You open Gmail. The subject line reads “Quick question” and your gut tightens. I saw the same string of identical, bland pitches and realized the inbox was quietly turning into an AI factory.
So I connected my account and watched what happened next.
At 9:12 a.m. my inbox showed a stack of suspicious cold emails — Pangram now labels what it thinks AI wrote
I want you to understand the simplest fact: Pangram will check new Gmail messages as they arrive and tag them. Once you grant access, the tool scans incoming mail and applies a label — AI or Mixed by default — and you can also enable a label for clearly human-written messages.
The workflow is blunt and quiet. New messages are checked, labels appear, and existing labels remain if you pause the integration. You can exclude specific senders, set flagged mail to archive or spam automatically, and watch a dashboard that summarizes labels over 24 hours, seven days, or 30 days.
On my desk, a cold outreach email got tagged — privacy worried me, then Pangram offered answers
Privacy is the conversation everyone expects when email gets scanned. Pangram’s co-founder and CEO Max Spero said the team debated this heavily and landed on a zero data retention policy: processed email data isn’t stored or used for training. The detector also claims it never receives sender, recipient, subject, or other metadata.
That doesn’t erase risk — you’re still routing message text through a third-party filter — but the design puts control in your hands. You choose labeling behavior; Spero personally prefers leaving flagged messages in his inbox for now, though he admits that could change if AI pitches increase.
On my screen, the numbers looked impressive — but detectors still miss and mislabel
Pangram advertises 99.98% accuracy. An independent University of Chicago Becker Friedman Institute study compared detectors and found Pangram the most effective of the tools tested.
Even so, Spero told Gizmodo that detectors will always trail generative models. I’ll be blunt: false positives are real, and your trusted contact could get tagged incorrectly. Use the tool to triage, not to prosecute.
How accurate are AI email detectors?
Short answer: very good but imperfect. Pangram quotes 99.98% accuracy and a University of Chicago study placed it ahead of rivals, yet the company itself warns against expecting perfection. The practical takeaway is to treat labels as signals, not verdicts.
Can I stop AI-generated emails in Gmail?
Yes, to a degree. Pangram lets you auto-archive or send flagged messages to spam, and you can whitelist senders you never want scanned. Think of the tool like a metal detector at the beach: it points you to suspicious items, but you still pick them up and inspect them yourself.
At a conference, someone called it a necessary guard — but the job keeps changing
I’ve told colleagues this feels less like a product and more like a stance: choose visibility over blissful ignorance. Pangram’s integration gives you a dashboard, sender exclusions, and a simple control panel so you can decide whether AI-labeled mail should stay, be archived, or be dumped to spam.
That choice matters because generative models improve fast. Spero said the company can only “accumulate more certainty using more data,” which means the tool will keep evolving. For now, it functions as a smoke alarm for your inbox, not a full security system.
References and context: Pangram’s Gmail integration page is here: https://www.pangram.com/solutions/gmail. Spero’s post on X is here: https://x.com/max_spero_/status/2100242436991705542?s=20. The Becker Friedman Institute study is available here: https://bfi.uchicago.edu/wp-content/uploads/2025/09/BFI_WP_2025-116.pdf.
You can add Pangram to Gmail if you want clearer signals — but will you trust the labels enough to stop reading unsolicited messages that might still be human-crafted?