile covers face-to-face meetings and deletes cached audio after transcription, and the transcription providers Deepgram and Assembly are named. We found no public information on the supported languages, German included, on accuracy per language, on mixed-language meetings, or on phone and dialer calls — the pieces my impact assessment would need next. 2 4 5 10
The IT Lead
Botless desktop capture works with any meeting platform including Teams, with speaker tags using display names for Meet, Zoom and Teams, live transcription, editable transcript chunks, and personal plus admin-managed workspace jargon lists — that covers my Microsoft estate well. But the captured pages state that importing pre-recorded audio files is not supported, mobile cannot capture audio from virtual meeting apps, and I found no public information on phone or dialer capture, a supported-language list confirming German, timestamps linked to the recording, or accuracy figures per language. 2 4 5 10
The Skeptic
Botless capture is documented in depth: audio taken from the device with any meeting app, live transcription on macOS and Windows, an in-person mobile app, speaker tags by display name on Google Meet, Zoom and Teams, and personal and workspace jargon lists with stated term limits. The transcription and AI providers are named — Deepgram and Assembly for transcription, OpenAI and Anthropic for summaries. We found no public information on supported languages including German, accuracy per language, timestamps tied to the recording, or phone-call capture, and the vendor states that importing pre-recorded audio files is not supported. 2 4 5
Summaries, action items & meeting knowledge
Show reasoningHide reasoning
How this is scored
What the tool makes of the transcript: summaries and minutes by meeting type with editable templates, decisions and action items with owners and due dates, summaries in the meeting's language or a chosen one, follow-up emails, clips and highlights, search across all meetings, questions answered over the archive with citations back to the transcript passage, shared team folders or channels, and the minutes exported to Word, PDF or a wiki.
0 — No summaries of any kind; only the raw transcript or recording.
3 — One generic summary per meeting; templates, action items, search across meetings and citations are basic or not described.
5 — Summaries with templates per meeting type, action items and decisions extracted, the summary language selectable with German available, search across meetings, questions answered over one or several meetings, and sharing or export of the minutes.
8 — Custom templates per team, action items with owners pushed to task tools, answers that cite and link the transcript passage they rest on, topic or keyword tracking across meetings, team libraries with permissions, and the AI models used for summaries named.
10 — Meeting knowledge a company can audit: every generated statement traceable to its source passage, hallucination handling and human review documented, summary quality by language stated, and the archive queryable through an API or MCP server with the same permissions as in the app.
The Management Consultant
Templates for sales calls, 1:1s and stand-ups plus custom templates, notes structured around people, decisions and intent, chat that queries all my meetings, follow-up email drafts with a review step, and the summarising AI providers named. What I need at midnight and do not find: we found no public information on exporting the minutes to Word or PDF, on action items with owners and due dates, or on answers that cite the transcript passage they rest on. 2 6 7 9
The Sales Team Lead
Notes come out structured with people, decisions and intent, templates exist for sales calls, 1:1s and stand-ups plus custom templates shareable across the organisation, and chat can query all my meetings, rewrite sections, reassign action points and translate notes; the workflow that drafts follow-up emails and sends Slack messages with a review step is exactly what I want after every discovery call. The models behind the summaries are named. But we found no public information on pushing action items into a task tool, on chat answers citing the transcript passage they rest on, or on exporting minutes to Word, PDF or a wiki. 6 7 9 2
The Works Council Chair
Templates per meeting type including custom ones, chat that answers questions across all meetings or a folder, note translation, follow-up email drafts with a review step, and the AI providers named. We found no public information on answers that cite and link the underlying transcript passage, on export of minutes to Word, PDF or a wiki, or on action items pushed to task tools with owners and due dates. 2 6 7 9
The Data Protection Officer
Notes come out structured with people, decisions and intent, templates exist per meeting type plus custom ones, chat answers questions over a single meeting, folders or the whole archive, notes can be translated, and follow-up emails, Slack messages and calendar events are drafted with a review step; the AI providers OpenAI and Anthropic behind the summaries are named. We found no public information on answers citing the transcript passage they rest on, action items pushed to task tools with owners and due dates, or export of minutes to Word, PDF or a wiki — traceability of generated statements is what I cannot yet verify. 2 6 7 9
The IT Lead
Templates per meeting type with custom and organization-shared ones, notes structured with people, decisions and action points, chat that queries all meetings or a folder, note translation, follow-up workflows, and the AI models named (OpenAI, Anthropic, Fable 5.1). I found no public information on answers citing the transcript passage they rest on, action items pushed to task tools, topic tracking, or export of minutes to Word, PDF or a wiki. 2 6 7 9
The Skeptic
Notes are templated per meeting type with custom templates shareable across the organisation, they carry people, decisions and intent, chat answers questions over a single meeting or the whole archive, and workflows draft follow-up emails, Slack messages and calendar events behind a review step. The AI providers behind the summaries are named. We found no public information on answers citing the transcript passage they rest on, export of minutes to Word, PDF or a wiki, or action items pushed with owners into a task tool. 2 6 7 9
Integrations, API & workflow
Show reasoningHide reasoning
How this is scored
How the notes reach the place where work happens: calendar connection with rules for which meetings are recorded, CRM fields filled from the call (HubSpot, Salesforce, Pipedrive, Microsoft Dynamics), project and task tools, Slack and Microsoft Teams channels, Notion, Confluence and SharePoint, Zapier, Make or n8n, a documented REST API and webhooks, an MCP server for AI assistants, and browser extensions and native apps.
0 — No integrations and no export; the notes stay in the app.
3 — A calendar connection and an email or Slack summary; CRM, API and task tools are basic or not described.
5 — Google and Microsoft calendars, a native CRM integration (HubSpot or Salesforce), Slack or Teams, a task or wiki tool, Zapier or Make, and a documented API or webhooks.
8 — CRM field mapping per meeting type, several CRMs including Microsoft Dynamics or Pipedrive, a full REST API with webhooks and reference documentation, an MCP server, SharePoint or Confluence export, and recording rules per calendar or meeting type.
10 — An integration layer a company builds on: every object reachable by API with rate limits and versioning stated, granular OAuth scopes, data residency kept across integrations, and the integrations administered centrally with logs.
The Management Consultant
The published API is unusually solid — endpoints, authentication, rate limits, note-access scopes, workspace keys and webhooks — and an MCP server with official connectors to Claude, ChatGPT, Cursor and more sits on top, with calendar integration and Slack follow-up sending shown. We found no public information on CRM integrations such as HubSpot, Salesforce or Dynamics, on Zapier or Make, or on export to Notion, Confluence or SharePoint, so the notes largely stay inside Granola's own apps. 2 7 8 9
The Sales Team Lead
The developer side is well documented: a REST API with full reference documentation, rate limits, scoped keys and admin-controlled access, webhooks with signature verification and retries, and an MCP server with official connectors for Claude, ChatGPT, Cursor and more, alongside the calendar connection and the Slack and email workflows. For my team's daily flow, though, we found no public information on a CRM integration — nothing on HubSpot, Salesforce, Pipedrive or Microsoft Dynamics — and no Zapier, Make or n8n bridge, so call outcomes and next steps would reach the CRM only through our own API work. Recording is manual by design since the app never auto-joins, and we found no recording rules per calendar or meeting type either. 8 9 7 2
The Works Council Chair
A documented REST API with rate limits, selectable scopes and admin-managed workspace keys, webhooks on the paid plans, and an MCP server with browser-based OAuth that is off by default for enterprises and turned on by admins. We found no public information on CRM connections, Zapier, Make or n8n, or export to Confluence, Notion or SharePoint; beyond Google and Microsoft calendar integration the in-app workflows draft emails and Slack messages with a review step. 2 7 8 9
The Data Protection Officer
There is a documented REST API with reference documentation, rate limits, note-access scopes, admin-managed workspace keys and enterprise scope control, webhooks on Business and Enterprise, an MCP server that enterprises have off by default, calendar sync, and workflow drafts for email, Slack and Google Calendar. We found no public information on CRM integrations such as HubSpot, Salesforce, Pipedrive or Microsoft Dynamics, on Zapier, Make or n8n, or on export to Notion, Confluence or SharePoint. 2 7 8 9
The IT Lead
A documented REST API with rate limits, note-access scopes, admin-managed workspace keys, webhooks with signature verification and retries, and an MCP server that is off by default for enterprises with admin control — the API layer is genuinely solid. But no CRM integration is documented anywhere, and I found no public information on SharePoint or Confluence export, Zapier, Make or n8n, task tools, or browser extensions; workflows reach only email, Slack and Google Calendar, which is thin for a Teams-and-SharePoint company. 2 7 8 9
The Skeptic
The programmatic layer is strong and unusually specific: a documented API with published rate limits, key scopes, admin-managed workspace keys and signature-verified webhooks, plus an MCP server that is off by default for enterprises, alongside calendar connection for Google Workspace and Microsoft and chat workflows for Slack and Google Calendar. We found no public information on CRM integrations such as HubSpot, Salesforce, Pipedrive or Microsoft Dynamics, on task or wiki tools like Notion or Confluence, or on Zapier, Make or n8n — here the API is the only bridge. 2 7 8 9
Consent, participant rights & works council
Show reasoningHide reasoning
How this is scored
Whether every person in the meeting knows and can decide, and whether the tool can be introduced under German co-determination: a recording notice to all participants including external guests, consent or opt-out before the recording starts, a visible bot or a stated notice for botless capture, rules for when the assistant joins automatically, stopping and deleting a recording on request, per-person analytics (talk time, sentiment, engagement scores) off by default or switchable per organisation, material or a template for a works agreement (Betriebsvereinbarung, § 87 (1) no. 6 BetrVG), and support for a data protection impact assessment.
0 — Built to record people without their knowledge, or sells covert recording or individual performance scoring as a feature.
3 — A general privacy statement and a bot that appears in the participant list; notices to guests, opt-out, per-person analytics and works councils are basic or not described.
5 — A recording notice to participants by chat message, email or the bot's name, a way for a participant to stop the recording or opt out, admin rules for when the assistant joins, recordings deletable on request, and per-person analytics described with how to switch them off.
8 — Consent collected before recording starts for external participants, per-person analytics off by default or switchable per organisation, botless capture with its notice method documented, a works-council information pack or works-agreement template published, DPIA support, and the EU AI Act classification stated.
10 — Co-determination built into the product: every analytic and its granularity documented for the works council, consent and opt-out logged per participant, configuration changes logged, minimum group sizes for any team statistics, and references of works-council-approved rollouts published.
The Management Consultant
No bot and no auto-join — every recording starts manually — plus pause mid-meeting, notes private by default, and a real transparency package for botless capture: a camera watermark and an automated chat message visible to all participants, which a workspace admin can require, with transcription on camera-off allowed only after transparency was established. We found no public information on a participant-side opt-out, on per-person analytics such as talk time or sentiment, or on works-agreement material and DPIA support, and the consent page defers to third-party recording-law resources rather than providing a template. 2 10 11
The Sales Team Lead
Botless capture gets real transparency tooling: a camera watermark visible to everyone while transcribing, an automated message in the meeting chat, workspace admins able to require the watermark with a chat-message fallback when cameras are off, and a consent guide that points to recording-law resources and lets you pause transcription mid-meeting. Manual start only — it never auto-joins or auto-records anything — is also a clean default for a works council conversation. But we found no public information on collecting consent from external participants before recording starts, on a participant opt-out, on works-agreement material or DPIA support, or on an EU AI Act classification. 10 11 2
The Works Council Chair
The botless notice method is properly documented: an optional camera watermark and an automated chat message visible to everyone in the meeting, which a workspace admin can require with a chat-message fallback when cameras are off, and transcription starts only when the user manually starts it, can be paused mid-meeting, with transcripts permanently deletable. For our works agreement under co-determination we found no public information on per-person analytics such as talk time or engagement scores and how to switch them off, on a works-council pack or works-agreement template, on support for a data protection impact assessment, or on the EU AI Act classification — and guests have no documented way to object before transcription starts. 2 10 11
The Data Protection Officer
The transparency tooling is real: an optional camera watermark visible only while transcribing, with admin enforcement and an automated chat-message fallback, an automated message in the meeting chat, strictly manual start with no auto-join, mid-meeting pause of transcription, and notes private by default; the vendor also points to third-party recording-law resources while stating this is not legal advice. We found no public information on consent collected from external participants before recording starts, on a works-agreement template or works-council information pack, on data protection impact assessment support, or on an EU AI Act classification — so my Betriebsrat conversation and my pre-recording assessment start from scratch. 2 4 10 11
The IT Lead
Manual start only with no auto-join, a camera watermark visible to every participant that workspace admins can require with an automated chat-message fallback, mid-meeting pause, and edit-or-delete of sensitive content — good transparency tooling for botless capture. But I found no public information on consent collected before recording starts for external participants, per-person analytics and how to switch them off, a works-council pack or works-agreement template, DPIA support, or an EU AI Act classification, all of which I need before our works council signs off. 2 10 11
The Skeptic
The assistant never joins automatically — every capture starts with a manual action — and botless transparency is unusually concrete: a camera watermark visible to everyone while transcribing, an automated meeting-chat message, admin enforcement with a fallback chat message, and the rule that transcription may continue with camera off only after transparency was established in that meeting. The note-taker can pause mid-meeting and edit or delete sensitive content, and a consent guidance page exists. We found no public information on consent collected before recording starts for external guests, per-person analytics and their defaults, a works-agreement template, DPIA support, or an EU AI Act classification. 2 4 10 11
Retention, AI training & administration
Show reasoningHide reasoning
How this is scored
What the employer can govern: retention periods and automatic deletion of recordings, transcripts and summaries, whether audio and video can be discarded after transcription, sharing defaults and link permissions, customer content excluded from model training by contract, the AI and transcription providers named with their region and their own retention, encryption, single sign-on (SAML), SCIM, roles and workspaces, admin policies on auto-recording and sharing, audit logs, export of the whole archive, and security certifications such as ISO 27001 and SOC 2.
0 — No administration and no deletion; every user decides alone and recordings are kept indefinitely.
3 — Users can delete their own recordings and a security page exists; retention policies, model training, SSO and audit logs are basic or not described.
5 — Retention settings per workspace, customer content excluded from model training, team workspaces with roles, single sign-on, admin control over sharing and auto-recording, encryption at rest and in transit, and ISO 27001 or SOC 2 stated.
8 — Automatic deletion by policy with audio discardable after transcription, the AI and transcription subprocessors named with region and zero retention, SAML with SCIM, granular roles, audit logs, legal hold or export of the full archive, and certifications with their scope published.
10 — Governance an auditor accepts: every retention, sharing and recording policy enforceable centrally and logged, models run on the vendor's own or the customer's infrastructure as an option, audit logs exportable to a SIEM, and the deletion of every copy, including at subprocessors, documented.
The Management Consultant
Audio is never stored on desktop and mobile deletes cached audio from Granola and third-party systems once transcription completes; transcripts auto-delete on a rolling basis from one day to one year at individual or workspace level, with enterprise policy overriding individuals; SOC 2 Type 2, encryption, audit logging, SSO including Okta, and a full archive export are stated. The sentence I must read to every client before signing: Granola itself trains on anonymized data unless you opt out, training is off by default only on Enterprise, and subprocessors are barred by agreement from training — close to the written promise I want, but not the default for every plan. 2 10 12 13
The Sales Team Lead
Transcript retention is genuinely governed: rolling auto-deletion at periods from one day to one year, workspace policies that override individual settings, permanent deletion, and audio that is never stored on desktop, with the mobile cache deleted after transcription — the captured pages do give different pictures of whether individuals can set retention periods or only Enterprise plans, but the mechanics are clear. Governance and security are documented too: SOC 2 Type 2, encryption at rest and in transit, SSO via Google Workspace, Microsoft and Okta, infrastructure audit logging, model training on anonymized data off by default for Enterprise with third-party training contractually barred, and a full export on request. We found no public information on SCIM, granular roles, legal hold or the scope of the certifications. 12 2 10
The Works Council Chair
Workspace admins can set transcript retention from one day to a year with the workspace policy overriding individuals, audio is never stored on desktop and is deleted from all systems after mobile transcription, model training is off by default for enterprises and contractually barred for subprocessors, and SOC 2 Type 2, single sign-on, encryption and audit logging are stated. We found no public information on SCIM, legal hold, ISO 27001, or the regions and retention periods of the named subprocessors, and notes are kept indefinitely by default outside enterprise retention periods. 2 10 12 13
The Data Protection Officer
Notes and transcripts are stored indefinitely by default with custom retention reserved for Enterprise, but transcript auto-deletion offers periods from 1 day to 1 year or Off, applies to individuals and workspace admins, and workspace policy overrides personal settings with permanent, irreversible deletion; audio is never stored on desktop and iPhone cached audio is deleted from all Granola and third-party systems after transcription. Subprocessor agreements prohibit model training with enterprise training off by default, audit logging is stated, SOC 2 Type 2 was achieved in July 2025, and full export plus account deletion within 30 days are documented — but the AI and transcription providers are named without region or their own retention, and we found no public information on SCIM, legal hold, or the certification scope. 2 10 12 13
The IT Lead
Enterprise transcript auto-deletion from one day to one year with workspace policies that override individuals, audio never stored on desktop and mobile cache deleted after transcription, model training off by default for Enterprise with third-party training contractually barred, SSO via Google, Microsoft and Okta, SOC 2 Type 2, and named AI and transcription providers. But SCIM is not documented, and I found no public information on admin policies for sharing, admin-facing audit logs of user actions, legal hold, granular roles, or ISO 27001 — for 400 seats the missing SCIM alone costs real onboarding effort. 2 8 9 10 12 13
The Skeptic
Retention is genuinely governed: transcripts auto-delete on a rolling basis with periods from one day to one year, enterprise workspace policies override individual settings, audio is never stored on desktop and mobile caches are deleted after transcription from all Granola and third-party systems, third parties are contractually barred from model training with an opt-out for anonymised-data training that is off by default for enterprises, SOC 2 Type 2 is stated, and a full export of notes, transcripts and account data is available. Notes themselves are stored indefinitely by default and only transcripts auto-delete. We found no public information on SCIM, legal hold, subprocessor regions and retention periods, or admin-facing audit logs of sharing and recording events beyond infrastructure-wide audit logging. 2 10 12 13
European sovereignty
panel opinion
Show reasoningHide reasoning
How this is scored
Who the contracting entity and the parent company are, where recordings, transcripts, summaries, voice profiles and account data are stored and processed, which language model and transcription providers read them and where, and who the other subprocessors are (hosting, email, support, analytics and payment tools). Independently sourced by the sovereignty pipeline. A conferencing platform outside Europe that the customer already uses is part of the meeting, not counted against the notetaker; where the notetaker stores and processes the conversation is.
0 — Non-EU vendor and contracting entity, hosting unstated, subprocessors unnamed, and no data processing agreement offered.
3 — A privacy policy and a data processing agreement exist, but hosting location is unstated or non-EU and the subprocessors are not listed.
5 — EU hosting available and an EU contracting entity or EU representative, but parts of the chain (language models, transcription, email, support, payment) run through non-EU providers without a stated safeguard.
8 — EU hosting on named infrastructure, EU contracting entity, subprocessors including the AI providers listed with their locations, a data processing agreement published, and security certifications (ISO 27001, SOC 2) stated.
10 — Sovereign end to end and evidenced: European ownership, entity, hosting and every subprocessor, with transcription and language models run in Europe by European operators, data export and deletion documented, and certifications published with their scope rather than implied.
The Management Consultant
Notes are stored in a US-hosted AWS Virtual Private Cloud with no EU option and private cloud stated as unavailable today, and the AI chain runs through OpenAI, Anthropic, Deepgram and Assembly with no stated regions; the pages also give different figures for the vendor's legal entity (Granola Labs Ltd versus Granola, Inc. in the copyright line), so jurisdiction stays open. GDPR commitment, SOC 2 and a DPA available on request are stated, and the transcription and AI providers are at least named — but nothing we read shows a European processing path, which for my German client conversations is the disqualifying point. 2
The Sales Team Lead
This is an openly non-European chain: notes are stored in a US-hosted AWS virtual private cloud, there is no private-cloud or EU-region option today (the vendor says it may consider this as a future enterprise feature), and the named transcription and AI providers — Deepgram, Assembly, OpenAI and Anthropic — are listed without processing regions. GDPR commitment, a data processing agreement on request and SOC 2 Type 2 are stated, but we found no public information on an EU contracting entity or representative, and where the AI providers read my calls is not documented. For my German prospects, everything visible points to the conversation being processed outside Europe. 2
The Works Council Chair
Notes are stored in a US-hosted AWS virtual private cloud with no EU hosting or private-cloud option today, the contracting entity is Granola, Inc., and a data processing agreement is available on request. The transcription and AI providers (Deepgram, Assembly, OpenAI, Anthropic) are named but their locations are not, so nothing in the captured pages evidences a European processing chain. 2
The Data Protection Officer
Notes are stored in a US-hosted AWS Virtual Private Cloud and a private-cloud option is explicitly deferred to a possible future enterprise feature, so no EU hosting option exists today; the captured pages also give different names for the contracting entity, Granola Labs Ltd versus Granola, Inc. GDPR commitment, a Data Processing Agreement available upon request, SOC 2 Type 2 and penetration testing are stated, and OpenAI, Anthropic, Deepgram and Assembly are named as the AI and transcription providers — but without locations, and we found no public information on a full subprocessor list, a transfer impact assessment, or an EU contracting entity or representative. 2
The IT Lead
Notes and transcripts are stored in a US-hosted AWS Virtual Private Cloud, SaaS only with no EU or private-cloud option, the subprocessors Deepgram, Assembly, OpenAI and Anthropic are named but without stated regions, the data processing agreement is available only upon request, and the contracting entity's jurisdiction is unknown to me. For a German engineering services firm this US-hosted, US-provider chain is the weakest link in an otherwise tidy product. 2
The Skeptic
As documented, the whole chain sits in the United States: notes live in a US-hosted AWS Virtual Private Cloud, the entity named on the security page is Granola, Inc., and the providers reading the conversation — OpenAI, Anthropic, Deepgram and Assembly — are named, with no EU or customer-run hosting option today and the data processing agreement available only upon request. Naming the subprocessors and publishing SOC 2 Type 2 with regular penetration testing is more than many vendors manage. We found no public information on an EU contracting entity or representative, on where the AI and transcription subprocessors process the data, or on transfer safeguards such as standard contractual clauses. 2
Pricing transparency
Show reasoningHide reasoning
How this is scored
Whether a buyer can compute the real annual cost from public pages alone: price per seat and tier, what counts as a seat (every invited user, only recording users), transcription minutes or credits included, storage limits, which features (CRM integrations, API access, SSO, custom retention, EU hosting, AI credits) cost extra, the free tier and its limits, monthly versus annual billing, minimum seats and VAT.
0 — No public prices at all; every customer is a sales call.
3 — A starting price or a free tier exists, but how cost scales with seats, minutes or features is unclear.
5 — Tiers are public with their seat price and main inclusions, but at least one commonly needed piece (SSO, CRM integration, API, custom retention, EU hosting, minute or credit limits) is unpriced or "on request".
8 — Every tier priced publicly with its scaling rule (seats, minutes, credits, storage), add-ons priced, what counts as a seat stated, overage handling stated, minimum term and VAT treatment stated.
10 — Complete price computability: annual cost derivable for a given team size and meeting volume, with every add-on, overage and support fee and the cancellation and data-return terms published.
The Management Consultant
Three tiers are public — free, Business at $14 per user per month, Enterprise at $35 per user per month — with the free tier's limit documented (notes older than 30 days need a paid plan) and several gates clear: API keys and webhooks on Business and Enterprise, custom retention periods on Enterprise. We found no public information on what counts as a seat, on an annual billing option, on VAT treatment, or on the price of SSO, which the pages route to the sales team — so I cannot compute a client's real annual cost from public pages alone. 1 7 8 12
The Sales Team Lead
All three tiers carry public per-user prices — Basic at 0 USD, Business at 14 USD per user per month, Enterprise at 35 USD per user per month — and the plan boundaries are documented: unlimited notes but only 30 days of history on the free tier, full history and model choice above, API keys and webhooks from Business, admin-controlled MCP on Enterprise. However, we found no public information on what counts as a seat, on annual versus monthly terms, on VAT treatment, on minute or storage limits, or on SSO pricing. So I cannot compute the real annual cost for my eight reps from these pages alone. 1 7 8 9
The Works Council Chair
The three tiers are public with seat prices — Basic free, Business 14 USD and Enterprise 35 USD per user per month — and the free tier's 30-day history limit plus the plan gating of API keys, webhooks and model choice are documented. Single sign-on setup runs through the sales team, and we found no public information on annual billing, VAT treatment, minimum seats, or what exactly counts as a seat. 1 7 8 9
The Data Protection Officer
All three tiers carry public per-user monthly prices — Basic at 0 USD, Business at 14 USD per user per month, Enterprise at 35 USD per user per month, all shown with monthly billing — and the free tier's limits are stated (last 30 days of notes, Auto model only), with API keys and webhooks tied to Business and Enterprise. Single sign-on setup routes through sales, and we found no public information on annual billing, what counts as a seat, minimum terms or VAT treatment, so the real annual cost for my desk is not computable from the captured pages. 1 7 8 9
The IT Lead
Three public tiers — Basic free, Business at $14 per user per month, Enterprise at $35 per user per month — with unusually clear free-tier limits (unlimited notes, last 30 days) and per-plan differences in note history, chat models, API and webhooks, and MCP coverage spelled out. What is missing for computing a real 400-seat annual cost: what counts as a seat, VAT treatment, minimum term, annual versus monthly differences, any add-on pricing, and SSO setup is routed to sales. 1 7 8 9 2
The Skeptic
Three tiers are public with exact prices quoted — Basic at $0, Business at $14 per user per month, Enterprise at $35 per user per month — and the free tier's limits are stated plainly: unlimited notes but only the last 30 days viewable and chat on Auto only, with plan-level mapping for MCP, webhooks, API keys and model choice. Only monthly billing is shown, and setting up single sign-on goes through sales. We found no public information on annual pricing, VAT treatment, storage or minute limits, or minimum seats. 1 2 7 9
European sovereignty — proven facts
3 of 6 dimensions provenBuilt only from facts shown on the vendor's own pages. A dimension we could not prove is left open, not scored as zero.
| Legal entity | Not determined | — | uncited Report an error |
|---|---|---|---|
| Hosting location |
Outside the EU
unverified
Why unverified?The statement covers note storage only; transcription and AI subprocessors process data too, and no statement addresses where all other data categories sit. |
0/4 pts | 2 Report an error |
| Hosting options |
Cloud service only
unverified
Why unverified?The page answers a private-cloud question rather than saying "self-hosted/on-premises" outright, and hints a private-cloud enterprise feature may come later, so no customer-run or EU-region option exists today. |
0/1 pts | 2 Report an error |
| Open source | Not determined | — | uncited Report an error |
| Data processing agreement |
On request
unverified
Why unverified?The footer of the same page also lists a Data Processing Agreement link, but its content is not in the excerpts, so only the upon-request statement can be relied on. |
0/1 pts | 2 Report an error |
| Transfer impact assessment | Not determined | — | uncited Report an error |
Questions about Granola
Is Granola European?
Not determined: the vendor’s pages as captured do not say where Granola is based.
Where does Granola store data?
Granola stores customer data outside the EU by default, per the vendor’s page captured 10 Nov 2026.
Does Granola have a free plan?
Not stated on the vendor’s pages as of 10 Nov 2026.
What are the best alternatives to Granola?
As of 18 Nov 2026, the panel rates MeetGeek (3.5/5), Noota (3.5/5) and Jamie (3.5/5) highest among the other 23 AI Meeting Assistants & Notetakers products evaluated. Ratings that look equal are rounded to half stars; the order uses the unrounded rating, then the name. The highest-rated European alternative is MeetGeek (EU-Made, Romania). Ratings are the AI panel’s opinion, formed from public evidence: how scores are made. The full list, with European products first: Granola alternatives.
Where this could be wrong
- Evidence ages. The oldest capture behind this page is from 10 Nov 2026. Vendors change pricing and policies without notice; every fact reflects its source as of the capture date shown in the registry.
- Weak sourcing — Data processing agreement. The footer of the same page also lists a Data Processing Agreement link, but its content is not in the excerpts, so only the upon-request statement can be relied on.
- Weak sourcing — Hosting location. The statement covers note storage only; transcription and AI subprocessors process data too, and no statement addresses where all other data categories sit.
- Weak sourcing — Hosting options. The page answers a private-cloud question rather than saying "self-hosted/on-premises" outright, and hints a private-cloud enterprise feature may come later, so no customer-run or EU-region option exists today.
- AI can misread a source. Extraction and judgement are automated; a citation guarantees traceability, not infallibility. If something here is wrong, say so — no account needed, every report is decided within 5 business days, and accepted corrections are published.
What we left out
A claim that does not survive our checks costs us the claim, not the page. This is what was taken off this one.
- 18 product facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 6 compliance facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 integrations facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 1 hosting fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 legal fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 pricing fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 support fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
Sources (13)
The pages every claim on this page was read from — each one checked, dated, and kept verifiable.
- 1 Vendor page www.granola.ai Checked 10 Nov 2026 Details →
- 2 Security / trust page — found from the homepage www.granola.ai Checked 10 Nov 2026 Details →
- 3 Open-source licence page — found from the homepage www.granola.ai Checked 10 Nov 2026 Details →
- 4 Recording, transcription & languages — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 5 Recording, transcription & languages — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 6 Summaries, action items & meeting knowledge — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 7 Summaries, action items & meeting knowledge — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 8 Integrations, API & workflow — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 9 Integrations, API & workflow — found from sitemap www.granola.ai Checked 10 Nov 2026 Details →
- 10 Consent, participant rights & works council — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 11 Consent, participant rights & works council — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 12 Retention, AI training & administration — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →
- 13 Retention, AI training & administration — found from sitemap docs.granola.ai Checked 10 Nov 2026 Details →