George TsagkarakisTwo months of cold emails, sorted into piles. That’s what Pangram’s co-founder and CEO Max Spero says he’s been living with before the company made the feature public.
“I’ve been using Pangram’s Gmail inbox labeler for about two months now. It’s been pretty huge to be able to see cold emails labeled as AI and use this info to inform how I want to engage,” Spero wrote Wednesday in a post on X.
That’s the pitch for the integration Pangram launched Wednesday: connect your Gmail account, and the AI-detection platform scans incoming messages and tags them by how much of the text looks machine-written.
What actually lands on your messages
By default, flagged mail gets a label reading “AI” or “Mixed.” You can switch on labels for human-written mail too, if you want the full picture rather than just the suspects.
Scanning happens on arrival, not retroactively. “New messages are checked and labeled as they arrive. Pausing the integration will stop new scans. Existing labels stay in Gmail,” the company wrote on its website.
So the labels are yours once they’re applied. Pause the thing and your inbox keeps its history.
Specific senders can be excluded from scanning, which matters more than it sounds like if you work with anyone whose writing gets mistaken for a model’s. There’s also a dashboard tracking how Pangram labeled your mail over the last 24 hours, seven days or 30 days.
The privacy question came up internally
Spero said there was significant internal discussion about privacy, given how sensitive email data is. The company settled on a zero data retention policy: no email data kept after processing, none used for training.
Pangram’s website goes further and says the detector doesn’t receive information about an email’s sender, recipient, subject line or other metadata. The classifier sees text and nothing else.
Autopilot is the risky setting
You can tell Pangram to automatically archive anything labeled AI, or route it straight to spam. Spero’s own preference right now is the tame one: label the email, leave it in the inbox. He said that could change in the future “if AI emails ramp up.”
Take the recommendation seriously before you turn on auto-spam. Detectors have improved, but false positives are still possible, and the cost of one isn’t a mislabeled message. It’s a real email you never see.
Pangram claims its detector identifies AI-generated text with 99.98% accuracy, and an independent study comparing multiple AI-detection tools found Pangram to be the most effective. Even at that rate, a busy inbox generates enough volume to eventually produce a miss.
The company doesn’t pretend the arms race ends
Spero has said detection tools will always be chasing generative models that keep getting better.
“I don’t think we’ll ever get to one hundred percent accuracy,” said Spero. What the company can do instead is “accumulate more certainty using more data.”
Which is the honest framing, and a reason to run this in label-only mode for a few weeks first. Watch what gets tagged, check the dashboard at the seven-day mark, and see whether the flags match what you’d have called slop yourself. If they do, then start talking about archiving.
