Pangram Gmail integration — Pangram’s Gmail integration wants to tell you which messages in your inbox were written by a machine. The AI detector company has released a browser extension that works inside Gmail, running incoming mail through its algorithms to flag suspected AI text. The feature requires at least the $20-per-month individual plan and is not part of the free tier.
I connected the tool to my own account to see what it would find. The results were less clear-cut than I expected: across 108 scanned threads, eight were labeled AI, 37 were labeled human, and the rest fell into a mixed category. That means more than half of my incoming messages carried at least some suspicion of machine authorship, with press releases the most frequent offenders.
How the scanner gets set up

Activating the extension is straightforward. From the Gmail integration page or the left-hand menu of the Pangram web dashboard, you click through to connect your account and complete the usual authorization screens. Once the connection is live, Pangram scans your 100 most recent email threads and then checks new messages as they arrive. It does not go back through your entire history.
Pangram says the contents of scanned emails are not stored after assessment and are not used to train models that look for AI text. Users can remove the plug-in from Gmail at any time. As far as I could tell, email scanning does not consume the monthly scan credits, so a busy inbox should not run into a limit.
After the initial pass, the Gmail Integration section of the dashboard shows counts for suspected AI mail, probably human mail, and messages flagged as a mix. Those three categories can optionally appear as labels inside Gmail. If you later remove Pangram, the labels it created remain until you delete them manually.
What the inbox results showed

The tool also offers controls for keeping flagged mail out of sight. Incoming messages judged to be AI-written can be sent automatically to the archive or to spam. Scans can exclude specific domains, Gmail labels, subject keywords, and senders, which is useful if you would rather not have messages from your boss or your mother scrutinized.
On accuracy, the small sample was encouraging. None of the emails I knew to be written by real people were marked as fully or partly AI. Messages from friends, family, and colleagues all passed the human test. I also sent two test emails to myself from different accounts. One was composed with Microsoft Copilot and was correctly flagged as AI. The other was a section of my own writing, typed by hand, and Pangram labeled it human.
There are limits worth knowing. The scanner appears to have a minimum email length, and messages below that threshold receive no verdict either way, which makes sense given that AI is harder to detect in a small sample. Only the first email in each thread is scanned and optionally labeled, so a correspondent who switches to AI partway through a conversation would not be caught.
Why the labels change how you read
The scanner stays in the background and does its job without much fuss, which is part of its appeal. Knowing that a message was assembled entirely by AI does shift how you read it. And Gmail being Gmail, you can always fire off an AI-generated reply of your own if the mood takes you.
