The product

From real demand to what AI actually answers about your brand

One chain, eight steps, sixty seconds. Every step below is what we run for a customer, and every example is a real number from our own published measurements in Portugal and Spain. Engines measured: ChatGPT, Google Gemini, Perplexity. Google AI Overviews and AI Mode are checked as search surfaces, not as chat assistants.

Step 1 — Demand

Every question we measure traces back to demand you can verify

We do not guess prompts. We start from observed search demand in your market and your language, and keep the derivation, so any question in your report can be traced back to a demonstrated buyer intent. Nobody outside the AI companies knows real ChatGPT prompt volumes — including vendors who imply they do — so we use the one empirical proxy that exists and label it honestly.

Real example. The Portuguese question set was derived from observed Portuguese category search volumes and written in European Portuguese — not translated from the Spanish set, because Portugal is not Spain and neither is Brazil.

Step 2 — Questions

Two kinds of question answer two different business questions

Buyers ask AI in two modes, and a measurement that mixes them tells you nothing useful. We run both and keep them separate.

Most tools report one blended number. A brand can look healthy because it defends its own name well while being absent from every question a new buyer asks — which is the only half that grows the business.

Step 3 — Measurement

One check is a coin flip, so we never run one

Identical questions return different brands a large share of the time. Every question is asked on every engine, in repeated runs, with the configuration frozen and hashed so two runs can be proven to be the same experiment. Every verbatim answer, position, classification and cited source is stored at measurement time.

Real example. The Portuguese study was five repeated runs of the same measurement, 72 answers per run, every answer and every source kept on disk.

Step 4 — Visibility

The same brand was in up to 88% of ChatGPT answers and 17% of Perplexity's

Ranges are the variation between identical runs, not error bars we invented. Two readings follow from this table. A brand measured on one engine is reading somebody else's market. And AI visibility is not a property of a brand — it is a property of a brand in a market: the same company, the same method, the same period, roughly twice as visible in Portugal as in Spain.

Step 5 — Recommendation

Being mentioned is not being recommended

We classify each answer as recommending, merely mentioning or discouraging the brand, and store the sentence that decided it. In the Portuguese study the brand was mentioned in about half of all answers but recommended in fewer than one answer in ten — most mentions place it mid-list, without the sentence that closes a decision. A single "visibility score" hides that gap completely.

Step 6 — Competitors

The answers you are not in are still naming somebody

Every brand named in every answer is recorded, so the brand-free questions produce the list of who buyers find instead of you, per engine and per stage of the decision. In the Spanish measurement the brand appeared in 24%–31% of answers; the remaining answers named other brands, and that field — not a score — is what a plan can act on.

Step 7 — Sources

Three in four cited sources were not the brand's own site

Every citation in every answer is logged with its owner. In Portugal, only about a quarter of citations (24%–26%) pointed at the brand's own sites, rising to 63%–67% inside ChatGPT alone; in Spain the own-site share was 21%–33% depending on the run. The rest were directories, press, comparison articles and reviews — pages the brand does not control but can influence. That list is the practical work: source by source, answer by answer.

Step 8 — Analytics

You also see which AI systems fetch your pages and which send you visitors

Underneath the measurement sits first-party analytics: a lightweight script on your site plus your server logs, showing which AI crawlers request your pages and which AI assistants refer visitors to you. It is a layer, not the headline — measurement tells you what AI says about you, analytics tells you what AI does with your site.

What you get, and what you decide

The output is a per-market brief: the visibility picture with its evidence, the competitor field, the cited-source map, and a prioritised list of what to change next — every number opening to the verbatim answer that produced it. See a worked example in the sample report, or the full method in the methodology.

Plans are priced by the number of markets you monitor, from €150 per month. See the pricing page for what each plan includes.

Do you score sentiment?

No. We quote the assistant's exact words as evidence — we don't score feelings. You get the sentence that called your brand "expensive" or "the safest option", with the engine, the question and the date. A sentiment number averages away the one sentence you needed to read.

Do you execute the changes?

We show you where you lose and why; what you change is your decision. Our work is the measurement, the evidence and the prioritised diagnosis. Your team or your agency executes — and the next run tells you, honestly, whether it moved.

How is this different from an AI visibility tracker?

Trackers start at the prompt. We start at demand, weight by it, split brand-free from brand-named, repeat every run and keep the evidence behind each number. Where a head-to-head helps, we keep an honest one rather than a straw man.