An audit of how AI answers see your business, and a prioritised list of what to fix, with every finding sourced and dated.
An audit of how AI answers see your business, and a prioritised list of what to fix, with every finding sourced and dated.
Generative engine optimisation, or GEO, is the work of making a business easy for AI answer engines to read, understand and describe correctly. Much of it is unglamorous: crawler rules, markup, and business details that agree with each other. The audit tells you which of it you actually need.
It looks at three things: what your site lets AI crawlers read, whether your details match everywhere they appear, and who gets named when a buyer asks the questions you would want to be the answer to.
When your buyers might ask an assistant before they ask a search engine, and you want to know what it can say about you. When a competitor gets named for questions you expected to own, and nobody can tell you why.
When it is not: if the site is about to be rebuilt, or you cannot yet say which market and services you want to be known for. The audit measures a site against a market. Without both, the honest answer is to wait.
An SEO audit asks whether search engines can crawl, index and rank your pages. This one asks whether AI crawlers can read them, whether your details agree everywhere, and who gets named when a buyer asks a specific question. The technical ground overlaps. The questions do not.
The reporting differs too. There is no score out of a hundred. Every finding carries its source and its date, and anything nobody measured is labelled as such.
It starts with up to 30 questions a real buyer would ask, in one language and one market, grouped by intent: recommendation, comparison, how to choose, and your brand by name. You approve the list before anything runs. Brand questions are there because an answer can describe you wrongly as easily as it can leave you out.
For every prompt we record who the web search results name, and how you compare with up to three competitors you choose or we propose. That is labelled a proxy throughout. It shows who the open web points to for the question, not what any AI engine cited, and the report never presents it as more.
The live site, read the way a crawler reads it. The rules robots.txt sets for GPTBot, ClaudeBot, PerplexityBot, Google-Extended and other AI crawlers. Whether llms.txt exists, what the sitemap lists, whether the structured data matches what the page actually says, and whether key pages show their text without JavaScript.
That last check is the quiet one. A crawler that does not run JavaScript sees only what the server sends, so a page that builds its text in the browser can look empty to it. We also check that your name, address, phone and profiles agree across the site and public listings.
Our tools cannot query ChatGPT, Gemini, Perplexity or Google AI Overviews directly. So by default the audit comes with a prompt sheet: the same buyer questions, set out for you to run on your own accounts, with space to record each answer and its date. Or we run it for you: we put those same prompts to ChatGPT, Gemini, Perplexity and Google AI Overviews ourselves, one dated run, logged as measured manually, for €299 net + VAT.
Whatever you send back goes into the report as answered, with its date. Any engine nobody ran is marked “not measured” rather than estimated. Answers also vary from one run to the next, so every result is a snapshot, not a verdict. A gap with a label on it is more use than a number nobody can check.
The audit costs €500 net + VAT. Within two weeks of a completed intake, a branded PDF. A readiness scorecard: each check pass, fail or not present, with its URL and date. A lost-prompt table: the question, who gets named instead, the likely reason, the fix. A fix list ranked High, Medium and Low, each marked on-site or off-site. Then a walkthrough call.
Reviews, listings and press are off-site: yours to act on, with our advice. If we built your site or look after it on a care plan, we can make the on-site fixes, staging first; otherwise the list is written for your developer. About 14 days after the fixes go live, we rerun the same prompts and checks and send a short before-and-after note.