What has actually changed
When a language model writes the answer it selects sources on different grounds than a results list. It looks for content that answers the question directly, can be lifted out of its context without losing meaning and comes from a sender the model can identify. Pages that rely on the visitor scrolling, clicking and interpreting disappear from that equation.
What has not changed is the foundations. A site that is fast, crawlable and clearly structured was an advantage in classic search and is a requirement in AI search. The difference is that the penalty for vagueness has become harsher: the model does not guess, it picks another source.
Three principles that steer the work
One answer per section
Every heading should read as a question and the paragraph below as the answer, complete in itself. That is how content gets cited, and it happens to be better reading for humans too.
Structured data that describes reality
Schema markup for the organisation, people, services and articles gives the model entities to anchor the answer to. The markup should mirror what is actually on the page, nothing else: inflated markup is a trust problem, not a shortcut.
An identifiable sender
Named authors with verifiable backgrounds, an organisation with a clear profile and consistent information across the web. Models weight the sender harder than classic search ever did.
How it is measured
AI search still lacks an equivalent of Search Console, so measurement has to be built: recurring control questions against the major models, logging which sources get cited and traffic analysis on the pages built to answer. The important part is measuring against the business, not visibility itself.
Being cited is a means. Pipeline is the measure, just like in every other kind of search.
That is why we write content that answers the question from the start, in parallel with the classic search optimisation. The same page carries in both channels: build once, be visible wherever the question is asked, whether a human or a model is the one asking. If you want to know what you can already be cited for, the AI-search work starts with exactly that review.