AI answers in Google and Yandex: fewer clicks, and how a site becomes a source

Search for something like "how to choose a stretch ceiling", and there is a good chance a ready-made answer appears above the links. In Google it is called AI Overviews. In Yandex, the main search engine in Russia, it is a quick answer from Alice AI.

Convenient for the user: read the answer, close the tab. For the site owner, rankings stay the same while clicks go down.

At RM Systems I build websites for businesses in Russia, so I watch both engines. Here is what these answers are, how many clicks they take, how they pick sources, and what you can check today.

What these answers are

Google shows AI Overviews above the results and has a separate AI Mode where you talk to search like a chatbot. According to Google, AI Mode is used by more than a billion people a month.

Yandex shows quick Alice AI answers above the results, with links to the pages the answer was built from. According to Yandex, in Q2 2026 quick answers appeared for about 42% of queries and reached 49.5 million people a month. Yandex estimates its own share of the Russian search market at about 70%, so for most Russian businesses Alice matters first.

How many clicks they take

The research points the same way:

• Pew Research, US, 2025. Almost 69,000 searches from 900 people. With an AI summary, people clicked a regular result in 8% of cases, without it in 15%. Links inside the summary got clicks in 1% of cases.
• Indian School of Business and Carnegie Mellon, 2026. A browser extension hid AI Overviews for some users. Where the summary was shown, clicks to sites were 39.8% lower.
• University of Pennsylvania and Northeastern University, 2026. AI Mode cut the share of clicks to external sites by 18.8 percentage points.
• Russia. According to Kokoc Performance, informational sites have lost about 30% of search traffic on average since early 2025, and from 15% to 60% in some categories.

Informational queries ("how", "what is", "why") suffer most. Commercial queries less so for now, but that is changing: according to Ashmanov & Partners, the Alice block in Yandex commercial results grew from almost zero to about 15% of queries between April and July 2026.

Google says clicks from AI summaries are "higher quality". The ISB and Carnegie Mellon experiment found no difference in click quality, so this claim is still open.

Where the answer gets its sources

Google says there is no special optimization for AI features: the page has to be indexed and eligible to show with a snippet, and normal SEO rules apply. No special files or markup "for AI" are required.

Yandex describes it the same way: search finds relevant pages, the model picks the most useful fragments and builds an answer with links. Only indexed pages are used, and well-structured, informative ones are picked more often.

Independent data is stricter. According to Ashmanov & Partners, an Alice answer relies on about 10 sources on average, and the share of sources outside the top 10 dropped from 40% to 10% between August 2025 and January 2026. For extended answers, more than 90% of pages come from the first page of results.

Short version: the AI answer is built from regular search results. If a page is not in the top 10, it is almost never in the answer.

How to check your site in 10 minutes

1. Ask your customers' questions. Write down 10 questions clients actually ask you, put them into Google (and Yandex, if you work in Russia). Look at what the answer says and which sites it cites.
2. Check indexing. In Search Console (and Yandex Webmaster) see which pages are not indexed. A page outside the index will not get into results or into an AI answer.
3. Use the visibility report if you have it. Yandex Webmaster has a report showing for what share of queries your site is cited in Alice answers, with example queries and competitors. Google has no separate report yet: AI Overviews traffic is counted inside regular "Web" search in Search Console.
4. Watch the trend. Since AI traffic is not separated anywhere, look at overall clicks per page. Usually informational articles lose traffic first, not service pages.

What to do now

1. Fix the basics. Important pages indexed, no old URLs returning 404, fast on mobile.
2. Fight for the top 10 on your main commercial queries. Without this the rest barely works.
3. Answer directly. Price, timing, process and who does the work should be plain text on service pages, not only inside images. An FAQ block at the bottom helps, including uncomfortable questions.
4. Add markup. Organization and FAQPage help search engines understand the page. It must match the visible text.
5. Keep company details consistent. Name, address, phone, services and description should match on the site, maps, directories and social profiles.
6. Do not block crawlers without a reason. In Yandex you can block the YandexAdditional robot in robots.txt to stay out of Alice answers without affecting regular results. For Google, nosnippet or max-snippet also limit the regular snippet. For a business this is usually a loss.

What I found on my own site

I went through this list on the RM Systems site and found dozens of old URLs returning 404, plus service pages that were crawled but not indexed. Those pages cannot appear in results or in an AI answer, however useful they are.

One small but telling number: over the last four weeks my site got 17 visits directly from ChatGPT. Small, but people already arrive from AI tools, not only from search.

Does "GEO" actually work?

Honest answer: there is no separate magic. A 2026 review of 45 studies found no technique that reliably and durably lifted a site in AI answers across platforms. What works is the usual: the page answers the question precisely and ranks in the top results. So for me, visibility in AI starts with regular SEO.

Indexed pages -> top 10 on main queries -> direct answers on pages -> consistent company details everywhere -> regular visibility checks.