Two measurements sit behind this. Pew found that where a Google search produced an AI summary, users clicked a result on 8 percent of visits against 15 percent without one, and clicked a source cited inside the summary on 1 percent. Semrush clickstream analysis puts growth in outbound referrals from language models at roughly 206 percent year on year, from a small base.
Together those describe a reader who increasingly meets your material as a paraphrase rather than as a page.
What a page has to do to survive that
Put the answer first under a heading that is the question. A summariser lifting a passage takes the one that resolves the query completely. If your answer is assembled across four paragraphs with the payoff at the end, there is no passage to lift and a competitor with a tighter opening is quoted instead.
Make the passage stand alone. A paragraph that depends on the two above it for its subject cannot be extracted. Naming the subject inside the sentence costs a few words and is the difference between being quotable and being skipped.
Include something specific in it. A definition, a boundary, a number with its source, a named method. General competence is unquotable because every competitor claims it in the same words, so a summariser drops it as redundant.
What this costs
It costs the withheld answer. Pages built to make a reader stay, or to trade the answer for an email address, are built for a traffic model in which arriving on the page was the event. Where the answer is being read inside a summary, withholding it means you are simply not the answer, and the buyer's question is resolved by someone else.
What we take from it
We hold the click data as strong evidence about direction and weak evidence about magnitude in any particular market, and the structural conclusion does not depend on the magnitude. A page organised around complete answers to real questions was already the better page for a human reader. What has changed is that the alternative no longer works at all.
The honest caveat is that this is a moving target set by systems whose behaviour is not published and changes without notice. Anyone offering a technique with a guaranteed result here is describing something they cannot know. Structuring for clarity is the version that survives the next change, because it is not a technique.