Most of us don’t have an email problem. We have a finding things in email problem.
The thread where a client agreed to the revised scope. The flight confirmation from four months ago. The three separate messages that, taken together, explain why a project slipped. It’s all in there. Gmail’s search will find it eventually, if you can remember the right keyword and the right sender and roughly the right month.
Connecting Gmail to Claude changes the shape of that problem. Instead of searching for keywords, you describe what you’re looking for in plain language and let Claude go read the thread for you. Ask it what a conversation concluded and it summarises. Ask it to reply and it drafts.
Here’s how to set it up, and — more usefully — what’s genuinely worth doing with it once you have.
Before you start
You’ll need two things:
- The Claude desktop app, or claude.ai in a browser. Both work; the desktop app is nicer for this because it stays open alongside your other work.
- A Google account you’re already signed into.
One caveat if you’re on a work account: if your organisation uses Google Workspace, your admin may need to approve Claude as a trusted app before the connection will go through. Same goes for Claude Team and Enterprise plans, where an account owner enables connectors centrally. If you hit a wall at the permissions screen, that’s almost always why — and it’s a conversation with your IT team rather than something you can fix yourself.
Connecting Gmail: the actual steps
The whole process takes about a minute.
1. Open Settings → Connectors.
In the Claude desktop app, click your name or profile icon in the bottom corner, then Settings. Find Connectors in the sidebar.
2. Find Gmail and click Connect.
You’ll see a directory of available connectors. Gmail sits under the Google group, usually near the top alongside Google Calendar and Google Drive.
3. Sign in and grant access.
A Google sign-in window opens. Pick the account you want to connect, then review what Claude is asking for. Read this screen rather than clicking through it — it’s the moment you’re deciding what Claude can see.
4. Check it’s enabled in your chat.
This is the step people miss. A connector can be connected to your account but switched off for the particular conversation you’re in. Look for the connector or tools control near the message box and make sure Gmail is toggled on. If Claude tells you it can’t see your email despite a successful setup, this is nearly always the reason.
That’s it. Ask Claude something about your inbox and it’ll start reading.
What you can actually do with it
Setup is the boring part. Here’s where it earns its place.
Find the thing you can’t find
The single biggest win, and the one you’ll use daily. You don’t need the sender or the subject line — just describe it.
“Find the email where someone sent me the updated pricing for the hosting renewal.”
“What did Sarah say about the deadline for the migration project?”
“Did anyone ever reply to my question about the licence count?”
That last one is quietly useful. Searching for the absence of something is close to impossible with keyword search, but perfectly natural to ask in a sentence.
Summarise a thread you were added to late
Being CC’d into a forty-message thread at message thirty-eight is a particular kind of misery. Rather than scrolling to the bottom and reading upward, ask:
“Summarise this thread and tell me what’s actually being asked of me.”
Claude reads the whole chain and gives you the shape of it: what was decided, what’s still open, who’s waiting on whom.
Pull information across many emails at once
This is where it stops being a nicer search box and starts being genuinely different. Claude can read across multiple messages and assemble something new from them.
“Go through my email from the last month and list every invoice I’ve received, with the amount and due date.”
“What are all the meetings I’ve been invited to next week that I haven’t responded to yet?”
“Pull out every action item assigned to me across the project threads from this week.”
You’re asking for a piece of information that doesn’t exist in any single email. It exists scattered across twenty of them, and Claude does the assembling.
Draft replies in your own words
Ask Claude to write a response and it’ll produce a draft you can edit before anything happens:
“Draft a polite reply saying I can’t make Thursday but Friday morning works.”
“Reply to this and ask for the three things we still need from them, in a bulleted list.”
The drafts land in Gmail as actual drafts, so you can open them in Gmail, adjust the wording, and send when you’re happy. If the tone isn’t right, say so — “make it warmer,” “much shorter,” “less apologetic” — and it’ll rework it.
Send, reply, and forward — with a checkpoint
Claude can send mail on your behalf, but not silently. By default it asks for your approval before anything leaves your account, and it shows you what it’s about to send. That confirmation step is the whole safety model, and it’s worth not getting into the habit of approving on autopilot.
Combine it with your calendar
Gmail on its own is useful. Gmail plus Google Calendar connected together is noticeably more so, because most email is secretly about time:
“I’ve got a message here proposing three meeting times. Which of them am I actually free for?”
“Summarise everything I need to read before tomorrow’s 10am meeting.”
What it can’t do
Worth being straight about the limits, because they shape how you use it.
It doesn’t act on its own. There are no triggers or automations. Claude won’t notice new mail arriving and respond to it. Nothing happens unless you ask, in a conversation, right then. If you were hoping to build an auto-responder, this isn’t that.
It can’t see images inside emails. Text, yes. A screenshot someone pasted into the body of a message, no. If the information you need is baked into an image, Claude will miss it.
It only sees what you see. Claude inherits your existing Gmail permissions exactly. It can’t reach a shared mailbox you don’t have access to, and it can’t see another person’s mail.
It’s only as good as your search terms. Vague questions get vague results. “Find the thing about the project” will struggle; “find the email from the accounts team about the Q3 invoice” won’t.
Privacy: what’s actually happening
Reasonable question, and Anthropic is fairly specific about it.
Claude only reaches into your Gmail when you ask something that requires it — it isn’t continuously reading your inbox in the background. It can only access the account you explicitly connected. Data it retrieves gets stored with the chat on Anthropic’s servers, which means deleting the conversation deletes the retrieved data along with it. And Anthropic states it doesn’t train its models on your Gmail or Calendar integration data.
Two habits worth adopting anyway. First, if you have both a work and a personal Google account, be deliberate about which one you connect and why. Second, disconnect it when you’re not using it — Settings → Connectors, and disconnect. It takes a minute to reconnect later.
When it doesn’t work
Three things account for nearly every failure:
“Claude can’t see my email” right after connecting. The connector is off for that specific chat. Toggle it on near the message box.
The permissions screen refuses you. Google Workspace admin restrictions. Your IT team needs to approve Claude as a trusted application.
It finds nothing when you’re sure the email exists. Give it more to work with — a rough timeframe, the likely sender, what the message was about. “Sometime in spring, probably from the finance team, about the renewal” beats a single keyword.
Worth it?
The setup is a minute. The habit takes a little longer, because the instinct to open Gmail and start scrolling is deeply worn in.
But the shift is real. You stop remembering keywords and start asking questions. You stop reading forty messages to find the one line that matters. And the small daily tax of where was that email — the one you’ve paid so many times you no longer notice it — quietly goes away.
Start with finding things. Once that clicks, the rest follows.
