ai-search-auditlisted

Audits and then fixes how a brand shows up in AI search and answer engines — ChatGPT, Google AI Overviews, Perplexity, Claude, Gemini, and regional assistants like DeepSeek, Kimi, Doubao or Qwen — across four layers (Discoverability, Clarity, Authority, Trust). Use whenever someone asks for an "AI search audit," "GEO audit," or "AI visibility" check, wants to know what ChatGPT or Perplexity says about their brand, worries AI states wrong facts about their company, asks if their site is "AI-crawler friendly" or "LLM-ready," wants schema/JSON-LD checked or corrected for AI citation, is setting a robots.txt policy for GPTBot/PerplexityBot/ClaudeBot, or wants to get cited and recommended by AI assistants rather than just ranked in search. Trigger without the word "audit" too — "why doesn't ChatGPT mention us," "should we block AI crawlers," "help us show up in AI answers," "fix how AI describes us." Always scope with the user first — which layers, platforms, competitors, who controls what — rather than assuming.
xwmtzcp/ai-search-audit · ★ 0 · AI & Automation · score 70

Install: claude install-skill xwmtzcp/ai-search-audit

# AI Search Audit ## What this does Being findable in classic search and being described correctly, cited, and recommended by AI assistants are different problems. This skill audits four of them separately, because each has a different diagnostic question and usually a different owner inside an organization: | Layer | The question it answers | Usually needs help from | |---|---|---| | **Discoverability** | Can AI systems find and read the content at all? | Whoever controls the website | | **Clarity** | Does AI describe the brand correctly? | Whoever owns positioning and messaging | | **Authority** | Does the brand look qualified enough to be included? | Whoever handles PR, partnerships, community | | **Trust** | Will AI actually recommend it, not just mention it? | Whoever handles reputation, reviews, support | The right-hand column is a prompt for discussion, not an assumption. A five-person company has none of those roles and one person doing all four — that's normal, and Phase A asks rather than guesses. **These four layers are the analytical frame, not the running order.** Executing them one after another means running the same prompt set three times. Phase C reorganizes the work into three parallel workstreams. ## How to grade every recommendation Most advice in this field is unproven. Sort each recommendation into one of three buckets and label it in the report, so the reader can tell evidence from speculation: | Grade | What it means | |---|---| | **Worth doing