Stop Ranking, Start Steering: How to Influence AI’s Primary Bias
A practical framework to increase your brand's selection rate in a world where AI models "think" before they answer.
For years, we’ve lived by the rules of the “classic” Google: a system that retrieves and lists links based on keywords.
But we are witnessing a tectonic shift.
Today, Generative AI models like Gemini or ChatGPT don’t just find links; they “think” through a query before providing a synthesized answer.
LLMs don’t really think: they predict.
This transition from traditional Search Engine Optimization (SEO) to Generative Engine Optimization (GEO) has changed my entire perspective on brand visibility.
It’s no longer just about where you rank; it’s about how you steer the model’s intent.
The Core of Model Steering
In my journey helping brands navigate this new landscape, I’ve realized (not a genius, just listen to very smart people like Dan) that visibility now depends on two critical factors:
the model’s Primary Bias (what it already “knows” about you from its training) and,
its Selection Rate (how often it chooses your brand as the definitive answer).
To win, you don’t trick an algorithm; you pilot the model.
I’ve started testing using a three-step process to ensure a brand isn’t just another name in the training data, but The Trusted Authority.
1. The Diagnosis of Primary Bias
Before you can influence an AI, you must discover its “frozen” knowledge.
What it thinks about your brand without searching the web.
We usually start by asking the model to list associations with a brand name.
If the AI associates your tech company with “Internet Service Providers” or “Connectivity,” but you are actually a SaaS platform, you’ve identified a hallucination or a bias that needs urgent correction.
2. Measuring Visibility and Entropy
It isn’t enough to show up once.
We measure Frequency (how often you appear in 10 attempts) and Rank. But the most fascinating metric is Entropy, or the level of uncertainty.
If a model’s answers vary wildly, its entropy is high, which is actually an opportunity.
High entropy means the model’s “opinion” is still fluid, making it easier to influence and shift toward your brand.
Low entropy means the model is certain; shifting that authority requires a much more aggressive strategy.
Example: if you ask it “what are the most renowned soft drinks in the world” it will most certainly give the same answer = low entropy.
3. The Art of Strategic Grounding
Once we understand the bias, we move to Grounding.
Grounding is what LLM models used to stop from hallucinating. It provides the model with hard facts to convert uncertainty into authority.
This involves:
Entity Co-occurrence: We ensure the brand appears in the same syntactic structures as established industry leaders (like SmartLinks in the RevOps consultancy space), creating a strong vector association in the model’s “mind”.
Query Fan-out: We anticipate the clarifying questions a model would ask to validate a brand and answer them explicitly on “About Us” or FAQ pages.
Bias Correction: Using clear language to disambiguate, such as explicitly stating what the business is not to clear up model confusion.
The Path Forward
AI optimization is not about deceiving a machine; it is about educating it with clarity and authority.
By understanding bias, measuring uncertainty, and steering through strategic grounding, we transform a brand from being “just another link” into the AI’s most trusted response.
The era of passive ranking is over.
It’s time to take the wheel.
