B2B / brands
The classic signals looked fine: brands were found, relevant pages were indexed, there was organic traffic. But that gave no answer to what a potential customer actually sees when they put their need to Google or to an AI system. Simply Dom built a measurement framework that keeps those different forms of visibility apart — by brand, language, market, intent and environment.
The situation
The starting question sounded simple: how visible are our brands, really?
There were enough classic signals to give a quick, reassuring answer. The website got found, brand names produced results, product and category pages sat in Google, and outside points of sale surfaced the brands too.
Except that one question turned out, in the meantime, to be two questions. How visible are we in classic search results? And: what happens when someone frames that same commercial need as a question to an AI system instead?
On that second question, the existing SEO overview gave no reliable answer.
The business question
The problem wasn't a lack of data. There was plenty of it: Search Console, rankings, website traffic, external mentions, product pages, dealers and other sources. On top of that came new observations from AI Overviews, AI Mode, ChatGPT and other systems.
The first risk was throwing all those signals onto one pile and drawing a single general conclusion from it — "we're well visible" — without knowing where, for which question, or against whom.
The second risk sat in the measurement method itself. One positive AI outcome proves little. But a self-invented percentage based on an arbitrary number of prompts doesn't automatically prove much either.
Before anything could be measured, what counted as a valid observation had to be defined first.
The choices made
Classic Search and AI environments weren't treated as one averaged channel. Google Search stayed its own measurement environment, Google's AI features were looked at separately, and other AI systems each got their own set of observations.
The commercial question came before the tool. Which search intents actually matter — brand awareness, product selection, comparison, purchase advice, technical information, finding a supplier? For which brand, in which language, in which market?
Only after that were test questions built.
Per run, it wasn't just recorded whether the brand itself showed up, but also which other brands got named, which sources were visible, and what kind of answer the system gave.
Assumptions were simply labelled as assumptions. A chosen number of runs is a working convention, not a law of nature. A language split is a research choice, not a proven weighting. And a self-built visibility score doesn't become official just because there's a percent sign after it.
My role
My role didn't start with "optimise the SEO" but with clarifying what we actually wanted to know.
I mapped out brands, markets, languages and commercial search intents, and looked up what Google and the other platforms themselves officially document about how their search and AI experiences handle websites and sources.
Claims from tools and vendor documentation were traced back to their original source wherever possible.
I then built the measurement framework: test questions grouped by intent, environments kept strictly separate, mentions, competitors and sources used all recorded.
Where an observation wasn't reproducible enough to hang a conclusion on, it wasn't presented as stronger than it actually was either.
And the methodology itself was challenged just as hard as the outcome.
An arbitrary measuring stick doesn't get better just because you measure with it very precisely.
Execution
The result wasn't a magic GEO number, but a usable picture: a matrix by brand, language, market, intent and search environment.
It made visible where the brand itself showed up, which competitors kept recurring, which own and external sources played a role, and where classic Search stood strong while other environments were less clear-cut.
What surfaced above all were concrete gaps that something could actually be done about. Sometimes that was technical findability, sometimes the structure or content of a page. Sometimes a clear source was simply missing around a question customers genuinely search for an answer to. And sometimes external information about the brand turned out to matter at least as much as what was on the brand's own website.
That didn't lead to a generic list of GEO optimisations, but to a priority list: act first where observation and commercial relevance meet.
Then measure again, with the same yardstick.
Only then can you tell whether something is genuinely changing — or whether you're simply looking at a different snapshot in time.
What was deliberately not done
No declaration that SEO is dead.
No promise that a schema tag pushes a brand into an AI answer.
No building hundreds of pages around presumed fan-out queries.
No presenting a single one-off ChatGPT prompt as market research.
No self-invented minimum run count subsequently dressed up as "scientific".
No arbitrary weightings to squeeze Google, ChatGPT and the rest together into one impressive percentage.
And no recommendations made simply because "GEO matters now".
Measure first, then see where the difference actually lies, and only after that decide whether — and what — needs to change.