Monitoring your AI visibility by running five or ten prompts a month has become fashionable. It's better than nothing. But it's statistically useless — and worse, it creates a feeling of control where none exists. Here's what you actually learn from a real GEO audit, the kind with hundreds of prompts, and why the findings tend to make clients uncomfortable.
There's a big difference between "checking whether you appear in ChatGPT" and "understanding what's happening to your brand in the AI ecosystem." The first takes five minutes. The second requires serious work. The results look nothing alike.
Why five prompts tell you nothing
Language models are probabilistic. The same question, asked twice, can produce different answers. Not radically different — the model has stable preferences — but different enough that a small sample lies to you.
If you run five prompts and appear in three, you might think you're in good shape. But if you run two hundred and appear in twenty-four, the real picture is 12% visibility. Extrapolating from five prompts is an optical illusion. Acting on an illusion is expensive.
The same applies to trends. Knowing whether you're "going up" or "going down" requires stable series. With five prompts you don't have a series — you have noise.
What kinds of prompts actually matter
A well-designed GEO audit combines several prompt types, because each measures something different. Without revealing the exact formula — which changes by sector and is part of the work — the categories are worth describing.
Commercial intent prompts are the ones that matter most for revenue: "recommend a restaurant in Mahón for an anniversary dinner", "architect for full renovation in Menorca", "tax advice for freelancers in Ibiza". The user wants to hire. Appearing or not appearing translates directly into euros.
Comparative prompts reveal the most about positioning against competitors: "compare the main architecture practices in Menorca", "differences between fine-dining restaurants in Mahón". Here you see which side of the table the AI puts you on, and on what criteria.
Local prompts expose presence gaps: "what to do in Sant Lluís", "best cheap places to eat in Ciutadella". Appearing here is easy if your local schema is solid; not appearing signals an obvious problem.
Long-tail prompts separate brands with real depth from those with surface-level authority: "Menorcan architect specializing in 19th-century house restoration", "restaurant with a celiac-friendly menu in Sant Lluís". If you appear here, you have genuine authority. If not, you have the appearance of authority.
What you learn when you do the audit properly
Things the client would never have spotted on their own. Two patterns come up repeatedly.
First: discovery of invisible competitors. There's almost always someone in your sector who doesn't show up in your traditional SEO reports but who AI models cite frequently. Often it's a smaller brand that has done solid editorial and schema work. That's gold — it tells you who's executing with discipline.
Second: gaps in your own prompt coverage. There are questions you consider central to your business where you systematically don't appear. Or you appear but with wrong or outdated descriptions. Knowing this is what lets you fix it.
What an audit can't do
To be honest, it's worth saying what it doesn't reveal. It can't tell you with certainty which specific changes will cause which specific results — AI models are partially opaque, and their criteria shift with every version. It doesn't give you magic formulas. It doesn't guarantee results in three weeks, though it usually accelerates the process significantly.
What it does give you is a map: where you are, where the competition is, what the gaps are, which priorities to move first. The equivalent of a proper medical diagnosis versus "I think something hurts." With a diagnosis you treat; without one, you're guessing.
When it makes sense to do one
If you already have llms.txt, schema and content running but aren't sure what's working — now. If you've never measured your AI visibility beyond asking a friend to search for you in ChatGPT — also now. If your business depends significantly on new customers finding you through search — definitely now.
What doesn't make sense is continuing to make content, schema, and product decisions based on gut feeling when there's a quantitative tool that can support them.
At A-Digital, GEO audits are our entry point with almost every client — in Menorca, across Spain, and in Europe. Not because we want to sell hours, but because without that diagnosis, any recommendation we make is groundless. We'd rather not work that way.
We run serious GEO audits: hundreds of real prompts, segmented by intent and location, with a findings report and action plan. The first layer is free.
Request your free audit →