AI and SEO: a guide for Ukrainian retail

Google is no longer the only entry point to a store website. Some buyers now receive a ready answer before they even click through: in the «AI Overview» block, in AI Mode, or in a chat with ChatGPT. Below is an explanation of what this does to the traffic of a Ukrainian store and which actions produce results.

How buyers search now

Both Google AI surfaces work in Ukraine. «AI Overview» shows a short answer above the search results. AI Mode conducts a dialogue with the user and supports the Ukrainian language. The AI Mode audience has exceeded one billion users per month, and the number of queries in it doubles every quarter.

Alongside search, a separate channel has appeared: chat assistants. In most commercial niches in Ukraine, ChatGPT accounts for more than 90% of all AI traffic. For Rozetka, this is about 93%. Gemini is in second place by popularity. Perplexity is used less often, usually for detailed technical comparisons.

The form of the query has changed. Previously, the buyer entered two or three words and browsed the results manually. Now they describe the situation in one sentence and expect a ready selection. In chat, it looks like this: «pick winter boots for a wide foot, up to UAH 3,000, available in Kyiv and with Nova Poshta delivery».

What happens to organic traffic

When an AI answer appears above the results, users click through to websites less often. Several independent studies have shown this. The numbers differ because the methodologies differ, but the direction is the same.

The last row is important. After a year and a half of decline, the indicator began to recover. The gap still remained: queries with an AI overview generate roughly one third fewer clicks. This is the difference worth accounting for in plans.

Reach estimates differ. Depending on the methodology and niche, researchers see AI overviews in between one quarter and one half of all queries. The average market number is not very useful. What matters is the distribution across your own semantics.

Which queries lose clicks the fastest

• Comparisons in the format «X or Y». An AI overview appears almost always.
• Questions with the words «how», «why», «what is better». Coverage is above 85%.
• Long queries of seven words or more. Previously, they were considered safe because of weak competition. Now AI overviews trigger for them more often than for short ones.
• Local questions with the word «nearby». Coverage is about 77%.

Practical conclusion for the content plan. Articles such as «how to choose winter tires» will bring fewer visits. Category pages and product cards, where people go with the intent to buy, hold up better. Therefore, reallocate time: first bring commercial pages into order, then informational ones.

How much AI traffic there really is in Ukrainian retail

The absolute numbers are still small. The share of AI visits in the total traffic of Ukrainian platforms mostly remains below 1%. In real estate, it is below 0.3%. The highest indicators among the researched players are Liki24 at 0.619% and Kasta at 0.444%. At the same time, the channel is growing, and the peak of queries falls in October and November, when buyers prepare for Black Friday.

In the United States, the market is about one or two years ahead of the Ukrainian market, and the picture there is different. According to Adobe, AI traffic to retail websites grew by 393% year over year. These visitors convert 42% better than other channels, generate 37% more revenue per visit, and spend 48% more time on the website. Previously, this same traffic converted 38% worse than average.

The reason is simple. A person comes from chat after already comparing options and making a decision. All that remains is to place the order. That is why each such visit is worth more than a regular one.

How to rebuild work on the website

Step 1. Divide queries into four groups

Export queries from Search Console for the last six months and sort them into groups. Each group then requires its own approach.

1. Commercial queries without an AI overview. Categories, brands, models, «buy». Everything here works as before and brings the main revenue. Do regular SEO: titles, category descriptions, filters, speed, product cards.
2. Informational queries with an AI overview. There will be fewer clicks. The goal of such pages changes: they must make AI name your store in the answer. Write a short direct answer at the beginning of the page, then the details.
3. Local queries. Tied to a city and district. Here, product availability and pickup speed matter. The amount of text on the page has little influence.
4. Brand queries. Count them separately. If the number of searches for your name grows, mentions in AI answers work even when direct visits have not increased.

Step 2. Add several metrics to the report

The number of visits remains a working metric for commercial pages. For the rest, add four more:

• how many times AI names your store from your list of 20 questions;
• who is named instead of you and how often;
• the dynamics of brand search in Search Console;
• conversion and average order value separately for the AI Assistant channel.

Check the list of questions manually once a quarter, on the same day of the month. Half an hour of work gives a picture that shows movement.

Step 3. Provide what competitors do not have

Google named content that cannot be taken from another website as the main factor of long-term presence in AI answers. For a store, these are specific things:

• real stock balances for each retail outlet;
• current prices and delivery times to a specific city;
• your own product photos taken in your store;
• warranty, exchange, and return terms in plain language;
• answers to questions about compatibility, sizes, and package contents.

AI can take a description copied from the manufacturer’s website from any other store. Your experience with the category cannot be taken anywhere else.

GEO: what affects whether a brand appears in the answer

Google published an official guide to optimization for generative search features. The document closed several disputes and removed from the agenda what agencies were actively selling.

Google directly called GEO and AEO part of regular SEO for its own surfaces. The working list is short: the site is open for crawling, pages are fast, the structure is clear, the content has unique value, and contact details are up to date. The money requested for llms.txt and special markup is better spent on product photos and website speed.

Why the top 10 no longer guarantees a mention

According to Ahrefs, 83% of links in AI overviews lead to pages outside the organic top 10. Among the pages most frequently cited by ChatGPT, 28.3% have no organic visibility in Google at all. The reason is that different systems build answers from different sources.

Another pattern from the Ukrainian study: AI is more willing to name stores with a transparent catalog and clear warranty and return terms. Platforms that chatbots almost never mention are usually poorly represented on authoritative third-party resources. Therefore, mentions in reviews, selections, industry directories, and customer reviews work. One publication on a prominent resource gives more than ten pages of text on your own website.

Local store and «nearby» queries

For a store with physical locations, local results matter the most. AI overviews appear in approximately 68% of local queries. A study of the restaurant niche showed that businesses from the top three local results are mentioned in AI answers in 25.9% of cases. In other words, map positions directly influence whom AI names.

The logic here is familiar. The AI answer for local search is built from the same signals as the map. The list of actions is short, and each item can be completed by the store itself.

The last point has direct monetary value. The study shows: if the query contains urgency, for example «delivery tomorrow» in a specific city, AI ranks local sellers with real stock data higher. A store with two locations can outrank a large marketplace precisely on such queries.

What to delegate to AI and what to do yourself

The division has become fairly clear. The model handles repetitive work with clear input data well. Decisions that create risk for the store remain with the human.

The risks of publishing without review are well described. The model invents characteristics that the product does not have. Competitors’ texts made with similar prompts turn out almost identical. Mass publication of unverified pages falls under Google’s policy on low-quality content. In health-related niches, the restrictions are even stricter: after rule updates, chatbots avoid direct advice on medicines, and traffic declined even for large pharmacy aggregators.

90-day plan

Frequently asked questions

The figures are taken from the studies listed in the sources. The indicators of an individual store depend on the niche, region, and query structure, so the basis for decisions remains your own Search Console and GA4 data.