In short
- At scale
- Many brands and many questions at once, instead of checking one brand by hand.
- Multi-engine
- Several engines side by side. Measuring one engine and generalising the answer is the commonest mistake in this field.
- Repeatable
- Someone else has to be able to redo the same measurement and land on the same result. Without that it is an anecdote.
- Comparable
- Only brands measured with the same questions and the same method may be put side by side.
What does the study say?
The contribution lies in the measurement framework, not in a single number. The study shows how to run the same measurement across brands and engines without skewing the comparison — which is harder than it sounds, since every engine answers differently and at a different moment.
| What was measured | How | What the study does not say |
|---|---|---|
| brand visibility across several AI search engines | a scalable, repeatable measurement framework with identical treatment per brand | which brand ought to win — it supplies a method, not a ranking |
(GEO research) (2026). Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines. arXiv 2606.20065
Why does this matter to you?
Because a number about your visibility only means something next to someone else's. And that is only possible if both were measured with the same questions and the same method. A ranking in which every brand is measured differently is a ranking of measurement methods.
What does ceeme do with it?
Our index applies this principle to sectors in the Benelux: the same question set, the same engines, the same month. And the report states which engines count towards the weighting and how much of the market that covers — measuring only Claude in Belgium is 4.2% of the market, and the reader should know that.
Frequently asked questions
Is this the same as your index?
It is the methodological grounding for it. Our index applies the principle to Benelux sectors; the study supplies the framework that keeps such a comparison fair.
Can I compare my number with a competitor's?
Only if both were measured with the same questions, engines and period. If that is not stated, you are comparing two different measurements — and the gap between them says nothing about the two of you.
Why not simply measure every engine?
We measure more of them than we weight. An engine whose market share in your market is unknown appears in the table as measured but not weighted. Giving it an invented percentage would pollute the number above.
Read on
Free, no account and no card.