ChatGPT security for meeting notes: What leaves your workspace when you connect Granola

TL;DR: The Granola MCP connector gives ChatGPT access to your personal notes (notes you own, notes shared directly with you, and notes in private folders shared with you) plus public Team Space notes on Business plans, scoped by plan tier, and never to raw audio. When ChatGPT queries the connector, the notes relevant to your question are sent to ChatGPT to generate a response. Whether ChatGPT trains on that data depends on your ChatGPT plan.

Back-to-back meetings create a documentation gap: you are either present or keeping up with your notes. Granola captures and enhances your notes so you can stay focused. Connecting Granola to ChatGPT via the MCP connector goes further, letting you ask one question across months of meetings and get an answer drawn from your notes, instead of reopening each note individually.

Before connecting, there are three things to confirm: what your meeting notes are exposed to, whether ChatGPT trains on them by plan, and how to protect confidentiality.

What the connector exposes, and what it does not

When you connect Granola to ChatGPT, you are not uploading a file or pasting a transcript. You are establishing a connection between your Granola MCP server and ChatGPT's interface. This section explains what data moves across that connection and what stays in your workspace.

What data Granola shares with ChatGPT

The connector exposes personal notes (notes you own, notes shared directly with you, and notes in private folders you can access), plus public Team Space notes on Business plans. Notes outside your plan's date range are excluded.

Granola gives Basic plan users 30 days of meeting history through MCP. Business and Enterprise plans add full history and transcript access. The specific data fields transmitted include meeting notes and transcripts (where available by plan). The MCP server reads your notes from Granola's backend and exposes them to AI tools through a read-only connection.

This differs fundamentally from copy-pasting raw transcripts into ChatGPT. When you paste a transcript, you lose control over what ChatGPT receives. When you use the MCP connector, access is scoped by plan and OAuth authentication.

ChatGPT data handling by plan

ChatGPT retains API inputs and outputs for abuse monitoring, then deletes the data unless legal obligations require longer retention, according to ChatGPT's API data usage policy. When Zero Data Retention (ZDR) is enabled on OpenAI's API Platform, customer content is excluded from abuse monitoring logs for eligible endpoints with a qualifying use case.

How to opt out of model training

1. For ChatGPT Free, Plus, and Pro: Navigate to Your Profile > Settings > Data Controls > Improve the model for everyone > Switch off the toggle. Once you opt out, new conversations will not be used to train models, as explained in ChatGPT's data controls guide.
2. For API users: By default, ChatGPT does not use data submitted via the API for training or improving models. If you want to help improve models, you can opt in through the API dashboard, per ChatGPT's API data usage policy.
3. For Enterprise and Business plans: ChatGPT does not use provided inputs and outputs to train models by default, according to ChatGPT's enterprise data commitment.

ChatGPT receives only the notes relevant to your question

Understanding the data flow requires separating the MCP connector (the bridge) from the underlying ChatGPT data policies on the other side of it. Your meeting notes are cached locally on your device and stored in a secure cloud on AWS servers in the United States, regardless of whether the connector is active.

When ChatGPT queries the connector, the notes relevant to your question are sent to ChatGPT to generate a response. Granola's copy of those notes stays in your workspace. What you are controlling is which notes the connector is permitted to expose, not whether data moves.

Two separate relationships: Granola has enterprise agreements with AI providers (such as OpenAI and Anthropic) that prohibit those providers from training on data Granola submits as part of its own backend processing. This is distinct from your personal ChatGPT account querying your notes via the MCP connector, which is governed by your own ChatGPT plan's data policy, not Granola's subprocessor agreements. Whether ChatGPT trains on notes returned through your personal connector depends on your ChatGPT plan.

ChatGPT can access only what the MCP connector exposes. The connector exposes your meeting notes and transcripts (by plan), but it does not expose your raw audio. The MCP server provides access to your meeting notes and transcripts, including attendees, from your Granola workspace.

This scoped access is enforced through OAuth 2.0 authentication and plan-level restrictions.

Enterprise plans: Centralized data governance

If your organization is on Granola's Enterprise plan, these controls are managed by your IT or security admin. The relevant question is whether MCP has been enabled for your workspace.

The Enterprise plan gives IT and security admins org-wide control over MCP access. Admins can enable or disable MCP across the organization. SCIM provisioning automates user provisioning and deprovisioning at scale, syncing accounts between the identity provider and Granola. When an account is deactivated in the identity provider, access is removed automatically without a separate admin action.

Granola stores meeting notes and transcripts in a secure cloud until you delete them or an auto-deletion policy applies. Enterprise plans include org-wide auto-deletion periods so data doesn't persist beyond your policy.

Participant transparency when using AI tools

Granola encourages transparency with meeting participants. Before running meetings with Granola active, let participants know at the start of the call. Granola offers two built-in options: an automated chat message that posts a notification when transcription starts, and a video watermark that displays a visible indicator throughout the call.

Querying those notes via ChatGPT is a separate step that happens after the interview. Granola's built-in transparency tools, the automated chat message and video watermark, cover the transcription step. What happens to notes after the meeting is governed by your ChatGPT plan's data controls and the access scoping enforced by the connector.

Whether ChatGPT trains on your notes depends on your plan, not the connector

Managing ChatGPT data by plan

Consumer plans (Free, Plus, Pro) may retain conversation content and use it to improve models unless you opt out before connecting. Enterprise, Business, Edu, and API plans do not train on your data by default. Check your plan before querying meeting notes.

User-level vs. workspace-level access

User-level connections place the responsibility for ChatGPT training opt-outs on each individual. A team member on a Free or Plus plan who connects without opting out may have their queries, including meeting notes returned by the connector, used for model training. Workspace-level governance, available on ChatGPT Enterprise, moves training and retention policy to the admin layer. Opt-outs are enforced org-wide by default, removing the dependency on individual action. If your organization uses ChatGPT Enterprise, ask your admin whether MCP is enabled and whether org-wide training opt-outs are in place before connecting.

How to monitor workspace activity

To monitor what meeting data is being queried through the connector, use Granola's audit API, available on the Enterprise plan, which pulls workspace audit events into your security and compliance tools. This gives your security team a Granola-side record of connector activity.

The usage analytics are part of Granola's Enterprise plan security controls, which also include SSO/SAML, domain management, and org-wide notification that Granola is transcribing.

When to pause before connecting

1. Your team uses ChatGPT Free, Plus, or Pro. Consumer plans may use conversation content to improve ChatGPT's models unless each user opts out individually. If a team member connects before opting out, any meeting notes returned by the connector during that session may be used for model training. Opt out in Settings before connecting, then verify the toggle is off before querying any meeting notes.
2. Your organization hasn't confirmed MCP status at the workspace level. On Granola's Enterprise plan, MCP is disabled by default and requires explicit admin enablement. Before connecting, confirm with your admin whether MCP has been enabled for your workspace and whether org-wide training opt-outs are in place on your ChatGPT plan.
3. You handle regulated or sensitive data. Granola is SOC 2 Type 2 compliant as of July 2025 and GDPR compliant across all plans, which covers the Granola side of the connection. On Granola's Enterprise plan, MCP is disabled by default and requires explicit admin enablement. Granola's built-in transparency tools, the automated chat message and video watermark, are available across all plans to inform participants that transcription is active. Your obligations on the ChatGPT side depend on your plan.

What to verify before querying meeting notes through ChatGPT

Before querying meeting notes through the connector, these are the four things worth checking.

1. Verify Granola enterprise data controls: If your org is on Granola's Enterprise plan, confirm with your admin that MCP has been enabled for your workspace. On Business and Basic plans, you can connect directly. Once enabled, admins control which scopes are available to users across the organization.
2. Assess ChatGPT enterprise data usage: Confirm that your ChatGPT plan does not train on your data by default. For Enterprise and Business plans, ChatGPT does not use provided inputs and outputs to train models, per ChatGPT's business data policy.
3. Inform participants that Granola is transcribing: Let participants know before or at the start of the meeting. Granola offers an automated chat message and a video watermark to inform participants. The automated chat message posts a notification in your meeting chat when transcription starts, while the watermark displays a visible indicator on your video.
4. Review workspace activity: Granola's Enterprise plan includes usage analytics that give admins visibility into workspace adoption and activity. Contact your account team for current audit log capabilities and roadmap.

Try Granola for free. Download the Mac or Windows app, connect your calendar, and run your next meeting to see it in action.

FAQs

Does ChatGPT train on my meeting notes?

No, if you are on ChatGPT Enterprise, Business, Edu, or the API. By default, ChatGPT does not use data from these plans for training or improving models, according to ChatGPT's business data policy.

Can I revoke ChatGPT's access to specific notes?

Not per note. You can only disconnect the whole connector. Disconnecting the app in your Granola settings removes ChatGPT's access to your entire Granola workspace at once.

How does this differ from Claude connector security?

Granola's MCP server works with both ChatGPT and Claude. The difference is in each provider's data policies.

What happens if a team member leaves?

When HR deactivates an account in the identity provider, Granola revokes their access automatically. Meeting notes remain accessible to authorized team members in shared folders, scoped by their plan and permissions.

Key terms glossary

MCP (Model Context Protocol): An open-source standard for connecting AI applications to external systems, as defined in the MCP specification.

Connector scope: The specific data fields and metadata that a connector exposes to an AI tool. For Granola's MCP connector, this includes meeting notes, transcripts (by plan), attendees, and calendar event data, but not raw audio.

API data retention: The period of time an AI provider retains API inputs and outputs. ChatGPT retains API data for abuse monitoring, then deletes it unless legal obligations require longer retention, as noted in ChatGPT's API data usage policy.

Admin-controlled access: Enterprise admins can enforce org-wide policies including MCP enablement, data residency, and retention. Individual user access is managed through SCIM provisioning tied to your identity provider.