Insights
Research is the engine, not the decoration. Everything below comes from our own measurements - real prompts sent to real models, with the full answers and sources stored. No estimates, no industry averages.
AI Search
In our test a fictional brand scored 36% AI visibility. Broken down, every mention came from prompts where its name had already appeared - the model was repeating it in order to say it knew nothing about it. Separated: 0% general visibility, 90% brand visibility. That is why we count the two apart. A metric you can raise by adding prompts with your own name in them is not fit to report to a board.
Content
Across 60 answers we measured that between 35% and 48% of brand mentions came from URLs in citations rather than from sentences. In Gemini, 62% of all characters in the answer were links. The conclusion for content is specific: a page a model cites as a source does not necessarily make the model recommend you. Those are two different things and you write for them differently.
Analytics & Measurement
Competitors detected automatically enter the set once per spelling - Alior and Alior Bank as two separate brands. Every such pair inflates the denominator and deflates your share of answers. On one project, merging the variants moved the share from 2.12% to 3.73% without a single change on the client's site. Measurement itself is sometimes the first thing that needs fixing.
Method
Prompts are sent to the models for real and the full answers are stored together with the sources they used. Every number here can be traced back to a specific sentence in a specific answer.
Insights will be organised into these areas as they are published. We would rather list three findings that exist than twelve categories that do not.