B2LLM: when the reader is a language model

What?

My doctorate (Hull York Medical School, 2025) found that COVID-19 completed a shift from a ‘deferrer society’ to a ‘referrer society’: people no longer simply accept what authority tells them; they check, compare, and increasingly ask an AI. In April 2026, on a Reuters panel asking what declining web traffic means for HCP engagement, I put a name to where this leads and coined ‘B2LLM’: the channel between an organisation and the large language models that now sit between it and its audience.

Today, Sacit Can Soyen, whose thesis on generative AI and Generative Engine Optimisation in cardiovascular therapeutics I second-supervised, defended his viva and passed with a double distinction: the top mark of 1.0 for both the written paper and the defence. I could not be prouder.

So what?

Here is what keeps me at it. A large language model answers in one confident voice, which means a falsehood dressed in the right words can reach a clinician looking exactly like sound advice: a wolf in sheep’s clothing, and the sheep’s clothing has never been easier to sew. If the careful people opt out of this space, the models get furnished by whoever optimises hardest, not by whoever is most correct.

So optimising for these engines, done honestly, is not a dark art. It is the same instinct that started WatMed Media: go to where people actually are, on the platform they actually use, and make credible health information legible there. The platform used to be social media. Now it is the model itself.

Now what?

Two commitments. First, I am upskilling myself in the open: this website is a live experiment in whether careful, well-structured content can surface accurately inside the models, and I am learning from what does and does not get cited. Second, I am turning that learning outward, building a curriculum to upskill my colleagues, so that the people with the most accurate information to share are also the most able to have it found.