Pangram's AI Detector Flags Eight of 108 Gmail Threads

Pangram, the company behind an AI-text and AI-image detection tool, has extended its scanning into Gmail inboxes, and a hands-on test by WIRED found the tool flagged eight of 108 scanned email threads as AI-generated, according to a report published Nov. 9, 2026, by WIRED senior writer David Nield (wired.com).

How does Pangram's Gmail integration work?

The feature runs as a browser extension that connects to a user's Gmail account through a standard OAuth authorization screen, Nield reported. Once connected, Pangram scans the 100 most recent email threads in the inbox, then continues checking new incoming messages going forward — it cannot retroactively scan the full history of an account, according to the report. Access to the Gmail integration requires at least Pangram's $20-per-month individual plan; the feature is not available on the company's free tier, Nield wrote.

How many AI-written emails did the test find?

Out of 108 email threads that Pangram scanned in Nield's inbox, the tool labeled eight as AI-generated and 37 as most likely human-written, per the WIRED account. The sample reflects one person's inbox at one point in time, not a controlled study, and the report did not include an independent accuracy benchmark for the detector's calls on real-world correspondence — information that would normally accompany a peer-reviewed evaluation of a classification tool.

What does Pangram do with the emails it scans?

Pangram applies one of three labels inside Gmail — AI, human, or mixed — and users can choose whether those labels appear at all, Nield reported. The company told him that email contents are not stored after being assessed and are not used to train the models that look for AI-generated text, though he noted that claim still requires a degree of trust from users, since the extension has access to inbox content to function. The integration can also be removed from Gmail at any time, though any labels it already created remain attached to messages until deleted manually, according to the report.

Can users control which emails get scanned or flagged?

Users can direct Pangram to automatically archive or send suspected AI-written emails to spam, according to Nield's test. The tool also allows exclusions by domain, Gmail label, subject-line keyword, or specific sender — a setting Nield pointed to as useful for keeping messages from a boss or a family member out of the detector's review. Scanning incoming mail does not appear to count against the monthly scan-credit limit tied to Pangram's paid plans, based on his observations, meaning inbox volume should not exhaust a subscriber's quota.

What are the limits of this kind of AI detection?

The test offers a single data point rather than a validated measure of how much AI writing moves through everyday email. Pangram's own dashboard sorts results into suspected-AI, probable-human, and mixed categories, but WIRED's account did not specify a count for the mixed category or detail how Pangram's underlying model distinguishes AI-assisted drafts from fully automated ones. Detection tools of this kind estimate likelihood based on statistical patterns in text; they do not confirm authorship with certainty, and no false-positive or false-negative rate for the Gmail feature was reported. Readers weighing whether to adopt the tool should treat the eight-of-108 figure as descriptive of one inbox, not as a benchmark for inboxes generally.

How can someone try the Gmail integration themselves?

The path described in the WIRED report starts at Pangram's Gmail Integration page, where a "Connect your Gmail" button initiates the authorization flow; the same entry point is also available from the left-hand navigation pane of the main Pangram web dashboard. After granting access, the dashboard updates with a running count of AI-flagged, human-flagged, and mixed threads, and those categories can be converted into Gmail labels at the user's discretion, according to the report.

The full WIRED account, including screenshots of the Pangram dashboard during setup and scanning, is available at wired.com.