How do I detect when an AI engine drops my brand?

Engines drop brands silently — there's no GSC-equivalent alert when ChatGPT or Perplexity stops citing you. Detection requires continuous prompt monitoring with thresholded alerting. The standard alert pattern: send a notification when citation share on a tracked prompt set falls 10+ percentage points week-over-week, or when share-of-voice vs. a named competitor inverts.

A roll-your-own detection stack: (1) Maintain a prompt registry of 100–500 prompts tagged by intent and priority. (2) Run them daily against ChatGPT, Claude, Perplexity, and Google AI Overviews via API or headless browser. (3) Parse responses for brand mentions — exact-match on brand name plus normalized variations (with/without trademark symbols, common misspellings). (4) Store time-series share data in a database with engine, prompt, brand-detected, brand-position-in-answer, and competitor-presence fields. (5) Alert on threshold breaches — 10pp share drop on a single engine, 5pp drop across 3+ engines, or new competitor appearance above 20% share.

Common causes when a drop is detected, in order of frequency: (1) competitor pushed new content that's now winning the retrieval slot, (2) your content went stale and freshness signals dropped, (3) model update changed retrieval behavior (often resolved by the next update cycle), (4) infrastructure issue — your site became slow, returned errors to crawlers, or accidentally blocked an AI user agent in robots.txt, or (5) third-party signal decay — a Wikipedia citation removed, a Reddit thread deleted, or a high-DR media mention dropped from view. Dedicated platforms like Surfaced run this whole loop and surface the likely cause alongside the alert.