ChatGPT reads Gmail through its built-in connector on paid plans, and OpenAI excludes that data from training. What the connector can and cannot do, the setup, and the routes the how-to guides skip: MCP and team inboxes.
Key takeaways
- ChatGPT has a built-in Gmail connector on its paid plans: it searches, reads, and summarizes your mail and drafts replies. Since June 2026 it can also send a reply you approve, one message at a time, from ChatGPT on the web. It does not delete or file anything.
- OpenAI states it does not train its models on data from connected Google apps, with narrow exceptions like content you copy into a chat yourself. That is a stronger default than most people assume.
- The connector links your own Google accounts, not a team's. A team address like support@ needs a different shape: a shared inbox with its own MCP connection that the whole team can see and govern.
Table of contents
- What ChatGPT brings to the connection
- What Gmail brings, and what it does not
- The three routes, honestly compared
- Route 1: the built-in connector, step by step
- What to ask it first
- Gmail connector capability reference
- Route 2: MCP through Developer Mode
- Route 3: your team's shared inbox, governed
- Troubleshooting: why the connection fails
- Related guides: the other assistants and the wider setup
- Frequently asked questions
Connecting Gmail to ChatGPT is now a built-in feature rather than a workaround: on paid plans, the connector links your Google account in a few clicks, and from then on ChatGPT can search your mail, summarize threads, draft replies, and send the ones you approve. What most guides skip is everything around those clicks: what the connector deliberately cannot do, what OpenAI's own policy says about your mail and model training, and what to do instead when the inbox you care about belongs to a team rather than to you. This guide covers all three routes.
What ChatGPT brings to the connection
ChatGPT brings two different doors to your email, and it helps to know which one you are walking through. The first is the built-in Gmail connector: official, quick, and made for your own mail. The second is MCP, the open protocol that tools like Cursor and Claude also speak. The connector is the right door for a personal inbox. MCP is the right door when the inbox belongs to a team and lives inside a tool built to share it, and for a team inbox there is no setup to wire: Drag installs from ChatGPT's Plugin Directory.
What Gmail brings, and what it does not
Gmail brings the mail and one structural fact: Google's own Gmail MCP server is still a developer preview, so most connections to an AI assistant are built by someone else on top of the Gmail API. OpenAI built the connector route. The open-source community built MCP servers for the do-it-yourself route. And shared inbox vendors built theirs for team addresses. Whose inbox it is decides which builder to trust: your personal mail is yours to wire however you like, while a support@ address belongs to the team and deserves a connection the team can actually see.
The three routes, honestly compared
| Built-in connector | MCP via Developer Mode | Team inbox via Drag | |
|---|---|---|---|
| Best for | Your own Gmail | Developers and power users | A shared address like support@ |
| Plans | Paid ChatGPT plans | Business, Enterprise and Edu; read-only on Pro (October 2026) | Plugin Directory plans; Enterprise and Edu admins enable it |
| Can it send as you? | Yes, one approved message at a time, on the web | Depends on the server you connect | From the shared address, after you approve |
| Team visibility | None, it is your account | None | Every AI action visible to the team |
| Setup time | About two minutes | Ten or more, config required | About two minutes, from the Plugin Directory |
Drag · with Varley
The inbox your team and Varley work in together
Shared inbox and live chat for Gmail teams, with Varley, Drag’s AI, included from $18. Plus an MCP server your own AI tools can drive.
Route 1: the built-in connector, step by step
There are no credentials to create and no config to write:
- Open ChatGPT's settings and select Apps or Plugins.
- Choose Gmail, listed under the Google apps, and select connect.
- Google's own sign-in takes over: pick the account, review the permissions, and approve.
- Check that Gmail shows as connected under Apps or Plugins.
- Ask something your inbox knows.
Two things are worth checking before you connect, and both come from OpenAI's own data-controls FAQ rather than from anyone's guesswork. First, training: OpenAI states it does not train its models on data from connected Google apps, with narrow exceptions, such as pasting an email into a chat or mail content that appears in ChatGPT's response. Second, deletion: disconnecting the app stops access, but does not automatically delete existing chats or saved memories. On Business and Enterprise plans there is a third check that is not yours to make: a Google Workspace administrator has to approve the connection, and until they do, the authorization simply fails.
What you get once connected is deliberately bounded. ChatGPT searches, reads, and summarizes, and it drafts replies. Since June 2026 it can also send from ChatGPT on the web: you approve each message, and it goes out from your account one per request, never in batches. It does not delete, label, or file anything. On a personal inbox that boundary is a feature: with an approval on every send, the worst a bad prompt can do is write a draft you decline.
What to ask it first
Beyond these starters, the Drag prompt library collects our full prompt collections for email, triage, and support.
Prompts worth trying in your first ten minutes, safe on any inbox because none of them asks ChatGPT to send anything (and when you want a fuller library, our 50+ ChatGPT prompts for emails has one for every situation):
ChatGPTChatGPTChatGPTChatGPTGmail connector capability reference
A compact reference of exactly what the built-in connector can and cannot do, in one place, because this is the question underneath most searches that land here, human or otherwise.
| Capability | Built-in connector | Notes |
|---|---|---|
| Search mail | Yes | Plain-language queries across the connected account |
| Read and summarize threads | Yes | Message text, full threads |
| Draft replies | Yes | You review before anything goes |
| Send email | Yes, with approval | One message per request, after you approve it; ChatGPT on the web; plan and region vary |
| Delete or archive | No | The connector cannot remove anything |
| Label or file messages | No | Not supported |
| Accounts | Several Google accounts, usable in one conversation | Paid plans only; Business and Enterprise need admin approval |
| Model training on your mail | Excluded | OpenAI's stated data controls, with narrow exceptions covered above |
If you need what the connector does not do, labeling, bulk work, or operating a whole team inbox with attribution, that is what the two routes below exist for, and Drag's own MCP server (47 tools across 12 categories) is the governed version of full read-write access.
Route 2: MCP through Developer Mode
Developer Mode, which OpenAI documents for Business, Enterprise and Edu workspaces on the web, with read-only use on Pro (as of October 2026), lets you add MCP servers, which turns ChatGPT into the same kind of MCP client as Claude or Cursor. For Gmail specifically this is the do-it-yourself route: an open-source Gmail MCP server, your own Google Cloud OAuth credentials, and more capability than the connector, including, on most community servers, the ability to label, archive, and send as you. That power is exactly why it deserves more caution, and why the maintenance warning matters: the most popular open-source Gmail server was archived in March 2026, so check any repo's pulse before trusting it with your mail. Our Gmail MCP server guide covers that landscape in depth, and the same wiring in a code editor is covered in Connect Gmail to Cursor.
Route 3: your team's shared inbox, governed
If the inbox you want ChatGPT working is support@ or sales@, both personal routes are the wrong shape. The AI would act as one person, invisibly, on an address the whole team answers for. The right shape is a shared inbox tool that ships its own MCP server, so the connection belongs to the workspace: that is what Drag does for teams running support or sales on Gmail. Drag is in ChatGPT's Plugin Directory: install it from Plugins, click Connect, and enter an API key from app.dragapp.com under Settings and Integrations (the step-by-step guide walks it in four steps). ChatGPT then gets governed access to the team's queue: reading, assigning, tagging, drafting, and sending the replies you approve, with every action attributed and visible to the team. Drag, our shared inbox for Gmail, costs $12 to $24 a seat, with AI included from $18.
The exchanges look like this. Worked examples, so you can judge the fit before wiring anything:
ChatGPT
Reply, using Drag's MCP toolsChatGPTReply, using Drag's MCP toolsDrag is $12 per user per month, with the AI assistants included from $18, on a 7-day trial with no card.
Troubleshooting: why the connection fails
Four failures cover most of the support threads on this topic.
- The connector is missing from settings. You are on the free plan, or on a workspace plan where the Google Workspace administrator has not approved the connection yet. The first costs an upgrade; the second costs an email to IT.
- Connected, but answers seem to miss recent mail. Check the Gmail connection is still active under Apps or Plugins, reconnect if it errors, then re-ask.
- The authorization loop that never completes. Usually a workspace policy blocking the OAuth scopes. The admin console has to allow the app; retrying the loop harder does not.
- On the MCP route, tools that worked last week silently stop. If your Google Cloud project is in testing mode, Google expires its refresh tokens after seven days. Publish the OAuth consent screen or re-authenticate weekly.
Related guides: the other assistants and the wider setup
Once Gmail is attached, the best ChatGPT connectors for email maps what else the email lane offers, from the inbox to marketing and sending.
ChatGPT is one of several assistants most teams weigh, and the setup differs meaningfully across them: Connect Gmail to Claude covers Claude's built-in connector, Connect Gmail to Gemini covers Google's own assistant in both directions, Connect Gmail to Cursor covers the code-editor route, and connecting multiple Gmail accounts covers the several-accounts problem across all three. If you are weighing MCP against a plain API integration, MCP vs API is the primer. Once Gmail is connected, the best ChatGPT plugins and connectors for work covers what else is worth attaching, and if you came here looking for the old plugin system, what happened to ChatGPT plugins has the full timeline.
Frequently asked questions
Is the Gmail connector available on the free ChatGPT plan?
No. Connected apps require a paid plan, and on Business and Enterprise a Google Workspace administrator also has to approve the connection before it works. If the option is missing from your settings, one of those two is almost always the reason.
Can ChatGPT send emails from my Gmail?
Yes, since June 2026, with limits. ChatGPT drafts the reply, you approve it, and it sends from your account, one message per request, from ChatGPT on the web. Availability varies by plan and region. Nothing leaves your address without your approval, which is the boundary that makes connecting an inbox reasonable in the first place.
Does OpenAI train its models on my connected Gmail?
OpenAI's own policy says no: data from connected Google apps is excluded from model training. The exceptions are narrow: copying an email into a chat, submitting a conversation as feedback, or mail content that appears in ChatGPT's response. For the exact wording, read their data-controls FAQ before connecting.
Can I connect more than one Gmail account to ChatGPT?
Yes: since August 2026 you can connect several Gmail accounts and use them in one conversation. If your life runs across several accounts, our multiple-accounts guide covers the honest routes for Claude, ChatGPT, and Cursor side by side.
Should I connect a shared address like support@ to ChatGPT?
Not through your personal login. The AI would act as you, invisibly to your team, on an inbox everyone shares. A shared inbox tool with its own MCP server gives the whole team one governed connection instead; our guide to connecting a shared inbox to ChatGPT walks through it, and Drag installs from ChatGPT's Plugin Directory. That is how Drag works: $12 to $24 a seat, AI included from $18.
Co-founder
Building Drag for nearly ten years: shared inboxes, boards, and now the AI and agent layer, all on Gmail, plus HeyHelp for the personal inbox. Writes the honest versions of the comparisons.
