How to get into AI answers.
MediaNation's podcast “Lida, where are the leads?” released an episode on AI search and GEO — optimizing for answers generated by neural networks. The main expert is Igor Sklyar, head of MediaNation's SEO department; Marina Shakhova hosts, and I asked the awkward questions. Here are the highlights for anyone without 43 minutes to spare. The episode is in Russian.
Search is no longer a list of links
Neural networks are the fastest-growing technology in history: by Igor's estimate, they add around 100 million users per quarter. Search engines are embedding AI answers right into results. So it's no longer enough for a business to rank a link — it needs to become the source of the answer itself.
The scariest numbers in the episode: in the US about 68% of Google searches end without a click, and when an AI answer appears above the results, the first position's click-through rate in some projects drops from 50% to 9%.
The market is still almost empty
According to research Igor cites, the share of websites that become sources for AI answers grew about fivefold in a year — from 1.5% to 6–6.5%. That's still very little. Igor compares the moment to the early days of SEO: getting in now is easier than it will be in a year or two.
Three steps to start with
- A visibility audit. Measure which questions AI already mentions your brand for, and which it doesn't.
- Presence on external platforms. AI trusts brand websites less, and review sites, forums, UGC platforms and authoritative niche media more.
- Put the answer in the first 30% of the text. In Igor's observations, the first 30% of a text becomes the cited source in 40–50% of cases. Answer first, then go deep. Long reads of 20,000+ characters and “best of” rankings work well.
The AI tells you where to publish
Platforms aren't chosen blindly. For priority queries you capture each relevant model's answers and collect the domains it cites. If a site is the source in 20% of answers on your topic, it's a top priority. Models have different tastes: ChatGPT likes encyclopedic, fact-rich texts without advertising hype, while Yandex actively uses data from its own services — Yandex Business and Dzen.
Speed differs too: Yandex picks up changes in about five days, ChatGPT in about 20, the rest in 20–25 days on average. And AI results are unstable — answers are assembled anew each time — so visibility needs constant monitoring, not a one-off check.
Myths not worth your time
- llms.txt. A robots.txt analogue for AI. Easy to add, but there are no confirmed cases of it helping — neither ours nor on the market.
- Structured data solves everything. It helps machine readability, but without the right text structure and external publications it does nothing on its own.
- SEO is dead, only GEO matters. The most harmful myth — more on it below.
My position: GEO alongside SEO, not instead of it
At a recent meeting a client said: “We're moving the SEO budget to GEO.” That's a mistake. Classic search still brings the bulk of traffic, leads and sales. And without a technically clean, SEO-optimized site, the chances of being cited are lower. GEO brings fewer people, but they arrive “warm”: they've already read about the brand in an AI answer and trust it.
The second mistake is measuring GEO only by website traffic. If an AI recommends your product, the person won't necessarily buy it on your site — they may go to a marketplace or a dealer. So the metric is share of voice — how often the brand appears in AI answers among all competitors — and the effect should be extrapolated to every sales channel, not just your own online store.
Cases from the episode
Which AI models matter in Russia
If a business targets the Russian internet, the priority is Yandex Neuro and Alice: Yandex holds about 70% of Russian search. Next, Igor recommends watching DeepSeek. Other models come second for now.