AI-based SMS opt-out detection is now available

Telnyx SMS now uses an LLM to detect unsubscribe intent in inbound messages, going beyond fixed keyword matching to catch natural-language opt-out requests like "don't message me again," "leave me alone," or "please stop texting," and handles them exactly like a standard STOP.

What's new

• LLM intent detection: An LLM evaluates each inbound message for opt-out intent, catching phrasing that fixed keyword filters miss. Your keyword handling still runs first, instant and free, and only messages that match no keyword are evaluated.
• Natural-language support: Recognizes unsubscribe requests expressed in unexpected language, spelling, or phrasing that traditional STOP keyword filters cannot match.
• Automatic opt-out action: When the LLM classifies a message as an opt-out with high confidence, the sender is blocked and your opt-out auto-reply is sent, with the same handling as a literal STOP. Your custom reply is used if it exists, otherwise the default. Everything else delivers normally.
• Per-inference billing: $0.002 for each inbound message the LLM checks, including ones it decides to deliver. Keyword matches, media-only messages, profiles with the feature off, and messages during an LLM outage are never charged.
• Safety rails: A high bar before the LLM acts, and if it's ever slow or unavailable, messages deliver normally rather than being held.

Why it matters

Traditional SMS unsubscribe relies on fixed keywords like STOP, END, or UNSUBSCRIBE. Real recipients don't always use those exact words. They say "don't text me," "take me off this list," or use language that keyword filters miss entirely. For high-volume SMS programs, missed opt-out requests mean continued messaging to recipients who asked to stop, which creates compliance risk and damages sender reputation. LLM-based detection catches the intent behind the words, not just the words themselves, so opt-out requests get honored regardless of phrasing.

Example use cases

• High-volume SMS programs where recipients express opt-out requests in varied and unpredictable language.
• Political campaign messaging where recipients may use informal or aggressive opt-out language.
• Marketing programs that need to maintain compliance by capturing every opt-out intent, not just keyword matches.

Getting started

• Pick the messaging profile where you want it. Each profile has its own toggle, off by default.
• Enable the feature with the Messaging Profile API: set features.ai_opt_out_detection_enabled to true.
• Configure your stop auto-reply if you want a custom confirmation. LLM-detected opt-outs use the same auto-reply and block behavior as keyword STOPs.

Learn more in the advanced opt-in/out guide or the messaging profile API reference.