AI models are learning to protect their answers from error. A wrong fact about a bank or an energy tariff is a complaint and a correction and sometimes a regulator. So the models are being taught to prefer sources that can be checked. A page with a named author and a date and a source behind each claim is cheaper for a model to trust than a page that asserts the same thing with none of those.
I think that shift will grow and that it changes what a brand's website is for. The job is no longer to rank for a query. The job is to be the page a model can safely repeat.
Two things follow. The first is about the data. Most of what a model says about a brand has never been checked by anyone. The brand should check it against its own facts on a schedule and that is a data quality job and not a public relations job. The second is about the content. Governing your own content means showing the expert who wrote each page and what qualifies them. It means showing when the page was last checked and what it draws on. It means naming who endorses it and what their stake is. It means one version of each fact across every page the brand owns. And it means saying all of that in structured data so a machine can read it as a person would.
What I have counted so far is that the models read a handful of pages and repeat them. Which pages win is the question the next readings will answer and I will publish the counts here.
My details are on the contact page if you have any questions or comments. Please get in touch.