GEO audit: a step-by-step method and prompt

A GEO audit checks how AI assistants such as ChatGPT, Perplexity, Gemini and Copilot answer the questions your buyers ask, whether they mention your brand, which sources they cite, and what they get wrong. It ends with a list of content and technical fixes that make you easier to find and cite.

What it is

AI assistants answer from a mix of training data and live retrieval. When they retrieve, they cite pages, and those pages shape the answer. A GEO audit treats each assistant like a new search engine: you define the queries, record the answers, and look for patterns in who gets mentioned and why.

Read our guide to getting cited by AI assistants for the strategy behind the audit.

When to use it

• Prospects say "ChatGPT recommended a competitor".
• You want a baseline before investing in GEO content.
• You changed positioning or pricing and want to see if AI answers caught up.
• Quarterly, alongside your SEO audit.

Step-by-step method

1. Write 20 to 40 buyer questions. Mix category questions ("best invoicing tool for freelancers"), problem questions ("how to chase a late invoice"), comparison questions ("Tool A vs Tool B") and brand questions ("what is [your brand]").
2. Run each question in every assistant. Use fresh sessions, run each question at least three times, and record the full answer plus every cited URL.
3. Score presence. For each question, note whether you are mentioned, your position in any list, and whether the description is accurate.
4. Map the cited sources. Count which domains get cited most. These are often review sites, forums, publications or competitor blogs. They are your real competition in AI answers.
5. Find accuracy problems. List wrong facts about your pricing, features or audience. Trace them to the source page if one is cited.
6. Check technical access. Confirm robots.txt does not block AI crawlers you want, key pages render content in HTML, and you have an llms.txt file.
7. Plan fixes. Write answer-first pages for questions where you are absent, correct outdated facts on your own pages, and pursue mentions on the most-cited third-party domains.

Ready-to-use prompt

You are a GEO analyst. I ran buyer questions through AI assistants. Analyze the results for [BRAND] ([DOMAIN]), which sells [PRODUCT] to [AUDIENCE]. Our main competitors are [LIST].

Data (question, assistant, run number, full answer, cited URLs): [PASTE]

Tasks:
1. Visibility table: for each question, how many runs mentioned us per assistant, and our typical position.
2. Top 10 cited domains across all answers, with counts and the type of site (review, forum, publication, competitor, our site).
3. Accuracy issues: every claim about us that looks wrong or outdated, with the cited source if any.
4. Gaps: questions where competitors appear and we do not, and what content or third-party mention would most likely change that.
5. A ranked action list of 10 items with effort (S/M/L).

Only use the data provided. Mark any inference as an inference.

Checklist

• 20 to 40 buyer questions across four types
• Every question run 3 or more times per assistant
• Mentions, positions and accuracy recorded
• Cited domains counted and categorized
• Wrong facts traced to their source
• AI crawler access and llms.txt checked
• Answer-first pages planned for missing questions
• Third-party mention targets listed
• Baseline saved for the next audit

How Thig automates it

The GEO agent runs your buyer questions across ChatGPT, Perplexity, Gemini and Copilot on a schedule, shows visibility per assistant, and posts fixes to your feed. The Writer agent then drafts the answer-first pages, following the answer engine optimization method.