AI Search Audit — Security skill for Claude Code

ai-search-audit

**A Claude skill for Generative Engine Optimization (GEO) / Answer Engine Optimization (AEO) that refuses to make anything up.** It audits how a brand appears in ChatGPT, Google AI Overviews, Perplexity, Claude and Gemini — and in the Chinese-market assistants **DeepSeek, Doubao (豆包), Qwen (通义千问) and Kimi**, which most tools in this space ignore — then produces the actual changes — across four layers: **Discoverability**, **Clarity**, **Authority**, **Trust**.

Every tool in this space says *audit, optimize and track*. Three things this one does that they don't:

• It cannot invent a platform answer, by design. Claude has no access to ChatGPT or Perplexity. So it designs the prompt set and the recording format, a human runs them, and Claude structures and classifies what actually comes back. It will not generate a response it never received — the single most damaging failure available to a tool like this, because the user cannot tell the difference between a real citation and a plausible one.
• It grades its own recommendations, including down. Every suggestion is marked worth doing, worth trying or not yet, with one test overriding all three: if the entire AI-citation story turned out to be false, would this still be worth doing?
• It sorts fixes by when you could know they worked — same-day, weeks-to-months, or never individually attributable — rather than by tactic. That ordering is what stops a team starting with the slow invisible items, seeing nothing for a quarter, and concluding none of it works.

Two stages, and the second one ships files rather than advice. The audit answers *what's wrong and why*. Optimization, entered on your say-so afterwards, answers *what to change, in what order, and how you'll know it worked* — and hands over the JSON-LD block, the `robots.txt`, the page specs and the verification sheet, not a list of things someone should do.

**Topics**: `generative-engine-optimization` `answer-engine-optimizat