When someone asks ChatGPT "what's the best project management tool?" and three brands come up — but not yours — what exactly happened? It wasn't random. There's a process behind it, and understanding it is the first step toward doing something about it.
Most marketing teams still treat AI visibility as a black box. This post opens it up, without oversimplifying and without requiring a background in machine learning.
There's no internal ranking list
Start here: LLMs don't maintain a ranked list of companies by category that they consult when generating a response. There's no live "top 5 CRMs" index being queried under the hood.
What exists instead is a high-dimensional representation of learned knowledge. During training, the model processes massive amounts of text — articles, forums, reviews, documentation, news coverage — and builds associations between concepts. "Project management" ends up close to "team coordination", "task tracking", and also to brands that consistently appeared in those contexts.
When a query comes in, the model navigates that network of associations to generate a statistically coherent response. Brands that surface are the ones that became strongly associated with the relevant context during training.
The three factors that matter most
Without turning this into a machine learning lecture, three dimensions carry the most weight:
Volume and consistency of independent mentions. How often and across how many distinct sources your brand appears connected to a given topic. A company mentioned in 50 independent publications has more weight than one mentioned 200 times only on its own website.
Source credibility. Models learn to weight content by origin. An article in a major industry publication carries different authority than a low-traffic blog post. Wikipedia entries, recognized trade publications, and sources with a history of accuracy tend to carry more influence.
Contextual specificity. A brand that clearly owns a specific niche — say, "churn analysis for SaaS companies" — tends to appear more reliably in queries about that niche than a generic analytics platform does. Clarity about what you do and for whom matters a lot.
Real-time retrieval changes the equation
Beyond what was learned during training, many modern models (particularly ChatGPT with web search enabled and Gemini) perform real-time retrieval — they consult live web sources while generating a response.
This shifts the calculus somewhat. Here, what matters is whether pages about your brand are recent, relevant to the query, and structurally easy for the model to parse. Structured data, clear page titles, and explicit context about what you do all help the model extract meaningful signals about you.
But retrieval doesn't override base knowledge. Models tend to favor sources that align with what's already in their training. A brand completely absent from training data will have an uphill battle appearing even through retrieval.
Why your brand might be disappearing
A few patterns that consistently cause brands to go missing from AI responses:
Your own content talks mostly about yourself rather than the problems you solve. "We're leaders in innovation" builds no semantic association with anything specific. "We help support teams cut response time in half" does.
External mentions are sparse or concentrated in a few domains. A press release syndicated across 30 outlets copying the same text isn't the same as 30 independent mentions. Models recognize syndication patterns.
Your brand name is ambiguous or shared with other entities. This creates semantic noise that dilutes the associations that should be anchored to you.
There's no clear primary use case or niche. Brands that try to appeal to everyone tend to get cited for nothing in particular.
What you can actually influence
The key distinction is between what lives in training data (retrospective, hard to change quickly) and what's available through real-time retrieval (present-tense, more actionable).
For training data in future models, the strategy is long-term: build consistent editorial presence, earn mentions in authoritative sources, participate in industry conversations that leave a searchable record.
For real-time retrieval, the timeline is much shorter. Well-structured pages, clear FAQ sections, schema markup, and current content all help models find and cite you when search is active.
And before any of that: knowing how you're being cited today is the baseline. Which models mention you? For what kinds of queries? With what framing? Without that picture, optimization efforts are working blind.
Genoma tracks exactly this — share of voice, sentiment, and citation accuracy across the major LLMs. If you don't yet have that visibility, that's where to start.