What is a GEO audit?
A GEO audit measures how often, and how accurately, AI answer engines mention and cite a brand for the questions its buyers ask, and explains why competitors appear instead.
Four questions, answered first
What is GEO?
GEO (Generative Engine Optimization) is the practice of getting a brand mentioned and cited in AI-written answers. A GEO audit measures where you stand before any of that work starts.
GEO vs SEO: what is the difference?
SEO measures positions in a list of links; GEO measures whether a written answer names or cites you. This audit checks both sides: your technical SEO foundations, because engines cannot cite what they cannot crawl, and your presence in the answers themselves.
How do I get cited by ChatGPT?
Start by making sure OAI-SearchBot, the crawler OpenAI uses for ChatGPT search, can reach your pages; then make those pages answer your buyers' questions directly. Step 6 of this audit checks access; steps 3 to 5 show which sources Perplexity and Google AI Overviews cite instead of you. In the paid audit we also ask ChatGPT and Gemini up to 10 of your questions by hand where possible; if we cannot, the report says so.
How do you measure AI visibility in the audit?
In the paid audit we run 30 buyer prompts 3 times on Perplexity and Google AI Overviews, which gives 180 answers, and count mentions, citations, recommendations, and the errors for you and 3 competitors. Where possible we also ask up to 10 of the questions in ChatGPT and Gemini by hand, 3 times each; if we cannot, the report says so. The rules for each label are listed below.
The seven steps
- Agree the scope. Your brand, 3 competitors, chosen with you from the ones you actually lose deals to, your markets and languages, and the topics you want to be known for.
- Build the prompt set. Questions written the way your buyers ask them, grouped by intent. We send you the list before we run it.
- Run it across the engines we measure. Perplexity and Google AI Overviews today, each prompt 3 times, in clean sessions with a fixed locale. In the paid audit we also check ChatGPT and Gemini by hand: up to 10 of the 30 questions, 3 times each, and the report says where they were asked from. If that is not possible, the report says which of them was not measured. Claude is not in our standard measurement yet; we say so rather than guess.
- Classify every answer. Each answer is tagged for mentions, citations, recommendations, and the factual errors it contains, using the definitions below.
- Map the cited sources. We list the domains the engines cite for your topics. This shows where competitors get their visibility.
- Check your site and brand signals. Crawl access for AI bots, rendering without JavaScript, structured data, page structure, and how consistently your brand is described elsewhere.
- Report and prioritise. A plain-language report with the baseline numbers and a ranked list of fixes, each tied to the finding that justifies it.
Which prompts do we run?
We group prompts by what the buyer is trying to do. The examples use placeholders; your set uses your category and your competitors.
| Intent | Example prompt | What it shows |
|---|---|---|
| Category discovery | “What are the best [category] tools for [audience]?” | Whether you are on the shortlist at all |
| Comparison | “[Your brand] vs [competitor]: which is better for [use case]?” | How you are positioned against rivals |
| Problem | “How do I [solve the problem your product solves]?” | Whether you appear before the buyer knows the category |
| Brand | “What is [your brand]? Is it reliable?” | What the engines believe about you, and errors |
| Local or regional | “Best [service] in [city or country]” | Visibility in each market you sell to |
Which platforms?
| Platform | What we record |
|---|---|
| Perplexity | Answer text, brands named, the numbered source list |
| Google AI Overviews | Whether an overview appears, brands named, linked sources |
| ChatGPT (paid audit, by hand) | Up to 10 questions, 3 times each: answer text, brands named, linked sources. If we cannot run it, the report says so |
| Gemini (paid audit, by hand) | Up to 10 questions, 3 times each: answer text, brands named, linked sources. If we cannot run it, the report says so |
Every record keeps the date, the platform, the surface (app, web or search feature) and the locale, so a result can be traced back to how it was collected.
What counts as a mention?
- Mention: the answer text names your brand or an unambiguous variant of it (a product name, a common short form).
- Citation: the answer links to or lists your domain as a source, whether or not the text names you.
- Recommendation: the answer tells the user to consider you, for example by listing you as an option for their case.
- Error: the answer states something false about you: wrong price, discontinued product, wrong location.
Not counted: your brand appearing only because the prompt contained it (we score brand prompts separately), and pages that belong to someone else with a similar name.
What we check on your site
- robots.txt and firewall rules: can the crawlers behind AI search (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, Bingbot) reach your pages?
- Rendering: is the main content in the HTML, or only after JavaScript runs?
- Structure: are there clear headings, short definitions, and lists and tables an engine can quote?
- Structured data: Organization, Product or Service, FAQ and Article markup that matches the visible page.
- Facts: do your prices, your locations, and your founder and product names stay consistent across your site and profiles?
What you receive
- A baseline table: mention rate, citation rate and share of voice per engine, against your competitors.
- The full prompt list and raw answers, so you can check our classification.
- The list of sources AI engines cite for your topics, with the gaps where you are absent.
- Every factual error the engines state about you.
- A technical checklist with pass or fail per item.
- A ranked 90-day plan.
The free audit covers 10 prompts, 3 competitors and one market: 10 prompts × 2 engines (Perplexity, Google AI Overviews) × 3 runs = 60 recorded answers.