“We’re in Yandex’s top results, so Alice recommends us too, right?” Copywriter-marketer Vyacheslav Savitsky says clients ask him this more and more. So he tested it, and published the results on vc.ru on 17 October. He took 30 local company websites from the first pages of Yandex and asked Alice AI and ChatGPT for a recommendation, the way a customer would. Across three rounds of checks, the two assistants named only 9 of the 30. The other 21 were never named.
The headline finding is that search position “decided almost nothing”. What did matter was Yandex Maps ratings, reviews and mentions in media round-ups.
How the test worked
He tested five niches, each in a different city: freight in Kazan, dentistry in Yekaterinburg, flat renovation in Novosibirsk, bankruptcy lawyers in Krasnodar and travel agencies in Samara. For each niche he took the first six company sites in Yandex results with a head office in that city. He left out ads, aggregators such as Avito, 2GIS, Profi and ProDoctorov, regional subdomains of national chains, and out-of-town firms. Even that step was revealing. For “грузоперевозки казань” (freight Kazan), the first page had no Kazan company at all, and he had to go to pages 2–5 to fill the list.
He asked each assistant questions like “Recommend a freight company in Kazan”. He used alice.yandex.ru and ChatGPT in a temporary chat, with a new window for every question. The checks ran on 3 and 4 October 2026. That was before the data-centre strikes that knocked out much of Yandex’s infrastructure, so this is Alice working normally.
What he found
- Answers vary. On 3 October Alice named 7 of the sampled firms and ChatGPT named 5. On 4 October Alice named 7 again, but one firm had dropped out and another had come in. Alice named 25 companies in total across the five cities. Most of them were national players, or local firms that ranked lower.
- Ratings were the common factor. In 4 of 5 answers Alice showed organisation cards with a rating and a review count, and every company it named was rated between 4.2 and 5.0. In Samara, three of six agencies were named by both assistants. Two of them have a 5.0 rating on Yandex Maps and between 165 and 194 reviews. ChatGPT also built every list from map data with ratings.
- Media lists get repeated. In Novosibirsk, Alice relied on a kp.ru ranking of renovation firms on both days. In Yekaterinburg, it repeated a clinic’s “No. 1 dentist according to Delovoy Kvartal” claim almost word for word.
- Technical factors didn’t make the difference. None of the 29 sites he could check blocked AI bots in robots.txt. Fourteen had Organization schema, but only 3 of the 8 firms named on day one did. Two sites had llms.txt, and one of them was named.
The informational questions were where small sites won. Alice was asked what renovating a one-room flat in Novosibirsk costs. It put seven footnotes on a single article from a small firm, Avantstroy. That firm’s site sits on page three of Yandex for “ремонт квартир новосибирск”. The article doesn’t even have an author or a date. Its title repeats the question, it is specific to the city and the year, and it gives figures that are easy to quote. For price questions, Alice took figures from local company sites in four of five cities. For “how to choose” and “what to watch out for” questions, no local site was cited at all. Advice came from media, bank magazines and clinic blogs in other cities. Across 73 links in ten Alice answers, local company sites and blogs made up just 11%.
Our take
It’s a small sample, and the author says so himself: 30 sites, few repeats and personalised accounts. But a 9-versus-21 split is too big to dismiss, and it fits what Yandex is building. Alice is already answering about 15% of ecommerce queries with product cards that aren’t from your site. Yandex’s own course on getting cited in Alice tells writers to give a precise answer backed by facts and experience, “not another retelling”. That is exactly what the Avantstroy price article did. The 295,000-answer Neuro scrape found the same pattern at scale: platforms such as Dzen, VK and Wikipedia were cited far more often than brand websites.
For local businesses, the order of work looks clear. Fix the Yandex Maps card and reviews first. Then publish prices on service pages in plain text. Write city-and-year articles that answer “how much does X cost in Y”. Then work on getting into the round-ups Alice reads, even though it warns users that many of those rankings are paid. And measure monthly, not once. Within a day, the list of firms Alice names has already changed.