in 93% of cases. In a four-set comparison, 222 of 446 cited sources were on domains from the first ordinary results page; 224 were not. The article also reports that 97% of sources for Russian queries had Russian-language titles.
The mechanism described is query fan-out: Google breaks a question into subquestions and searches its index for each, potentially widening the set of pages that inform the overview. The article says a page must be indexed and eligible to appear with a snippet; it does not establish how each of the 224 sources outside the first page was discovered. The implication is that ordinary first-page ranking alone may not describe which sites can be cited in an AI Overview. That is a visibility hypothesis, not evidence that a particular optimization reliably earns citations.
Why now: this study supplies Russian-language observations in a search feature often discussed through Western-market examples. Its evidence boundary matters: the Russia setting was configured in the query rather than tested from a Russian IP, and the questions came from client niches, not a random sample of searches. Watch whether independent, locally executed samples reproduce the citation patterns across broader query sets.