How AI assistants are changing e-commerce: what brands and marketplaces should rely on

In the spring of 2025, the Russian e-commerce market took an important step into the future: an AI assistant, a chat assistant based on the YandexGPT 5 Pro neural network, has appeared on Yandex.Market. It helps customers find what they need not through a filter search, but through a dialogue: you can describe what you need or formulate the purpose of the purchase — and get a selection with explanations in seconds.

The trend towards the introduction of AI assistants is global. But a full-fledged dialogue with the assistant remains an experiment everywhere. At Amazon, voice purchase via Alexa remains a niche for now: according to Marketplace Pulse reports, only 2-3% of users actively use it for purchases. Chinese platforms like JD.com and Pinduoduo are only testing visual search. Therefore, we can say that Yandex Market is at the forefront among Russian players.

This launch is not just a technological novelty, but an attempt to shift the logic of online search. Previously, a user could endlessly scroll through collections of dozens or hundreds of products. Now he just describes the task — for example, “to pick up a monitor for work and games.” YandexGPT 5 Pro is a powerful model, however, it is not a complete replacement of filters, but an additional interpretation of them.

According to Adventum’s internal tests, only 61% of the collections satisfy the user from the first response. Errors occur due to weak product descriptions, a lack of usage scenarios in the cards, and sometimes “hallucinations” of the model (for example, it may attribute a function to a product that does not exist).

How AI assistants are changing the user experience

The main advantage for the buyer is saving time and fewer unnecessary actions. The assistant takes into account not just the request itself, but also the order history, individual preferences, context, and usage scenario. For example, a user writes: “What should I give a ten-year-old girl?” and receives not a random list of products, but a selection based on age, trends, and past purchases.

But for brands, this is not only a step forward, but also a new competition. AI no longer displays a long feed of dozens of products — it shows 3-5 that, in its “opinion”, correspond to the query as much as possible. And in the voice interface, there is only one. To get on this short list, it is no longer enough just to “be on the site” and do classical optimization. The battle will not be for visibility, but for the right to be chosen.

What is changing in optimization approaches

Classic SEO focuses on keywords and structure. That’s not enough for the assistants anymore. Now it’s important how useful, personalized, and understandable the content is. Previously, it was necessary to enter keywords. Now it is necessary to explain who needs the product and why. For example, you should not write “EVA case”, but “hard case so that the laptop does not crash during the trip”, not “weight 3 kg”, but “you can carry it with one hand and not get tired”.

AI doesn’t just search by words — it analyzes meaning, compares products, and suggests use cases. The content should not answer the question “what is it?”, but “why do I need it?”. This approach requires a deep understanding of the target audience. Here, the focus is not only on the product’s functions, but also on the buyer’s motivation: his pain points, needs, objections, fears and beliefs (PNOFB). Such an analysis will help to adapt the content more accurately to the expectations of different segments. In this sense, AI recommendations are closer not so much to classic SEO as to a well-thought-out marketing hypothesis, where each product card is the answer to the buyer’s real task.

There is also a technical nuance. The AI on the marketplace does not “see” the photo like a human. It analyzes the description, image metadata, behavioral signals (card viewing time, adding to cart, refunds), and user responses in Q&A. If the description is laconic, the photo has no alt tags, and there are few reviews, then the product will not be included in the recommendation, even if it has perfect characteristics.

How to adapt product cards for AI

There is no analysis on the issue yet, but it is already clear that the output may be affected by:

Completeness of the data. Specify everything. The more distinct the characteristics, the higher the chances.
Context and scenario. The card should answer the question: why and to whom?
User-defined language. The AI is guided by conversational queries. Therefore, “holds A4” is better than “size — 30×21 cm”.
Visual. Not photos only, but also videos, 3D, infographics. Even if the AI doesn’t “look” at the image directly, the user is watching, and his behavior affects the training data.
Reviews and ratings. The assistant can “learn” from the negative. In case of repeated complaints, it is highly likely that he will not show the product at all.

Where are the risks and how to deal with them

The platform does not explain the logic. There is no transparency yet: it is unclear why one product is in the top and the other has disappeared.
“Medium range”. Algorithms tend to make an “average” choice: non-standard or niche products may fall out of the search results.
The AI may confuse the positioning. For example, a premium product will be called “budget” because of the simplified description.

And something else: AI has not replaced trust yet. Research shows that 40% of users are annoyed by the lack of a live consultant. Therefore, the brand strategy today is not only about visibility, but about the balance between automation and humanity.

What should companies do to get into the AI recommendations?

Learn to experiment and adapt quickly. At Adventum, we recommend that you allocate at least 10% of your marketing resources to testing new tools, formats, and channels. These “exploration percentages” are investments in resilience to change and a chance to catch the wave before others. And right now we can start:

Enter standard queries and check if you are in the search results.
Improve the cards. Specify not just the characteristics, but the context, emotions, scenarios, photos, and videos.
Publish FAQ, instructions, and Q&A — useful content. Assistants search for answers in open sources and can “extract” the answer from these blocks.
Work with consumer behavior. Cards with a high viewing depth and a good conversion rate are more likely to be included in recommendations.
Follow the initiatives of the priority partners. Perhaps there will be analogues of Amazon’s Choice in Russia.

Where are the marketplaces going?

AI assistants are becoming part of the user’s journey, and it is already clear which directions will be decisive for the platforms themselves:

AI integration across the entire funnel. From search and recommendations to payment and post-service. The more stages an assistant accompanies, the higher the retention and loyalty.
Quality control and transparency. It is important to let brands understand why the product was not included in the recommendations. Mistakes, distortions, and opaque logic can undermine trust.
Localization of AI. Customization for language, mentality, culture, interfaces is a strategic resource, the key to audience trust.
Preservation of humanity. In niches like electronics, furniture, cosmetics, AI is not yet a substitute for humans. Live contact will remain necessary, especially for expensive and complex purchases.

What’s next: the 2030 horizon

Gartner predicted that AI would automate 85% of customer interactions in retail by the mid-2020s. In 2018, the figure seemed too ambitious. It’s the same now, but it’s gradually becoming plausible. Customer scenarios are changing, and today people, especially young people, start the day not with a search, but with a neural network question. By 2030, this is likely to become the norm. And for businesses, now is the period of initial adaptation of product cards to the new search logic and the waiting period for transparent conditions and analytics to be included in the shortlist of the issue from the AI assistant.