Methodology
Every number we publish or deliver follows the same chain: observed market demand → buyer questions → repeated AI measurements → verbatim evidence → market intelligence. This page states the method and, just as importantly, what each step does and does not allow us to claim. If a vendor cannot show you this page for their own numbers, ask why.
Nobody outside the AI companies knows what people type into ChatGPT at scale — including the vendors who claim to. We start instead from demand that is observable: real search volumes in your market and language. Every question we measure traces back to a demonstrated buyer intent observed in search demand, with the derivation kept.
What this licenses us to say: "this question represents demand people demonstrably have." What it does not: "this is what people type into ChatGPT" — nobody has that data, and we do not pretend to.
Demand is translated into the questions a buyer would actually ask an AI — in the market's own language and phrasing, across the stages of a real decision: discovering options, evaluating them, pricing, trust, comparison, choice. A question set is written for one market; Portugal is not Brazil, and Spain is not "Spanish-speaking users". A stiff translated question often returns no answer where natural native phrasing returns a full one.
Each question is asked on each engine — ChatGPT, Google Gemini, Perplexity, Google AI Overviews — in repeated runs. Repetition is not thoroughness theatre: identical prompts return different brands 40–60% of the time, so any single check is a coin flip. The measurement configuration is frozen and hashed; when we compare two runs, we can prove they were the same experiment, and when the configuration changes, the comparison says so instead of pretending.
For every answer we store, at measurement time: the verbatim response, whether the brand was named, its position, whether it was recommended / merely mentioned / discouraged, the sentence that decided that classification, every cited source with its owner, and the timestamp. Every number in a BrandsNode report opens to the answer that produced it. A number that cannot be opened is not shown.
We report rates across runs with the observed range — "present in 47–54% of answers" — because that is what the data supports. A brand's month-to-month change is only called movement when it clears the noise we measured for that market; otherwise it is reported as within noise. We would rather tell you nothing changed than sell you a story.
| We don't | Because |
|---|---|
| Publish a single "AI visibility score" as the product | One number hides the stage where buyers lose you, and hides the engine differences that decide what to fix. |
| Claim to know real ChatGPT prompt volumes | Nobody outside the AI companies has that data. Our demand figures are labeled with their real source. |
| Present inference as observation | "In the buyer journeys we tested, the brand appeared in X% of answers" — never "X% of buyers chose you". |
| Measure your market from somewhere else | Location and language decide whether AI answers at all, and with which brands. Each country is its own measurement. |
| Invent customer results | Research pages are labeled as our own measurements; the studied brands are not clients unless stated. |
One brand, one market: demand-derived questions, repeated runs across engines, every answer and source logged, and a prioritised list of what to fix — with this methodology behind every number.
Apply for a free auditNo cost, no card, no obligation. We reply with the audit or with an honest reason it would not tell you anything useful.