How AI search works

No. Measurement found nearly 30% of cited domains did not appear in the first-page results shown alongside the Overview, so selection is not a re-ranking of the visible top results and position alone does not decide it.

Not completely. Google describes them as built on its core ranking systems, which likely determine the candidate pool. Independent measurement indicates the step that chooses which candidates get cited applies criteria distinct from ranking order.

In a large study of trending queries, they appeared on 13.7% of searches overall but 64.7% of question-form queries. Activation varies considerably by topic, so measuring your own category matters more than the average.

It helps systems understand and index your content, which supports being considered. Controlled testing of content factors found formatting-level changes had little direct effect on which candidate gets cited, so treat schema as groundwork rather than strategy.

There is no good evidence that they do. Google has repeatedly stated it does not use those analytics metrics as ranking signals, and the apparent link is better explained by good answers holding attention rather than attention causing selection.

Yes. Researchers found 11.0% of atomic claims were unsupported by the cited pages, with omission the most common failure, meaning qualifications can be dropped. Keeping conditions inside the same sentence as the claim reduces the risk.

Audits consistently find a narrow group of domains cited repeatedly, alongside a very large tail of domains cited only once or twice. Recognised reference sources, established publications and organisations with clear subject authority dominate the repeatedly cited group.

Not reliably. Research auditing four engines found roughly 16% of cited sources showed evidence of being AI-generated, so citation indicates that a page was retrievable, well structured and topically matched, rather than independently verified as authoritative.