I Let Pangram Loose on My Gmail Inbox. Here’s How Much AI It Found

Save Story Save this story Save Story Save this story AI-generated content is everywhere now: From restaurant menus and student essays to full-length movies and social media posts , AI is churning out huge amounts of text, images, audio, and video. In some digital spaces, it's overwhelming human-made content. But what about your inbox? Just how many of your friends, family members, clients, and colleagues are using AI to compose their emails? Is that store newsletter you just signed up for written by a real person or a bot? It's not always easy to tell. The Pangram AI text and image detector tool has now extended its reach into your Gmail inbox. The company makes a browser extension that sits in Gmail and runs incoming messages through its algorithms to try to detect AI writing. You will need at least the $20-per-month individual plan on Pangram to use the Gmail integration—it's not available on the free tier. I tested it to see how accurate it appears to be based on my recent Gmail conversations. Setting Up Pangram in Gmail After connecting, Pangram gets to work. Courtesy of David Nield You can let Pangram loose in your Gmail inbox by heading to the Gmail integration page and clicking Connect your Gmail . You can also click the Gmail Integration entry on the left-hand navigation pane in your main Pangram dashboard on the web. You'll need to go through the standard authorization screens to confirm you want to make the connection. Pangram says the contents of your emails aren't stored after being assessed and won't be used to train models designed to look for AI text. There's still a level of trust here, but you can quickly remove the Pangram plug-in from your Gmail at any point. After you connect your Gmail account, Pangram gets to work. It'll scan your 100 most recent email threads and then any new messages from then on—you can't get it to check all of the emails you've ever received. As far as I can tell, email scanning doesn't count against your scan credit limit for the month, so you're covered no matter how busy your inbox is. After scanning, the Gmail Integration part of your Pangram dashboard changes. You can see at a glance how many of your emails are suspected to be AI, how many are most probably human, and how many have been flagged as a mix between the two. Inside Gmail, these three categories can be applied as labels, though this is optional. (If you decide to remove Pangram from your inbox later, any labels it's created will stick around, but you can delete these manually.) You can also choose to send incoming AI-written emails straight to the archive or to spam if you want, which is handy for keeping AI content away from your eyeballs. You can also choose to exclude specific domains, Gmail labels, subject keywords, and senders from Pangram's scans, if you don't want emails from your boss or mom scrutinized. Scanning My Inbox With Pangram The Pangram verdict on my inbox. Courtesy of David Nield So the moment of truth: Out of the 108 threads that Pangram scanned, eight of them were labeled as AI, 37 got the human label, and the remainder were marked as mixed. It's a little dispiriting to discover that way more than half of my incoming messages are probably AI-written, with press releases the most common culprits. In terms of accuracy, none of the emails I know to have been written by flesh-and-blood people were flagged as fully or partly AI. Everything from friends and family and colleagues I know passed the human test, so for this small sample size at least, Pangram's detection algorithms seem to be working. I sent a couple of emails to myself from different accounts to see what Pangram would make of them. One was written with Microsoft Copilot, which the detector correctly flagged as AI. The other was a section of this article, written with my own brain and keyboard taps, and again Pangram got it right—it was marked with the Human label. There does seem to be a minimum email length, and if an email’s word count falls below that threshold, Pangram won't make a decision either way, which is understandable. (AI is harder to detect with a smaller sample size). Only the first email in each thread is scanned (and optionally labeled), so if your correspondent switches to AI halfway through a conversation, it wouldn't get spotted. I like the way the scanner stays in the background and just gets on with its job, and I can see myself using it long-term. It really does affect your mindset if you know a message has been entirely put together using AI. Of course, Gmail being Gmail, you can quickly fire off an AI response as well, should you want to.