Connecting ChatGPT to your inbox (Gmail, Outlook): how to do it

“Can ChatGPT read my email?” Yes — and since connectors arrived, without any hacking. But two routes coexist, with very different outcomes: the official connector, quick to switch on, and third-party automation, slower to build but able to trigger work without you. Confusing the two is the main source of disappointment.

The two ways to connect ChatGPT to email

Before configuring anything, pick your route — they answer different needs.

• The official Gmail or Outlook connector. A few clicks from settings. ChatGPT can then search your email during a conversation. It is the default choice, and the one this guide covers.
• A third-party automation (Zapier, Make and friends) pushing each new email to OpenAI's API. Something does then happen without you — but you build and maintain the plumbing, manage API keys, and the output rarely travels further than a Slack message or a spreadsheet row.

The first route is an assistant. The second is the beginning of automation, provided you accept becoming its integrator.

Switching on the Gmail or Outlook connector

1. 1Check your plan includes connectors

   Availability depends on the plan and, at work, on what your workspace admin allows. Settings are the source of truth: if the section is missing, it is not enabled for you.

2. 2Open Settings, then Connectors

   Depending on the version the section is called “Connectors” or “Connected apps”. Available sources are listed there.

3. 3Start the connection and approve consent

   You are sent to Google or Microsoft. Check two things: the account shown — work, not personal — and the permissions requested.

4. 4Point at the source in the conversation

   This is the step everyone skips. Depending on the interface you sometimes have to name the source explicitly, otherwise ChatGPT answers from memory instead of searching the mailbox.

5. 5Test on a real request

   “Find the last exchange with this customer and draft a polite follow-up” tests search and writing at once. A plausible answer with no precise quote from the thread means the source was never queried.

Not working? The three most common causes

• The connectors section is missing. Ineligible plan, or the connector is disabled by your workspace admin. Nothing to fix on the user side.
• The connection succeeds but ChatGPT never opens the mailbox. The source is not selected in the conversation, or the request is too vague for it to bother searching. Name the sender or the period.
• Google or Microsoft consent fails. Company policy on third-party apps: it goes through IT, and no workaround survives for long.

What it looks like day to day

Once the source is properly queried, the profitable uses look alike from team to team:

• Reconstructing a deal's history before a call, without reopening twenty messages
• Spotting requests left unanswered for more than three days
• Drafting a follow-up that reuses the wording actually used in the thread
• Pulling the figures and dates out of an exchange to record elsewhere

One habit worth forming: always ask it to cite the messages the answer rests on. It is the simplest way to tell a grounded summary from a plausible rewrite.

Limits to know before depending on it

• You remain the trigger. Nothing happens until someone opens a conversation. Over the weekend, nobody is processing the mailbox.
• Scope is personal. Everyone connects their own mailbox; shared ones, which often receive the bulk of inbound volume, stay outside the loop.
• No writing elsewhere. The assistant reads and drafts, but updates neither your CRM, nor your quoting tool, nor your sales tracking.
• Instructions do not survive. Your handling rules must be restated every conversation, and nobody else on the team benefits from them.
• Supervision is hard. At team scale, there is no way to say who answered what, based on which information.

A connector answers “help me with this email”. A team's real question is “who handles the two hundred emails arriving today”.

The alternative: connect the mailbox, not the assistant

That is Periscorp's stance. Rather than giving a mailbox to an assistant, you plug the mailbox into the platform: each incoming message triggers an agent that has your company's context, not just the email's text.

1. 1Plug in the mailboxes that matter

   Gmail or Outlook over OAuth, shared addresses included. No scripts, no API keys to manage, no plumbing to maintain.

2. 2Write the instruction once

   “Qualify the request, identify the customer, prepare a reply, create the opportunity if it is a quote.” It then applies to every email, for the whole team.

3. 3Let processing happen on receipt

   The agent reads, finds the customer, gathers the history, drafts and updates records — including when nobody is at a screen.

4. 4Supervise instead of executing

   You choose what goes out alone and what waits for approval. Every run is auditable, showing what the agent read and what it inferred.

Frequently asked questions

Do I need a paid plan to connect my mailbox?

Connectors depend on your subscription, and the tiers move often. The reliable test is not the pricing page but your own settings: if the section is not there, the feature is not open to you.

Is my email used to train the model?

It depends on the account type and privacy settings, which differ between consumer and business plans. Settle that with your IT department before connecting a company mailbox, not after.

Can I connect a shared mailbox like contact@?

Technically you can connect the account, but processing still starts manually, inside one person's conversation. That is not team processing — it is precisely the use case that justifies a platform.

Is it different with Claude?

The principle is close, the screens and permissions differ. Our dedicated Claude guide is linked below.

Connecting ChatGPT to Gmail or Outlook saves time immediately, and it is a good first step. It simply reveals the next limit: as long as processing depends on someone opening a conversation, your inbox stays a queue.