The devil is in the details of GEO.
Realistic prompt set
Real conversations, no brand names.
Prompts should reflect how people actually talk to AI models: conversations and long, detailed questions, not short Google-style queries.
A perfectly realistic prompt set doesn’t exist, because what people ask ChatGPT, Claude and Gemini stays private. The one exception is Google AI Overviews, which answers regular Google queries. So any prompt set is a benchmark: an approximation of reality, not reality itself.
We build your prompt set from datasets of real conversations between people and AI models and from modern intent taxonomies, which classify what users want from an AI model.
Our prompts contain no brand names. A prompt that names a brand gives it close to 100% visibility and says nothing about the market. We want to know when the AI model names a brand on its own.
Beyond visibility
An integral index of brand attention in AI answers.
Brands don’t care about mentions for their own sake. They care about what mentions produce: visits, consideration and, eventually, purchases. In AI systems, these outcomes can’t be measured directly yet.
Most GEO tools rely on proxies such as visibility, average position or sentiment. Each of them captures only one aspect of a mention.
At Citeus, we believe the closest measurable proxy for purchase intent is the attention people pay to brands in AI answers. Brand attention depends on whether the brand appears, where the brand stands, how well the mention matches the user’s request and how much detail the brand gets. We combine these signals into a single attention index, measured across many responses from multiple LLMs.
We also calculate source weights: which pages and domains drive this attention for your brand, for each competitor and for the category as a whole. This shows where to focus first.
Tested edits, not urban legends
A test environment built on your category.
We test every GEO edit on pages from your own category instead of relying on generic recipes. GEO advice often rests on such recipes, and, as everywhere else, universal advice can work very differently in your industry. Our test environment is built from pages in your category: the pages yours actually compete with in AI answers.
The test environment rules out outside factors such as search index updates and new competitor publications, so you can focus on the effect of the edit itself. It helps optimize existing pages and create new ones and shows a page’s GEO potential before the page goes live.
Grounded in science
GEO methodology based on research.
Our GEO methodology is based on the best available research, and we keep updating it as new papers come out. GEO interests not only marketers but also researchers. We read scientific papers ourselves, with our own eyes, the way it was done before AI: this way we don’t miss details or make mistakes in our conclusions.
We are not looking for what “works”: put that way, it’s a conversation for marketers. We want to understand what “works” means, how we know something works, whether there is evidence that this particular thing works and not something else, and why it works. We bring academic knowledge into practice and adapt it to the applied task of GEO.