The seller's AI glossary for 2026: 25 terms that decide whether the marketplace shows you or your competitor

If you sell on Wildberries, Ozon, Yandex Market, or Megamarket in 2026, you're no longer "running a product listing." You're managing an AI interface to your brand. What brings a buyer to you isn't a banner or intuition — it's an algorithm: the marketplace's internal search, the recommendation feed, and — more and more — an AI assistant that a person simply tells, "find me one."

The problem is that half of all sellers play this game blind. They "guess" where a competitor knows the lever and pulls it.

This glossary isn't theory. It's a checklist. Go through every term and answer honestly: do I have this — or am I hoping to get lucky?

---

Layer 1. Marketplace visibility — how people find you at all

This is where the money lives. If the listing isn't in the results, nothing else matters.

1. Ranking (search results) — how WB or Ozon decides which position to show your listing for a query. This is the main "AI judge" in Russia, not ChatGPT. Everything else in this block is a signal for it.

2. Listing SEO — the title, description, and keywords tuned for the marketplace's internal search. Not for Google — for WB and Ozon, which have their own logic.

3. Semantic search — the algorithm understands intent ("a warm winter jacket for the city"), not just an exact word match. Ozon and WB already search this way. That means your listing needs not only keywords but also the context of use.

4. Relevance — how well the listing matches the query by text AND by attributes at the same time. Empty attributes = low relevance, even with a perfect description.

5. Listing CTR — the share of clicks from search results. The algorithm lifts what gets clicked and sinks what gets ignored. Your main weapons here are the first photo and the price.

6. Conversion to order (CR) — the path from click → cart → purchase. The strongest ranking signal: it pays the marketplace to show what people actually buy.

7. Buyout rate — a critical factor for WB. A high return rate sinks a listing even if sales are high. Size charts, honest photos, and accurate descriptions feed ranking directly.

8. Ad share of revenue (ACoS) and boost — a paid visibility lever on top of organic. It works, but without strong organics it just burns budget.

---

Layer 2. Listing content — what the algorithm and the human both read

Visibility brought the buyer. Now the listing has 3 seconds to convince both of them — the human and the algorithm.

9. Attributes (characteristics) — the Russian equivalent of schema markup. The more completely they're filled in, the better the listing lands in filters and in AI recommendations. This is the most underrated and cheapest lever: everyone ignores it, and it's free.

10. Rich content and infographics — a structured listing that both the algorithm and a human-on-the-go can "read." Text no one finishes loses to an image that answers the question instantly.

11. Catalog enrichment — rewriting product data so that search and recommendations understand it. Not "pretty for a human," but "readable by machine and human at once."

12. Reviews and questions (UGC) — an independent trust signal. Both the algorithm and the buyer trust them more than your own description. Working with reviews isn't a "nice to have" — it's a ranking factor.

13. Content rating — the marketplace's own score of how well the listing is filled out (Ozon has one). A direct hint from the platform: here's what you're missing. A free roadmap almost nobody opens.

---

Layer 3. Internal AI assistants — new traffic inside the marketplace

14. AI visibility — how often the marketplace's built-in assistant, Alice, or GigaChat mentions your product in a recommendation.

15. Recommendation feed — "bought with this item," "similar products." The main source of extra sales inside the marketplace, and it's almost entirely on the algorithm's side.

16. Agentic commerce — the AI picks and places the order on the buyer's behalf. In Russia it's only emerging, through marketplace assistants, but the direction is set: fewer human clicks, more algorithmic decisions. In detail below.

17. Zero-click purchase — an order placed by voice or in chat, without ever opening the listing. The buyer doesn't see you — only the AI does. Which means it isn't the listing design that wins, it's the data inside it. In detail below.

18. AI Share of Voice — how often the AI names you, and not a competitor, in its answer to a query. The new market-share metric: not a shelf in the store, but a line in the answer. In detail below.

---

Layer 4. External AI assistants — GEO, Russian style

Yes, technically ChatGPT, Perplexity, and Gemini are blocked in Russia. But a large and — more importantly — solvent audience uses them through a VPN, and it really does ask AI "what to buy" and "which is better." You can't ignore this channel. But don't overrate it either: for a Russian seller this is a supplementary layer, and it works on a fundamentally different principle.

19. GEO (Generative Engine Optimization) — getting your brand into the answers of ChatGPT, Perplexity, Gemini, and DeepSeek. For the Russian version, add Alice and YandexGPT here too. In detail below.

20. AEO (Answer Engine Optimization) — content built so the AI serves you up as the ready answer ("the best thermo pot under 5,000 ₽"), not as one of ten options. In detail below.

21. Citation frequency — how often the AI cites your site, review, or listing as a source.

22. Entity strength (brand as an entity) — whether the AI recognizes your brand as a real, trusted entity. What works: reviews on external platforms, mentions in the media and blogs, and your rating on the marketplace itself.

23. GEO audit — ask ChatGPT, Perplexity, and Alice "how do I buy [your product type]" without naming your brand and see who they name. That's your list of competitors as seen by the AI.

The nuance that changes everything

External AI barely sees the marketplace internals. ChatGPT doesn't crawl WB's search results in real time. It knows your brand from what lives outside the marketplace: your site, reviews, ratings on external platforms, media mentions, recognition.

That's why the GEO lever for a seller isn't the listing — it's the brand's external digital footprint. And here a neat circle closes:

Inside the marketplace you optimize the listing. Outside, the brand as an entity. These are two different jobs, and whoever does both wins.

---

Layer 5. Data and economics — without them, everything else runs at a loss

24. Unit economics — your margin after commission, logistics, storage, and ads. Any visibility without it is just a paid path into the red.

25. Listing funnel — impressions → clicks → cart → buyout. Find the step where you lose the most, and fix that one — not the one that's more pleasant to fix.

---

How to use this glossary

Don't reread it. Apply it in a single evening:

1. Run every one of the 25 terms as a "yes / no" against your listings. Each "no" is a lever you're not pulling.
2. Visibility first: ranking, attributes, CTR, buyout rate. This makes money the fastest.
3. Then content: catalog enrichment, rich content, working with reviews.
4. Then data: unit economics and the funnel — so growth is a profit, not turnover for turnover's sake.
5. The external brand last: GEO, citation frequency, recognition. It's a long game, but it's exactly what feeds organics.

A 1-hour audit

Open WB and Ozon search. Type the query people should find you by — without your brand name. Look at the top 10: what photos, price, and rating do competitors have, and how are their attributes filled in? Then ask Alice and ChatGPT the same question.

Everything you see there that you don't have — that's your roadmap for the next 90 days.

---

Six terms in detail

Five of them are named above in a single line — here they are taken apart down to the lever: how it works, what the algorithm sees at that moment, and what to do about it. The sixth, Yandex Commerce Protocol, isn't on the checklist: it isn't a term but a protocol, launched in February 2026, that gave agentic commerce in Russia its first working channel.

16. Agentic commerce

Agentic commerce is when the AI itself picks the product and places the order on the buyer's behalf — not just advises. A person tells the assistant "order me a cheaper thermo pot, but one with good reviews," and the algorithm makes the decision and completes the purchase. For a seller this changes who you're selling to: the listing used to be chosen by a human's eyes, now the product is increasingly chosen by AI from data.

In Russia this is no longer "tomorrow." Yandex has the YCP protocol and thousands of stores selling through Alice; Wildberries launched AI-driven selection in June 2026. The logic is the same: the buyer states an intent instead of scrolling the results, and the assistant matches the query to products by attributes, price, reviews, and rating. When the algorithm decides, it isn't a pretty listing that wins — it's the data inside it. What a human would have "felt out" from a photo, the algorithm takes only from structured fields. A request to "put together a cart for the dacha under 10,000 ₽" walks straight past a listing with empty characteristics, even when the product is a perfect fit.

It differs from the recommendation feed in who presses the button: the feed advises and the human decides — agentic commerce carries the order through itself.

17. Zero-click

A zero-click purchase is an order placed by voice or in chat, without ever opening the listing. The buyer says "order the same as last time" or "add laundry detergent," the order goes through, and the listing is never opened. They don't see you — only the AI does.

In a normal purchase the person scrolls the results, looks at photos, reads the description, and every visual lever does its job. In a zero-click purchase that stage is simply gone. The first photo, the infographics, the well-written text play no part at that moment — only machine-readable fields do. Empty characteristics mean invisibility.

It hits repeat orders hardest, and repeat orders are the bulk of the channel. Once a month a buyer asks Alice to "order cat food, the same one." If the product was chosen once and its data is honest and complete, you land in those orders automatically; if the data is weak, the assistant swaps you for a competitor with a fuller listing on the next request. Rating does the work photos used to do — it is the only thing the assistant shows in place of an image, which makes unanswered negative reviews more expensive here than anywhere else.

18. AI Share of Voice

AI Share of Voice is the metric of how often the AI names you, and not a competitor, in its answer to a buyer's query. Market share stopped being a shelf in a store and became a line in an assistant's answer.

You measure it by running queries, not by exporting a report. Give Alice and the platform's assistant ten typical category queries without naming your brand and count how many name you. Five out of ten is a 50% share against competitors in that sample. A quarter later you repeat the measurement. There's no universal benchmark: what matters isn't the absolute number but the growth of your share from one measurement to the next.

The metric is often confused with citation frequency (term 21). Share of Voice is how often you are named in the answer; citation frequency is how often you are cited as a source. They're related but grow from different things: the first from brand strength and listing completeness, the second from whether you have anything worth citing at all.

19. GEO (Generative Engine Optimization)

GEO is optimizing your brand and product for the answers of generative assistants: ChatGPT, Perplexity, Gemini, DeepSeek — and, in the Russian version, Alice and YandexGPT. Classic SEO fights for a spot in a list of links; GEO fights to make the AI name you specifically in a ready-made answer to "what should I buy."

The assistant doesn't show ten links — it gives one answer and names a few brands inside it. Getting into that short list is the whole job. The lever isn't the obvious one: external AI barely sees the marketplace internals. ChatGPT doesn't crawl WB's search results in real time; it knows your brand from what lives outside the platform — your site, reviews, ratings on third-party platforms, media and blog mentions, recognition. So the work is on the external digital footprint, not on the listing.

A buyer on a VPN asks ChatGPT "which thermo pot should I buy under 5,000 ₽." If the brand has reviews and mentions, the AI may name you; if there's nothing outside the marketplace, it names a competitor who has that footprint — even when your WB listing is better.

Technically ChatGPT, Perplexity, and Gemini are blocked in Russia, and the channel shouldn't be overrated: for a Russian seller GEO is a supplementary, slow-to-pay-off layer, not a replacement for working on the listing. But the VPN audience is large and solvent, and it really does ask AI what to buy.

20. AEO (Answer Engine Optimization)

AEO is building content so the AI serves you up as the ready answer, not as one of ten options. GEO is about getting into the answer at all; AEO is about becoming the answer itself.

A generative assistant loves structure: a clear question, a clear answer. Content built for AEO answers a specific query directly and in the first lines — a definition, a short takeaway, a fact. A phrasing like that is easy to lift whole. Instead of a "Thermo pots — big selection" page you build a piece that answers the buyer's question directly: "Which thermo pot to choose under 5,000 ₽," with the takeaway in the first paragraph, specs, and a short comparison.

You check it in one move: ask the assistant your buyer's exact query and see whether you're named as the ready answer, as one of the options, or not at all.

Yandex Commerce Protocol (YCP)

YCP is Russia's first standard by which an online store connects to AI and starts selling directly in a chat with Alice and in Yandex Search. It isn't a checklist term but a channel: a new storefront where, instead of a shelf and a listing, there's a line in the answer of an assistant that picks the product itself and walks the buyer to the order. About 19 million people use Alice AI in chat every week, and some of them ask not "show me" but "buy."

Technically YCP is a way to hand Yandex the structure of your catalog: products, prices, stock, attributes. The protocol was announced on February 27, 2026; at launch about 3,500 stores could already sell through Alice and Search. There are four ways in: automatically, if the store runs on Yandex KIT; with a ready-made module for 1C-Bitrix; through the Yandex Tovary account with fulfillment via Market; or over the API for a custom CMS. Auto-connection and the module cover small stores too — this isn't a big-players-only story.

It differs from connecting to Market in kind: Market is a platform with search results and listings, while YCP is a standard by which the product is sold inside Alice's answer, where the buyer may never open a listing. From there the rules of agentic commerce apply: completeness and honesty of data decide.

Whether to connect now is a question about comparing channels. YCP covers what external AI can't give: those work in Russia only over VPN and barely see your internals, while Alice is a domestic channel that is already live. Yandex forecasts global agentic sales at $3–5 trillion by 2030; in Russia the layer is only forming, and early connections are cheaper and more visible.

---

In 2026 the winner isn't the one with the most inventory. The winner is the one the algorithm understood and chose — on the marketplace shelf and in the AI's answer.