The LLM Citation Audit
You can't optimize what you don't measure. Before running an AEO program, run an LLM citation audit — a fixed methodology that shows exactly what ChatGPT, Perplexity, Claude, and Google AI Overviews say about your brand and category today. This is the same audit we run for every MiPharos engagement.
Step 1 — Build the prompt set
Assemble 15-25 prompts that a real buyer in your category would ask an AI assistant. Include: 'best [category] for [use case]' comparisons, 'what is [problem]' primers, 'how do I choose a [category] agency' evaluators, and direct brand queries ('is [your brand] good').
The prompt set should be fixed — you'll re-run it monthly to track share-of-voice movement. Do not change prompts once the baseline is set.
Step 2 — Run the prompts
Run each prompt in ChatGPT (web search enabled), Perplexity (default and Pro), Claude (with web search), and Gemini. Log the full response and every URL cited.
For each cited source, record: domain, page URL, position in the citation list, and whether it's you, a competitor, or a third party.
Step 3 — Compute the scorecard
Share-of-voice per engine: how many of the 15-25 prompts cite your brand vs. total citations across all brands.
Citation position: are you cited first, third, or buried?
Sentiment: are you described as a category leader, one of many, or negatively?
Competitor breakout: which competitors show up most often, and on which prompts?
Step 4 — Diagnose the gaps
For every prompt where you don't get cited, ask: does the citation come from a page you don't have? Structured differently than yours? Referenced from a source you're not on?
Common gap patterns: no comparison content, no primer/definition content, missing from key third-party directories, no `llms.txt`, thin schema markup.
FAQ
- 01
How often should I re-run the audit?
- Monthly is standard. LLM retrieval indices update frequently, and month-over-month movement is the primary metric we report to clients.
- 02
Can I run the audit myself?
- Yes — the methodology is transparent, and running it once yourself is a great forcing function. Where agencies add value is in the diagnosis and remediation (which schema gaps, which content to publish, which third-party mentions to earn) rather than the audit itself.