- Use a fixed prompt set of fifteen to thirty buyer questions and run each at least three times.
- Score three things: mentioned, cited with a link and described correctly.
- Report ranges and patterns. Single screenshots mislead because answers vary.
An LLM visibility audit answers one question: when buyers ask AI about your category, are you in the answer, and is it right? You can run a basic version yourself in an afternoon.
Step 1: write the prompts
Collect real questions from sales calls, support chats and Search Console. Group them by stage: discovery, comparison and decision. Include place, budget and constraints, as buyers do. Do not use your brand name in the prompt, except in a separate group that checks what engines say about you directly.
Step 2: choose engines and settings
Pick two or three: for example Google AI Mode, ChatGPT with search and Perplexity. Use a clean browser or a signed out session, and note country, language and date. Keep settings the same each month.
Step 3: run each prompt more than once
Answers change between runs. Three runs per prompt is a workable minimum. With twenty prompts and three engines that is a hundred and eighty runs, which is enough to see a pattern.
Step 4: score every run
| Field | What to record |
|---|---|
| Mentioned | Is your brand named in the answer text? |
| Cited | Is a link to your domain shown? |
| Described correctly | Are the services, location and claims accurate? Mark wrong or missing facts. |
| Competitors | Who is named instead, and which sources are cited? |
| Sentiment | Positive, neutral or negative, with the quote that shows it. |
Use plain definitions for sentiment and let a second person score a sample to check you agree.
Step 5: read the pattern
- Named but not cited: your facts are known but your pages are not the source. Strengthen the page that should own the topic.
- Cited but described wrongly: fix inconsistent facts across your site and listings.
- Never named: check crawl access and outside proof first.
- Competitor always named: study the sources they share.
Step 6: fix one thing and retest
Change one variable, wait a few weeks and rerun. If you change everything at once you will not know what worked.
Want it done for you? See our LLM visibility audit.