Methodology
Goppa asks five AI engines the questions a buyer in your market would ask, once a week, stores every answer, and reports a count with its sample size. This page is the whole method, so you can check any number we show.
Five engines, each counted on its own and never blended into a single score:
The answers are collected through DataForSEO. Copilot is announced and not measured yet; Claude is not measured. The free account measures one question on ChatGPT only.
The questions are the ones a buyer would type, written from what your site sells, in the language of the market you choose, or in English when we do not have that language. You can add your own questions and categories, and yours come first. What each plan includes, per site:
| Plan | Questions per site | Engines | Cadence |
|---|---|---|---|
| Free | 1 | ChatGPT | Once a month, 3 measurements in total |
| Growth | 75 | All five | Weekly |
| Pro | 65 | All five | Weekly |
| Agency | 40 | All five | Weekly, up to 25 measurements a month across the workspace |
These are the numbers each plan is sold with (see pricing). A site with few product categories can get fewer questions than the maximum, and the result then says so: the “n” in every “k of n” is the number of answers actually collected, never the plan maximum. We never fill the gap by repeating a question.
On paid plans each question is asked once on each engine every week. A measurement takes a few minutes. Every answer is stored with its measurement, and an older point on the chart opens the answers behind it.
An answer that only links to your site counts as Cited your site, not as Named. Until 2026-10-12 a link alone also counted as Named; older measurements are recounted with the new rule when their saved answers allow it, and the chart marks the change.
Every number comes with its total: 3 of 20 means 3 of the 20 valid answers on that engine. A percentage is shown only from 10 valid answers; below that you see the count only. A percentage comes with a Wilson 95% confidence band, rounded outward, because a small sample cannot support a sharp number. An answer that failed, or an engine that showed no AI Overview, is never counted as “not named”; it is marked as failed or not measured.
To say that something changed, Goppa compares the same questions on the same engine in two measurements and runs a paired test, McNemar’s exact test, on the questions whose result flipped. The result is Real gain, Real drop or Normal variation. When the questions, the engines or the Named rule changed between the two, the result is Not comparable. The report places work and results side by side and never says that the work caused a result.
You choose where to measure: a country, or a region such as Europe or Worldwide. The questions are asked in that country’s language when Goppa has it, and in English otherwise. Google always searches from one place, so Google AI Overview and Google AI Mode are measured from a country; for a region, Goppa measures the other engines. An AI answer also depends on who is asking, so we ask neutral buyer questions from one place, and the result describes that place, not every user.
An AI provider can change the model behind an engine without notice. Goppa stores the model version each answer reports. When the version changes, the series starts again from that point, the chart marks it as Model changed, and measurements from different versions are not compared.
ChatGPT, Gemini, Google AI Overview, Google AI Mode and Perplexity, on every paid plan. The free account measures one question on ChatGPT. Claude and Copilot are not measured today.
Each question is asked once on each engine every week on paid plans, and the answers are stored. The free account measures once a month, on ChatGPT only.
Named means the answer mentions your business in its text. Cited your site means the answer links to your site or uses it as a source. An answer that only links to your site counts as Cited your site, not as Named.
A count with its sample size is the honest number. A percentage appears only from 10 valid answers, with a Wilson confidence band, and each engine is shown on its own.
It compares the same questions on the same engine in two measurements with a paired test (McNemar) and says Real gain, Real drop or Normal variation. If the questions, the engines, the Named rule or the model version changed, it says Not comparable.
About Goppa · Get started · Read your results · Study · Plans and pricing