ChatGPT Work and Email: What It Can Actually Do in 2026
The question people actually want answered about ChatGPT and email is whether it can hit send. As of 2026 it can, from Gmail and from Outlook, and that changed more recently than most write-ups reflect.
What has not changed is that sending is the easy part. The interesting constraints sit in the permission model, in what happens with shared mailboxes, and in the fact that nothing about this setup notices mail arriving. All of the below was checked against OpenAI’s own help centre and release notes on August 7, 2026.
Yes, it sends now
From OpenAI’s release notes, verbatim:
“If you’ve connected Gmail or Outlook, you can now ask ChatGPT to draft and send emails directly from the same conversation. ChatGPT will create a draft and you can choose to send it without leaving ChatGPT.”
Sending is available on the web for Plus, Pro, Business, and Enterprise plans. The draft step is not optional, which is the right design: you see the message before it goes.
Sending is no longer a clean dividing line. ChatGPT sends from either Gmail or Outlook once connected, showing you the draft before it goes. Claude sends through Microsoft 365, off until you enable it, while its Gmail connector remains draft-only.
What ChatGPT Work can do with an inbox
ChatGPT Work is the agentic surface rather than ordinary chat. You hand it an outcome and it works through the steps. Applied to email, the documented capabilities are broader than the send headline suggests.
Reading and searching. It retrieves messages with their metadata (subject, sender, recipients, timestamps, plain-text body) and supports structured search syntax, so from:, subject:, and date filters all work. It returns full message text for anything it finds. Useful prompts from OpenAI’s own library include “List unread emails from yesterday,” “Identify action items from the last three emails about [project],” and “Create a table of sender, subject, and date for emails about [topic] in the last 7 days.”
People and contacts. It searches your work or school directory and saved mailbox contacts, and when actions are enabled it can create, update, and delete Outlook contacts and contact folders.
Drafting and replying. Summarising a thread, drafting a reply covering specific points, and critiquing your own email style are all standard.
Shared and delegated mailboxes. This is the capability most people do not know exists. For a shared mailbox you have access to, ChatGPT can read messages, browse folders, mark mail read or unread, move messages, and send plain-text mail from or on behalf of that mailbox. You must supply the exact mailbox owner email address to reach it.
The permission model is the real control surface
Every action ChatGPT takes in a connected app runs through a permission level, and understanding these four is worth more than any feature list.
| Setting | Behaviour |
|---|---|
| Always ask | Asks before any app action, including reading |
| Any changes | Reads automatically, asks before changing anything |
| Important actions | Reads automatically, asks before important actions. This is the default |
| Never ask | Reads and acts automatically, with no confirmation |
Sending or editing an email is explicitly classified as an important action, which is why the default setting still stops and asks you. OpenAI describes an important action as one that “could have a meaningful effect outside ChatGPT, expose sensitive information, or be difficult to undo.”
The classification is contextual rather than a fixed list. OpenAI notes that “saving a private draft is generally lower risk than sending that content to another person,” so the same underlying capability can pass silently or prompt you depending on what it is about to do.
Never ask carries elevated risk by OpenAI’s own description, and may not be available as an account-wide default at all, though it can sometimes be set for an individual app. Leaving email on the default is the correct call for almost everyone.
For workspaces there is a second layer. Admins configure which actions an app may take at all, choosing between all actions, read-only, or a custom set, and can decide how newly added actions are handled. Workspace agents are governed separately again, through per-agent controls set by whoever built the agent.
The Microsoft setup step that blocks people
Connecting Outlook is not purely a user action. The app requests a specific set of Microsoft Graph scopes (including Mail.Read, Mail.ReadWrite, Mail.Send, Mail.Send.Shared, and the shared variants) and those scopes must be reviewed and approved by a Microsoft Entra administrator. Without that consent you get an error when you try to connect.
Approval is also not the finish line. OpenAI is explicit that approving scopes “does not immediately make all actions available to users,” because admins still review app actions separately in workspace settings. After an admin enables additional actions, users may need to reconnect the app.
If you are hitting a wall connecting Outlook to ChatGPT at work, this two-step gate is almost always the reason, and it is an IT ticket rather than a settings problem.
The limits worth knowing before you rely on it
A handful of documented constraints that will otherwise surprise you mid-task:
- Attachments in shared mailboxes cannot be retrieved. The attachment actions work only with messages in your own signed-in mailbox.
- Write actions are unavailable through Company Knowledge. If you are reaching Outlook that way, it is read-only.
- Broad requests can fail outright. Summarising a long stretch of mail may exceed Microsoft’s service limits and return an error. Narrowing the date range or splitting the request usually fixes it.
- Very recent mail may not appear, since search precision follows Microsoft Graph indexing latency.
- Your admin may restrict which accounts you can connect to an approved set of domains.
On plans: Free and Go get limited ChatGPT Work through the desktop app only. Plus and Pro get it on desktop, web, and mobile. Since sending is a web capability on paid plans, the practical entry point is Plus at $20 a month. Our ChatGPT Work pricing breakdown covers how the metering behaves, and ChatGPT connectors covers the wider app directory.
Nothing here starts on its own
Every capability above begins with you opening a conversation and asking. That is the shape of the product, and for email it is the constraint that matters most, because email is the most reactive thing in your working day.
Both ChatGPT and Claude will now do work on a schedule, unattended. Neither reacts to events. A task that runs hourly still finds out an hour late, and one that runs each morning finds out tomorrow — the trigger is the clock, never the thing you actually care about.
The practical version: ChatGPT will happily draft a reply to the contract that came back signed, once you notice it came back and go ask. It will not update the CRM, generate the invoice, and offer three kickoff times because the contract arrived. The message landing is an event, and events are not in the vocabulary.
Running a scheduled task every fifteen minutes to check the inbox approximates this badly. It burns metered allowance on every empty check and still finds out up to fifteen minutes late.
The layer that reacts to mail arriving
Carly sits at that trigger layer rather than replacing your inbox or your assistant. It has its own email address, so people and systems can send work to it directly, and it fires on inbound events across 260+ integrations instead of waiting to be asked. Pricing is free Zapier-style workflows, with AI agents from $35/month, and the routing and filing steps sit in the free tier.
It also composes with ChatGPT rather than competing for the same slot. Carly runs an MCP server at https://carlyassistant.com/mcp/, which you add in ChatGPT under Plugins, so a ChatGPT Work session can call Carly’s email, calendar, contacts, CRM, and workflow tools inside a run. ChatGPT does the drafting and judgment; Carly notices the message and completes the work across the rest of your stack. To see how that maps onto your own setup, book a call with the team.
For the product overview, what is ChatGPT Work covers the agent itself, and ChatGPT Outlook integration goes deeper on the Microsoft side.
FAQ
Can ChatGPT send emails for me?
Yes. With Gmail or Outlook connected, ChatGPT drafts the message and you choose to send it without leaving the conversation. Sending is available on the web for Plus, Pro, Business, and Enterprise plans, and the draft-and-confirm step is built in rather than optional.
Does ChatGPT work with Outlook as well as Gmail?
Yes, both. The Outlook side additionally supports shared and delegated mailboxes, where ChatGPT can read, browse folders, mark messages read or unread, move them, and send plain-text mail on behalf of that mailbox.
Why can’t I connect Outlook to ChatGPT at work?
Almost always because the Microsoft Graph scopes need approval from a Microsoft Entra administrator, and then the workspace admin has to enable the specific actions separately. Both steps are required, and you may need to reconnect the app after the second one.
Will ChatGPT send an email without asking me?
Not on the default setting. Sending is classified as an important action, and the default permission level stops and asks before those. A “Never ask” option exists but OpenAI describes it as carrying elevated risk, and it is not available as an account-wide default everywhere.
Can ChatGPT read my email and act on it automatically?
It can read and act when you ask it to, in a conversation you start. It does not monitor your inbox, so nothing happens because a message arrived. There are no event triggers, only your requests and scheduled tasks you set up yourself.
Can ChatGPT handle email attachments?
In your own mailbox, yes. In shared or delegated mailboxes it cannot retrieve attachments, since those actions work only with messages in the signed-in user’s mailbox.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
