What GEO is and why you should do it now

Your site is in the Yandex top-3, yet Alice recommends a competitor in its answer. Familiar? A ranking position is no longer a guarantee that the client will even see your brand. A neural network now stands between the person and the site, and it decides whose text to quote and whose to leave on the second page, where almost no one goes.

GEO (Generative Engine Optimization), or optimization for neural search, is the work on content and the site so that generative systems cite you: Yandex Neuro and Alice on YandexGPT, Google AI Overviews, ChatGPT, Perplexity, Gemini, GigaChat. The goal of GEO is not a position among the ten blue links, but getting into the generated answer itself, which the user reads first. A business in Kazakhstan and the Russian-speaking web needs to do this now for one simple reason: in the informational segment of Yandex, AI answers reached up to 68% of the results by the end of 2025, and when they appear the click-through rate of organic results drops by roughly a third. Whoever is not present in the AI answer loses not a position, but the client itself.

What GEO optimization is, in simple terms

GEO answers one question: what do you need to do with your content so that a neural network takes a fragment from it and shows it to the user with a link to you.

The client's path used to look like this: query, list of links, visit to the site. Now a layer of generation has appeared between the query and the site. The user asks Alice or ChatGPT, gets a ready coherent answer assembled from several sources, and often closes the question without going anywhere. In this scheme the site works not as a place people come to, but as a source that gets cited.

GEO is sometimes called AEO (Answer Engine Optimization) or LLMO (LLM Optimization). The essence is the same: you optimize not for a keyword, but for a question and an intent, because a neural network answers questions rather than matching documents to keywords.

How GEO differs from SEO

The difference is not in the tools, but in what you fight for and how you measure success.

The key consequence: in GEO you have less direct control. You do not govern which exact paragraph the neural network pulls out and in what context it shows it. But you do govern how easy your content is to cite. More on that below.

Why GEO needs to be done now, not in a year

Because traffic is already being redistributed, and the count is not in percentages but in multiples.

The picture for Yandex neural search, which matters more for a business in Kazakhstan and Russia than Google:

• According to Yandex itself, 27 to 40% of queries already get an AI answer, and the share is growing.
• In the informational segment the share of AI blocks reached 68% by the end of 2025.
• The share of commercial queries handled by Neuro search doubled over six months, meaning the trend has reached selling queries, not only reference ones.
• When an AI answer is present, organic CTR in Yandex drops by about 34.5%.
• The Yandex Neuro audience is around 80 million users, of which about 12 million come in daily.

The picture for global AI systems, if you work not only for the Russian-speaking web:

• In early 2026 about 68% of Google searches ended without a single click (SparkToro / Similarweb).
• AI Overviews now appear in more than 20% of queries, and when they are present the click-through rate of the first position drops by almost 60%.
• In Google AI Mode, 93% of queries end without a click (Semrush, September 2025).
• ChatGPT handles queries from more than 800 million users a week.

Now the good part, the one all of this is for. AI referral traffic grew by 357% over the year (Similarweb). And the user who does click through to a site from a generative block converts noticeably better: by various estimates such traffic gives a conversion 4.4 times higher than ordinary organic, because the person arrives already warmed up; they read the answer, saw your brand as the source, and clicked deliberately.

The conclusion is simple. Informational-query traffic is shrinking for everyone. But inside this shrunken flow, the winner is whoever the neural network names as the source. This is an early-adopter position, comparable to those who took up SEO in the mid-2000s. In a year these spots will be taken by competitors.

How neural networks choose whom to cite in an AI answer

They take not the prettiest text, but the one easiest to extract and most worthy of trust.

In January 2026 Semrush compared 304,805 URLs from AI answers with 921,614 URLs from regular Google results and identified the factors that correlate with getting into an AI citation. In first place is clarity and the content's ability to be summarized: modular paragraphs, each understandable without context, give about a 33% boost to the chance of being cited.

What this means in practice:

1. The answer up front. The first one or two paragraphs under each heading should give a direct answer: to the point, with figures, no warm-up. The neural network cites what is easy to pull out.
2. One idea per block. Each fragment closes one specific question. Not a wall of text, but separate clear chunks.
3. Structure for the machine. H2-H4 headings, lists, tables (especially comparative ones), highlighting the key points. Schema.org markup (Article, FAQPage, JSON-LD) on the development side.
4. Evidence and E-E-A-T. A real author with credentials, links to authoritative sources, your own cases, figures, studies. AI relies on trust in the source.
5. Named entities. Explicitly name companies, products, people, dates, cities and their properties. Neural networks build the answer around entities, not around abstractions.
6. Freshness. An update date, regular additions of new data to old materials. AI values relevance and more often cites what was updated recently.
7. Content that cannot be retold in full. Original research, cases, calculators, proprietary data. If an answer cannot be boiled down to two lines, the neural network is forced to cite the primary source, and the reader comes to you.

This work, rebuilding the content structure for extraction and laying in the markup, we at 4U handle together with classic SEO, because one does not work without the other.

What a business in Kazakhstan should do right now

Start with diagnostics, not with rewriting the whole site.

• Check where you are. Ask 10-15 key queries for your niche in Yandex with Alice, in Google with AI Overview, in ChatGPT. Record whether you are mentioned, whether there is a link, which competitors are already in the answer. This is your baseline.
• Rethink the metrics. If 60-90% of people close the question in the results, the "clicks from search" metric stops reflecting reality. Add to it the citation frequency in AI answers, the growth of branded queries and the share of voice relative to competitors.
• Rebuild the format, not the volume. Start every piece with a direct answer, break it into self-contained blocks, add tables and FAQs. Volume for the sake of volume is not needed; extractability is.
• Build an evidence base. Real authors, cases with figures, your own analytics, links to primary sources. This is what competitors do not have and what cannot be retold.
• Go beyond the site. Mentions on trusted platforms (VC.ru, Habr, industry media, Dzen, 2GIS, Yandex Business) increase the chance of getting into answers, especially for systems that rely on brand citation. For a local business in Astana and Almaty this is also proof of the company's reality for Yandex.

What to take away

Three decisions worth making this week:

FAQ

Sources

• Yandex: data on the share of queries with an AI answer and the growth of Neuro search usage (2025).
• Semrush: study of citation factors in AI answers, comparison of 304,805 URLs from neural networks with 921,614 URLs from Google (January 2026); AI Mode data (September 2025).
• SparkToro / Similarweb: zero-click and AI referral traffic statistics (2025-2026).
• Search Engine Journal, Search Engine Land: AI Overviews and CTR dynamics (2025).