Answering is not ranking
A ranked list rewards the page that best matches a query. A generated answer rewards the source that is easiest to quote accurately. Those are related but not the same, and optimising for the first does not automatically deliver the second.
The practical consequence is that a page can rank well and still never be drawn on, because its claims are vague, unsourced, or buried in copy written to fill a word count.
What assistants actually use
In our experience three things matter more than they used to: an unambiguous definition of what an organisation is and does, specific claims that carry their own evidence, and structure that makes both machine-readable.
None of that is new advice. What is new is that failing at it now costs visibility in a surface where there is no second place.
Where to start
Ask an assistant what your organisation does, then ask it where that came from. The sources it names are usually the real problem, and they are frequently not pages you control.