GEO
GEO is the practice of getting your brand selected, cited, and accurately represented when generative engines answer a question. It is the same discipline we have always called AI SEO, approached with real machine learning instead of guesswork.
Generative Engine Optimization, or GEO, is search optimization for a world where answers are generated rather than listed. Traditional SEO competes for a position in a list of links. GEO competes to be the source a language model chooses when it writes the answer.
GEO and AI SEO name the same work. We use both terms because our clients do. We do not treat generative search as a separate discipline with its own rules. It is search optimization for language models, and that is what our whole practice is built around.
When someone asks a generative engine, whether Google’s AI answers, ChatGPT, Perplexity, or Gemini, the model moves through four stages:
1
Interprets the question
2
Retrieves source material (grounding)
3
Synthesizes an answer from multiple sources
4
Selects which sources to cite
GEO optimizes every one of these stages. The goal reaches past appearing. We work to have your brand represented accurately in the answer a reader actually sees.
Our approach comes from taking language models apart. DEJAN founder Dan Petrovic trained a language model from scratch, from raw noise, with a custom tokenizer and masked language modeling, to understand how these systems decide what to say. That foundation shapes how we read and influence a model’s view of your brand.
We start by probing the model directly, borrowing methods from mechanistic interpretability, the field devoted to understanding the inner workings of deep learning models. We measure what a model already believes about your brand, your products, and the entities around them.
We then steer it. In machine learning this is called model steering. We use what the probing revealed to build a strategy that corrects the weak, missing, or wrong associations in how AI perceives your brand.
Every engagement follows the same measured sequence, from diagnosis through to tracking:
PHASE 1
We measure what models currently believe about your brand. Token probability analysis and our Tree Walker map where a model is confident, where it is uncertain, and where it is simply wrong. Brand Relevance Scoring turns that into an exact probability for the question, “Is this brand relevant for this entity?”
PHASE 2
Language models understand the world as entities and the links between them. We map your core entities, the associations that should exist but are weak or missing, and any negative associations working against you, using our Query Fan-Out model to probe the full scope.
PHASE 3
When engines ground an answer, they retrieve and cite sources. Our Citation Mining tool reveals which domains and URLs get cited for your topics, how strongly, and what the model retrieved but chose not to cite. That gap is a selection rate opportunity.
PHASE 4
Not every question triggers a search. Our Query Deserves Grounding models predict which questions cause Google and OpenAI to retrieve external sources, so effort goes only where grounding actually happens.
PHASE 5
With the diagnosis complete, we optimize on-page content, off-page citations, and link signals, then measure the result. AI Rank tracks your visibility in generated answers over time, and AI Flux measures how volatile those answers are.
Book a call with our senior strategy team to talk through your project.
