AI detector Pangram labelled eight of 108 Gmail threads as AI and 37 as human-written in technology writer David Nield’s personal trial. Many press releases received the AI label, while messages from people he knew did not. The integration examines only the first message in each thread and lets users exclude senders or subjects from scanning.
Writer David Nield tested the text-detection service Pangram through its Gmail integration. Among 108 threads, the system labelled eight as AI, 37 as human-written and the remainder as mixed. Many press releases received the AI label. Messages from Nield’s friends, relatives and colleagues did not. The observations concern the labels in his own inbox, without independently establishing the origin of every message in that sample.[1]
The integration initially scans the latest 100 threads and then assesses incoming mail. It examines the first message in a thread, rather than later replies, and requires a minimum text length before making an assessment. Nield also sent himself an email created with the Copilot assistant and a passage of his own writing. Pangram assigned the former an AI label and the latter a human label.[1]
Users can exclude senders, domains, labels and subject lines from scanning. Applying labels inside Gmail is optional, as is automatically archiving messages or sending them to spam. Pangram says it does not keep email content after assessment or use it for model training. Users can later revoke Gmail access, although labels already attached to messages remain until removed separately.[1]
