A new Gmail tool flags which of your emails were written by AI

A tool called Pangram can now scan your Gmail inbox and tell you which messages were probably written by AI. It works as a browser extension and labels each email as AI, human or a mix of the two. It is not free. You need at least the $20-per-month individual plan.

How it works

After you connect your Gmail account and approve the standard permission screens, Pangram scans your 100 most recent email threads. After that, it checks new messages as they arrive. It cannot go back through everything you have ever received.

The labels show up inside Gmail, though that part is optional. You can also send emails flagged as AI straight to the archive or to spam. Specific domains, Gmail labels, subject keywords and senders can be excluded, so mail from your boss or your mom is never scanned.

What the test found

A Wired writer tried it on 108 threads. Eight were labeled AI and 37 were labeled human. The rest were marked mixed. Press releases were the most common source of AI writing.

On accuracy, every email from friends, family and colleagues known to be written by people passed as human. A test email written with Microsoft Copilot was flagged as AI. A passage written by the reviewer by hand was labeled human. That is a small sample, so it is a good sign rather than proof.

Limits to know

• Only the first email in each thread is scanned, so a switch to AI mid-conversation would be missed.
• Short emails fall below a minimum length, and Pangram makes no call either way.
• Pangram says it does not store email contents after checking them or use them to train models.
• You can remove the plug-in at any time, but labels it created stay until you delete them.

Who it is for

The tool suits anyone who wants to know whether a client, vendor or newsletter came from a real person. It does require giving a third party access to your inbox, which takes some trust, even with Pangram's privacy promises. The reviewer said the scanner works quietly in the background and plans to keep using it. Knowing a message was fully AI-written, the reviewer noted, changes how you read it.