I Let Claude Answer My Emails and Here’s How It Went.
If AI is going to take away all our jobs, at least it could make them a little easier while we still have them. And email administration is one of the biggest time wasters in my daily life. I spend hours a week reading and responding to emails, when I really need that time writing about new Pixel or Mac apps .
I was intrigued by the news that the Claude AI chatbot can now reply to and send emails from a connected Gmail account—and without your express consent, if you’re comfortable with automated work.
Claude has been able to integrate with Gmail accounts for some time now, so you could already use it to read and summarize incoming messages—but any email composition had to be done in Claude’s bot and then copied. Now, the AI can compose and send replies directly in Gmail, without having to switch to Google’s email client.
Could this free up a little more time during the day? Is it really wise to hand over email handling responsibilities to an AI bot? I tried it to see how it might impact my productivity (and sanity).
How to connect Claude to your Gmail account
Currently, this requires a paid Claude plan, so you’ll need to use the Pro plan for $20 per month or more. To connect your Gmail account, if you haven’t already, open Claude on the web and click your profile photo (bottom left), then select Settings > Connectors and find Gmail . (If you don’t see your profile photo, click the menu button in the top left corner.)
To do something in Gmail, simply mention it in a prompt. If you’re not ready to compose and send an email, try a prompt like “Who emailed me most today” or “Briefly describe the emails I received this week.” Claude will scan your Gmail inbox and send you the response you requested.
When you’re ready to start sending, you can use fairly informal and ambiguous wording. Phrases like “reply to my last unread email,” “reply to David Neild,” and “reply to the article edit discussion” are all appropriate for Claude. However, the more specific you are with your email address, for example, the better to avoid errors.
If you say “email Dave,” and Claude finds three people with that name, he’ll ask you which one you mean. However, if he finds the wrong Dave and misses the other two for some reason (perhaps due to formatting or address construction), he’ll make the wrong choice and won’t let you know. So be as specific as possible.
Next, you tell Claude what to write: you can dictate the message word for word, ask Claude to be creative and use his imagination, or a combination of both. This works just like any other text request—you can specify as much as you like about the content, style, purpose, and other characteristics of the email.
You’ll be happy to know that automatic sending is disabled by default. If you go back to Settings > Connectors , you can click Gmail to set permissions. Various actions, such as searching your inbox, sending emails, and editing labels, can be configured to one of three options: “Never allow,” “Always allow,” or “Always ask for permission.”
This is what happened when I let Claude control my email.
I took a risk and allowed Claude to send emails automatically without approving every draft, and got to work. I considered contacting Lifehacker’s editors, but wanting to remain on the author list and avoid any awkward misunderstandings, I decided to communicate with myself via email, using a secondary address.
Claude is actually quite good at navigating my Gmail inbox: he understands labels, dates, contacts, subject headings, topics, and nuances. He easily identified “the email I just sent myself from my Proton account” without even having to open Gmail, and I gave him a command to respond—for example, to agree to a meeting to talk about the weather.
One immediate problem: my email ended with the phrase “cheers,” which I would never say except on special occasions involving bars and clinking glasses. I don’t think I’ve ever used it in emails, but apparently many people in Claude’s training set did, so it was included. Keep in mind that Claude may not recognize your standard tone and personal nuances when you let him craft your messages.
The rest of the letter was fine, if a bit formulaic: I agreed to a meeting (with myself) and mentioned the weather (a standard British conversation topic). I then repeated this process several times, giving Claude varying degrees of freedom in composing his responses.
From a technical standpoint, it’s all quite impressive. Claude can accurately read and relay the contents of email correspondence and compose responses according to instructions. I asked him to generate a message summarizing my work for the week (via Google Drive access) and add a question about the weekend football match, and he handled both tasks perfectly. We know AI makes mistakes, but I didn’t find any in my testing.
However, despite the cleverness of this idea, the main problem is that I don’t see the point in using it. Firstly, these letters aren’t written in my style, and I’d constantly be afraid that Claude is about to make a terrible mistake (perhaps misinterpreting an incoming message or making a mistake in an outgoing one).
Another question: where is all this heading? Will AIs send emails to other AIs, ad infinitum? I already suspect that many of the emails I receive are generated by chatbots, which immediately devalues them. If I’m going to communicate with someone, I feel obligated to compose the message myself. While I think I’ll keep Claude’s Gmail connector for searching and summarizing emails, I’m disabling the sending and drafting functions for now.