AI-visibility audit — which sources LLM-search actually cites

Run a category-specific audit of which sources ChatGPT, Gemini, and Perplexity actually cite when answering buyer-intent prompts in your vertical. Build a citation table (top 3 surfaces per LLM × vertical), then reallocate brand-presence budget to the surface that dominates your space. Existing tools (searchable.com, Amplitude AI Visibility) shorten the audit but the table itself is the deliverable.

Cross-LLM the consistent top citation sources are YouTube, Reddit, Wikipedia — order varies by vertical and LLM; Perplexity no longer cites Reddit after a lawsuit, which materially changes spend allocation. A founder discovering their category's LLM-search citations are 70% YouTube reroutes content investment from blog SEO to a YouTube series. This is the cheapest piece of GEO research you can run and the most actionable.

1. Compile 30–50 buyer-intent prompts a real prospect would ask (queries, not keywords).
2. Run each through ChatGPT, Gemini, and Perplexity. Capture the cited URLs / domains for each response.
3. Tally citations by domain class (YouTube channel, subreddit, Wikipedia, blog, news, vendor docs).
4. For each LLM separately, identify top 3 surfaces. Note divergences (Perplexity's lack of Reddit citation; ChatGPT's Wikipedia weighting).
5. Build a one-page citation table per vertical; share with the team.
6. Reallocate next quarter's content budget toward the dominant surface(s). If YouTube wins your vertical, fund a YouTube series before another blog post.
7. Optional: run searchable.com or Amplitude's AI Visibility tool for a faster first cut; treat results as a starting point, not a verdict.

Sample size matters. Auditing 5 prompts gives noisy citations; 30+ stabilises the ranking. Also: re-audit quarterly — Perplexity dropping Reddit shifted the table overnight for several verticals, and policy changes will keep happening.