Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article3 min readViews0

AI Catalog Search: What an Online Store Can Learn from the New Apteka ru Assistant

Yandex Cloud shared details about an assistant for searching products in a large catalog. I analyze what data and constraints are needed for a similar solution in a standard online store.

A magnifying glass highlights a suitable item among neatly arranged products
In this article

Buyers rarely phrase queries the same way suppliers name products. They might search for "a mount for a narrow shelf" or "a lamp for evening work," while the catalog expects an article number. Sales are lost at this gap, even with a good assortment.

On September 21, 2026, Yandex Cloud announced an AI assistant for the Apteka ru service. The solution relies on the platform's catalog and proprietary content; the database includes approximately 100,000 products and over 6,000 articles. This demonstrates handling a large assortment. Here, I will examine the technical takeaways for general retail, without providing any medical advice.

Start with card quality

If two products share the same short description and lack specifications, models have no source for exact differentiation. An eloquent dialogue cannot recover unknown size, compatibility, or kit composition.

For a hardware store, material, dimensions, and intended use are critical. For electronics, interfaces, power requirements, and compatibility matter. First, standardize key attributes and verify their completeness on popular items. This benefits both traditional filters and AI search.

The response must be based on the current assortment.

A model may know that a certain product exists in the world. However, this does not mean the store sells it. Therefore, search must be tied to a specific catalog and stable card identifiers.

Prices and stock levels require separate freshness checks. An old description is acceptable for an immutable attribute, but an outdated price already alters the buyer's decision. Before checkout, significant conditions must be confirmed by the store's active mechanism.

When a clarifying question is needed

The query "find a charger" is too broad. A useful assistant will clarify the device and requirements rather than selecting the first visually similar item. For business, it is important to define such decision points in advance.

In some categories, a compatibility error leads to returns or equipment damage. In such cases, it is better to honestly request additional specifications or pass the question to a specialist. A confident answer without sufficient data does not make the service more convenient.

What to include in the validation sample

Take real anonymized search queries that currently result in empty results. Add colloquial names, typos, incompatible requirements, and out-of-stock items. A correct result is sometimes a refusal to recommend rather than a found card.

Evaluate not only the visual appeal of the response. Ensure the suggested product is correct, free of fabricated attributes, and allows the user to proceed to a clear next action. It is useful to compare AI search with the current search using identical queries.

How to launch without major store rework

A pilot can use a single well-described category. The assistant can initially only explain the selection and show suitable items, leaving payment and ordering to the standard interface. This reduces the number of new failure points.

The effect should be evaluated based on successful search sessions, support inquiries, and returns due to incorrect selection. Then AI becomes a way to help customers navigate the assortment, rather than just another chat window on top of unresolved catalog issues.

Discussion0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.