How to Get Cited in Google AI Overviews Without Ranking Top

Google's AI Overviews — the generated answer block above the normal results — are usually discussed as an English-language, American problem. A new analysis published on Habr suggests otherwise. The author ran 1,095 Russian-language questions through Google with the region set to Russia and found an AI Overview on top of the results in 93% of cases. More interesting for a small business: half the sources cited in those answers were not on the first page of regular results.

What the data shows

The questions came from 55 sets across different niches. For four of those sets, the author compared the domains cited inside the AI answer with the domains ranking on page one for the same question. Out of 446 cited sources, 222 sat on first-page domains. The other half came from somewhere else in the index.

That matches how Google describes the mechanism in its own documentation. A user's question is split into several sub-questions, each one runs as a separate search against the index, and the answer is assembled from what comes back. Google calls this query fan-out. Because the system searches by sub-question, it reaches past the usual ten blue links.

The requirement for a page to be eligible is narrow: it has to be in Google's index and able to appear in results with a snippet — the short description under the link. No llms.txt, no special markup for AI answers.

Two more findings are worth holding onto. First, 97% of the sources cited for Russian questions had Russian-language titles, so the answer was built from local content rather than translated English articles. Second, no one owns a permanent seat: ru.wikipedia.org, vc.ru and habr.com appeared most often, but even Wikipedia showed up only 44 times across 1,095 questions. Everything else was spread across hundreds of sites.

Why it matters for a small business

The honest version is that this is about visibility more than traffic. Pew Research tracked almost 69,000 searches by 900 users in the United States. When an AI Overview was present, people clicked a regular result in 8% of visits, compared with 15% when there was no overview. Links inside the overview itself were clicked in 1% of visits. So the answer block takes attention away from the results below it, and being named in it mostly means being seen, not being visited.

Seen still counts. If a prospect asks a question about your category and your company is the one named in the answer, that is brand presence you did not have to buy. And since half the citations came from outside the top ten, a small site that answers one narrow sub-question well has a real chance where it has no chance of outranking large portals.

Scale, where the study measured it: per StatCounter, Google accounted for 28% of search referrals to Russian sites in September 2026, with Yandex at 70%; other counters put Google closer to a quarter. Google's own help lists Russia among the countries where AI Overviews run, and Russian among the supported languages. Wherever your customers search in a language other than English, the same logic applies — the answer is assembled from content in their language.

What to do about it this month

• Confirm you are in Google's index at all. Open Search Console, check how many pages are indexed, and make sure Googlebot is not blocked in robots.txt. Sites maintained only through one search engine's webmaster tools often never check this.
• Audit your snippet directives. Ask your developer to look at the page templates for noindex, nosnippet, max-snippet and the data-nosnippet attribute. A page-wide nosnippet makes the page ineligible for snippet display, and therefore for the overview.
• Write down the sub-questions. Take five questions customers actually ask, search them in your customers' language, and read the overview: it is normally split into blocks with subheadings, each with its own links. Give every sub-topic its own section on your site, with the direct answer in the first paragraph after the heading.
• Make answers short and checkable. The model lifts paragraphs that stand on their own — a number, a deadline, a price, a condition. "It depends on many factors" gives it nothing to quote.
• Look at who is being cited in your niche and work on appearing there. The list of sources in the overview for your own questions is the list.

One caveat the author states plainly: the region was set through a query parameter, not by searching from a Russian IP address, and a real user there may see overviews less often. The questions came from client niches rather than search suggestions, so the 93% figure would look different for everyday consumer searches. Also, when the same region was queried in English, only two .ru domains appeared among 341 sources — which is the whole point. Test in the language your customers use, not the one you read industry news in.