GEO audit: what AI says about your business

A GEO audit shows what ChatGPT, Perplexity and Google AI say about your business when a buyer asks about your category, and where those answers come from. It is a one-off piece of work with a written result: the questions we asked, the sources the assistants cited, the facts they got wrong and a prioritised plan of what to fix on your site and beyond it. Ongoing work on AI search can follow, but it does not have to.

Why an AI visibility audit comes before ongoing work

Many teams notice the problem by accident. Someone asks an assistant about the company and gets an old address, a service you dropped years ago or a competitor’s name. One answer like that tells you little. An AI visibility audit replaces the anecdote with a fixed picture: the same questions, the same assistants and the same market, recorded on a known date. Only then can you decide whether the gap justifies ongoing effort or whether two corrected pages solve it.

How the GEO audit runs

The question set comes first. We agree the questions with you before anything is measured. They cover what buyers ask before they choose: who provides this service in a given region, which provider fits a specific problem, how two options compare, what is included. We take them from your sales conversations, enquiries and search data, not from a generic template, and document the set so that a repeat check compares like with like.

Then the answers. For every question and assistant we note who is named (you, competitors or nobody), which sources are cited and whether the description of your business is accurate. We separate three cases that often get mixed up: you are missing, you are mentioned with wrong facts, or you are mentioned correctly through a weak source. Each needs a different fix. Every observation carries its date, assistant and market, because AI answers change from day to day.

Then the sources. On your site we look for the pages that should carry the key facts: what you do, for whom, where and with what proof. Typical findings are outdated service pages, conflicting contact details, vague regional wording, facts locked in images or PDFs and missing structured data. Beyond the site we check the directories, profiles, partner pages, review platforms and articles assistants draw on in your category. There the fix is usually a correction request or an update to a profile you already own. We do not create fake listings, reviews or mentions.

Finally the plan and the repeat check. You receive the question set, the observed answers and sources, the list of wrong or missing facts and a plan sorted by priority, with an owner for each task: your team, us or another contractor. After the agreed fixes we run the same set again and describe what changed in mentions, sources and facts.

What drives the cost of an AI search audit

The price is set by the scope, not by a package name. The main factors are:

• how many markets and languages are checked, since answers differ between them;
• the size of the question set;
• which assistants are included;
• how many pages carry your core facts and how many outside sources need reviewing;
• whether a repeat measurement is part of the job.

We put the scope in writing after a short review. We do not promise a citation, a mention in a given assistant or a date by which answers change. Assistants choose their own sources; the repeat check shows what actually happened.

Who it suits and who it does not

The audit suits a company whose buyers compare providers and ask assistants before they make contact, especially in B2B, where a wrong fact can drop you from a shortlist without anyone telling you. It also suits teams that want evidence before paying for ongoing AI search work, and that can confirm public facts quickly: services, regions, terms, proof.

It is the wrong first step if the offer itself is still unclear, because assistants cannot describe what you have not decided. If Google does not index the site properly, a technical SEO audit comes first. And if what you need is a guaranteed mention, no honest audit can give you that.

Common mistakes when reading AI answers

• Treating one screenshot as a result. A single favourable or unfavourable answer says little; only a fixed set repeated over time shows a trend.
• Counting mentions only. Being named with a wrong price or region can cost more than not being named at all.
• Fixing the site and ignoring outside sources. Assistants often lean on directories and third-party pages that still carry old data.
• Reading an index as revenue. In our case for a German B2B platform, an AI visibility index across Google AI Overviews, ChatGPT and Perplexity rose from 100 to 420 over five months. That is a visibility measure on a fixed question panel, not sales and not a promise of citations.

For a first look on your own, the AI search visibility checklist is a good start.

Next step

Write to us through the contact page. Send your category, priority market, a short description of the offer, five to ten questions buyers already ask and, if you have one, an AI answer that got you wrong. We will reply in writing with the proposed scope: markets, assistants, question set and whether a repeat check makes sense.