Does Granola train on your data?
Granola trains AI models on de-identified meeting data with an opt-out in settings, and bars OpenAI and Anthropic from training on your data at all.
A self-serve opt-out exists
"We only use de-identified data to train AI models, which you can opt-out of within your Granola account settings ."
- 14 Sept 2026
- 5 Oct 2026
- All plans
- global
- Fetched from the company
- 28320CB3…C1A5
How we read it: Graded C rather than D because the control is named as a setting in your account rather than deferred to an article, and the scope is limited to de- identified data. The policy is also unusually frank about what de- identification does not undo: it says individual data points may be "incorporated into aggregate model weights and parameters during the training process", which is not reversible by later opting out. The separate bar on OpenAI and Anthropic training is a real protection for a product built on their models.
opt-out, opt out, opted in, flagged, Feedback, Enterprise
What the document does offer
- The clause's setting, "which you can opt-out of within your Granola account settings"
- The policy's list of what is "Collected or Received for AI Model Training" puts user data last and conditionally — "When opted in, data that our Users or workers provide, including inputs and outputs from our Services, that is de-identified and aggregated with other data" — after two sources the setting does not reach: "Materials flagged for safety, security, or policy review" and "Feedback that Users explicitly provide about our Services"
- "Enterprise Workspace Products have admin-enforced settings and are opted-out by default" — the one plan where the default runs the other way
Why none of it applies: The toggle is real and named in the policy, and it governs the de-identified use of what you provide. Two sources sit outside it by the policy's own list — material flagged for safety or policy review, and feedback you give explicitly — the same shape of carve-out Anthropic's consumer tier carries. Enterprise workspaces are opted out by default; everyone else is opted in until they act.
What each plan can refuse
The same questions, asked of every plan we could identify. Every cell comes from a quoted clause or a documented search - see the evidence above.
- Self-serve toggle
- Not stated
- 32 / 100
- Yes, stated
- Unknown
How to opt out of Granola AI training
Plan by plan, from what Granola's own policy says about refusing. Where it publishes the steps, they're listed here with the limits they carry. Where there's nothing to switch off, or nothing graded, this says so instead of inventing a menu path.
All plans
Yes. Granola has a self-serve setting.
- Open Settings, then Preferences.
- Find the setting "Use my data to improve models for everyone".
- Turn the toggle off.
The help page rewritten on 21 September 2026 names the setting and where it lives; the earlier version said only that every plan could opt out in settings. Material flagged for safety review and feedback you give explicitly sit outside it, by the policy's own list.
What this policy has done since we started watching
2 changes detected since 14 Aug 2026. Each one is a diff between two archived copies, and each links to the passage that moved.
Vendor response
We haven't yet written to Granola about this verdict, and they haven't disputed, clarified or announced a change to any clause quoted above. When either happens it's published here, word for word.
Report a wrong clause