AI VISIBILITY AUDIT FOR SMEs
See the terrain before you move.
Your customers no longer read the tenth Google result — they ask ChatGPT and Gemini who to turn to. I measure what these systems actually answer about your business, in two modes: what the model knows on its own, and what the apps say today with live search. Run by run, verified live, never guessed.
From live measurements · June 2026
search-off = logged model measurement · with live search = how the apps answer todayhow I measure →
What does the AI say about you today?
It depends entirely on how the customer asks — and there are two very different realities, which is exactly why I measure both. Ask a model with no live search and it works from memory: it often invents a local business that doesn’t exist, stays silent, or names a different set every time. Ask through the consumer apps (Gemini, ChatGPT) the way a real customer does in 2026, and they pull live search — they name real businesses, drawn from a finite layer of third-party pages.
The decisive fact isn’t that AI is “blind” to you — with live search it can see plenty. It’s that the apps assemble their answer from reviews, directories and independent mentions, not from your website. Whoever is weak in that external layer is the one left out of the list.
So the honest question is not “am I invisible to AI?” but “who does AI name on my market, in each mode, and from which sources?” In a May 2026 measurement I put the same buyer question to several leading language models — without search, meaning in the logged mode where the model answers only from its trained knowledge, with no live internet search — the kind of question your customers would ask. I asked each for five named businesses so the results would be comparable across models.
In that no-search, logged model measurement, ChatGPT named one long-established, many-thousand-review Budapest clinic first. Gemini led with a completely different name for the same question — one ChatGPT never mentioned. With live search the answer may differ; that is exactly why I measure in both modes.
What matters even more: when I asked the same local-level question about a smaller, rural town, most of the real, well-run local practices were named by no model at all. For that kind of town-level question, a small-town SME’s chance of appearing approaches zero — regardless of how good their website is.
That is exactly what the audit measures: which model names whom on your market, in what order, in which mode, and — crucially — the named pages the answer was built from.
same question, two different names
There are two ways to ask an AI — and each reveals something different about you.
Here the model answers only from what crawlers gathered about you across the web and built into its knowledge. This is the real measure of your GEO/SEO work: has your business made it into the AI’s knowledge, or does it effectively not exist to the machine?
Here the model searches the web on the spot (for example, on Google) and answers from the pages it finds — how most apps work today.
Today live search usually decides. But as AI increasingly handles the purchase itself — think of Google’s new Universal Cart — what the model already knows about you, without searching, matters more and more.
How the audit works
Three steps, no risk. The first one costs nothing and already shows where you stand.
- Free mini-check. I run the questions your customers would ask about your industry and town through ChatGPT and Gemini — in both modes — and report back: do you appear at all, or does a competitor’s name come up instead of yours?
- Full audit with a competitor comparison. I score your site along seven weighted dimensions, then show it head-to-head against a specific competitor: who is visible to the AI today, who isn’t, and the source pages that decide it.
- Action plan, vendor-independent. Concrete, prioritised steps — in a form any developer or agency can price and deliver. You are never locked to a single supplier.
Every dashboard number is dated, comes from a logged and reproducible query, and every named competitor is verified with a live map search. The full four-step process — including honest timelines for each workstream — is on the how it works page; prices are on the pricing page. The sample report and its guided tour are currently in Hungarian — take a look — an English sample is on the way.
step · 1
Free mini-check
I check what ChatGPT and Gemini answer for your industry and town — and report back who gets named instead of you.
step · 2
Full audit + Competitor Map
Seven weighted dimensions, then head-to-head against a specific competitor — in both modes
. Every named business verified with a live map search.
step · 3
Action plan
Vendor-independent, priceable steps — any developer can turn them into a quote. You never get locked to a single supplier.
The methodology is public
The seven-dimension weighted rubric I measure with is not a secret. Technical signals matter, but the heaviest weight sits on presence in independent sources — because that is what truly decides who the AI names. I hold this site to its own scorecard, too: a newborn site honestly cannot top the chart on external presence, and I don’t dress that up. Showing my own weak spots is what makes the measurement of others credible.
the full rubric is public — alongside the technical signals, external presence decides the most
- 20% D1 · Reachability & crawlability — can the AI crawler reach the site at all, and is the content visible without JavaScript.
- 15% D2 · Structured data — does the site identify itself machine-readably: organisation, address, offering, related links.
- 15% D3 · Answer-ready content — can a standalone, quotable answer be lifted for the question — or does it get lost in long copy.
- 15% D4 · Entity & NAP consistency — do name, address and phone match everywhere, so the model doesn’t confuse you with another business.
- 25% D5 · External presence — is the business talked about elsewhere: reviews, directories, independent mentions, press. This is the heaviest weight.
- 5% D6 · E-E-A-T signals — is credibility visible: HTTPS, a named author, legal pages, cited sources.
- 5% D7 · Content depth — is there substantive, on-topic content, and is it linked together across key pages.
I don’t promise a guaranteed first place, because no honest person can. I promise that you will see your real position precisely — in both modes — and get a realistic, step-by-step plan to enter the answers deliberately instead of by luck. The full rubric is on the methodology page; the SEO-vs-GEO distinction is explained here.
Implementation partners
The audit is deliberately vendor-independent: any developer can price and deliver the action plan. If you don’t have a trusted team, I recommend two partners. SmartArt has been building websites and applications — and designing for screen and print — for fifteen years; DataStuds specialises in data and AI integration: automated reporting, data pipelines and analytics for SMEs. The recommendation is disclosed and optional — the plan works with or without them.
the plan is vendor-independent — it works with them or without them
Are you a developer or an agency interested in partnering? Get in touch.
