A Weighted Sum of Everything Ever Written About You
What Bayesian linear regression, copied by hand from a textbook I've owned since January 2007, tells us about getting cited by AI.
Ben Rees - 13 July 2026
The mechanism underneath everything I've written about AI visibility, worked through properly, with the maths on the page rather than gestured at. Why a language model's answer about your brand is a similarity-weighted sum of its training data, why the standard Bayesian model of brand choice is missing a term that matters once the thing doing the choosing is a language model, and what a diagram I drew by hand, in a textbook I've owned since a month after it was published, has to do with any of it.
↓ Get the PDF (free, just your email)Whitepaper on How to Optimise for AEO and GEO
Results from a two-month experiment testing what actually affects AI answers across ChatGPT, Gemini, Perplexity, Copilot and Google.
Making decisions in a Bayesian world
Most marketing decisions can't be A/B tested. Bayesian logic, combining prior knowledge with whatever data you do have, fills the gap.
My AI visibility went from zero to 26% in six weeks. The one category still at zero is the one this post is about.
Six weeks after publishing a zero-appearance rate across every AI platform, most categories moved. The one still at zero is AI Search Visibility itself.
Your brand doesn't get its own space inside an AI model
Language models pack far more concepts than they have space for, a phenomenon called superposition. Generic content competes for that same overcrowded space as everyone else's, and usually loses.