Checking for Spam Content with Chrome AI (www.raymondcamden.com)

A developer ported a server-based spam-detection demo to run with Chrome AI’s on-device capabilities, using the Prompt API to classify whether text would be flagged as spam and return reasons. The idea is to provide real-time, in-browser guidance — for example, flagging email or CMS copy that might get routed to spam — without round trips to a server. The demo produced sensible results: aggressive, vague debt-collection copy was flagged as spam with specific reasons (threats, vague details, pressure tactics), while a polite conference follow-up was correctly labeled non-spam. Technically, the proof-of-concept uses window.LanguageModel.create with a system prompt to steer behavior, monitors model download progress via downloadprogress events, and enforces structured output via a JSON schema (top-level boolean spam and an array reasons). Calls look like session.prompt(input, { responseConstraint: spamCheckResultSchema }). That structured-response pattern is important — it makes programmatic handling and UI integration straightforward. Note the Prompt API is currently behind a Chrome flag and available to extensions but not yet in the main browser. The demo shows on-device, privacy-friendly, low-latency spam signaling is feasible in web apps and could be integrated into editors and moderation pipelines once Chrome’s APIs are more widely available.