Ozon search and SEO: titles, keywords and how a card gets indexed

A card shows up in Ozon search on two counts: which field holds the word the buyer typed, and how closely the card matches the query. All of this comes from seller videos; it is not an Ozon document. Limits and weights belong to each video's period, and the sellers contradict each other on several points, which we flag.

How search works

The most detailed account comes from a marketplace marketer. First, candidate gathering: the query's words are sought across the card's fields and attributes — name, category, brand, description, colour, size, material, type — and the pool for one query can reach 500,000 products, a figure a second source repeats. Then base ranking: a formula counts each word's hits across those fields, a 0-25% multiplier is added for the rating, and the 2,000 most relevant are kept. The rest is machine learning plus preset boosting and debusting coefficients; at the end of the 2,000, the next 2,000 load.

On the text's share, two sources land close: one, citing the ranking formula Ozon publishes, says query relevance is given up to 40%; the other puts textual relevance at 20-40%.

Before ranking, the query is normalised. One trainer names three rules: synonyms — "телефон" also pulls in cards saying "смартфон", which rank lower for missing the exact word; form normalisation — "подушки" reduces to "подушка", so writing out every inflection is pointless; exact matching — a card catching the tail of a long query earns a boosting coefficient.

The title: the most argued-over field

One source's formula for a hand-written title: product type + brand + key characteristics, starting with the highest-volume word that is still on target. Volume alone is the wrong sort order: "витамины" has the higher frequency, yet the name "омега-3" goes first.

On length the sources disagree: one puts the ceiling at 200 characters, another at "around 230 or 240". Both say length is not the goal, and the first explains why: roughly the first 70-75 characters show on desktop, about 45 on mobile. If the buyer cannot see the spec they came for, in the photo or that line, they scroll past — hence splitting the title into readable blocks with punctuation (semicolon, comma, slash, all permitted). On Latin spelling: transliterated words are rejected in moderation outside certain categories, and Ozon translates the word itself anyway. The same video checks it: "омега-3" in Cyrillic runs to roughly 24,000 against roughly 3,500 in Latin.

Which field the keyword counts in

Which fields are indexed is where the videos disagree most.

Field	What is claimed
Product name (title)	All four agree: the first-priority field, and the strongest word belongs here.
Characteristics	Everyone says fill every field that can be filled; one recommends 95%+ attribute fill depth. The right product type matters most: it puts the card into filters and extra categories.
Description	Three say it is indexed, one adding that it comes second in priority. One trainer says the opposite: not indexed for organic search, it only feeds the ad tools.
Rich content	The split is sharper. One says "wherever you put the keys, the marketplace picks them up"; two say it is not indexed.
Keywords field	The most contested of all; its own section below.
Reviews and questions	The one source that addresses them says they are not indexed at all.

Two recommendations on description length: one calls 1,000-1,300 characters optimal and says repeating title words gains nothing, another orders "a 2,000-character description" from an AI model. Both insist the words sit naturally in sentences; two sources note the algorithms dislike keywords strung together with commas. Bolding keywords, once a trick, now counts as spam.

Finding the queries, and measuring whether they index

Free side: Yandex Wordstat for synonyms (its volumes differ from Ozon's), competitors' cards, Ozon's search suggestions, an AI-generated keyword list, and the search-query analytics in the seller panel. One seller works from Ozon Data with two caveats: the frequency there covers the last 7 days, and the numbers are Ozon's own — another service may show a figure four times larger. Paid side: MPStats for volume, SalesFinder for queries and position tracking, XWAY for pulling clusters out of competitors' cards, SellMonitor for the demand-to-supply ratio. Cluster logic: show up for one phrase in a cluster and you show up for all of it.

The indexing checks sellers name:

• "Products in search" in the panel. Type the query, sort by match, and the cards Ozon considers most relevant come up; one seller reads his own place and those cards' titles.
• The "queries for my product" report. The same seller runs a new card for a week, then checks which words it actually draws traffic from and clears the rest.
• Impressions and average position. An agency owner watches both after new words go in, shifting to mid and low frequency if the high-frequency positions are poor.
• Measuring old and new sets side by side. One seller sets up a position-tracking project per key set in SalesFinder and compares after a week.
• Splitting organic from paid. Another separates the two: if advertising brings 90-95% of sales, organic is not working.

Myth or verified

The keywords field. Four sources say four different things. One asked support directly: the field is not ranked, it exists only for advertising — but the ad logic predates the field, so it does nothing in practice; he fills it anyway, in case the rules change. A second says it barely affects organic search. A third takes it seriously: irrelevant queries there cut conversion and drag overall position down. A fourth notes that Ozon has never commented on the field: everything said about it is guesswork. On format two sources agree: semicolon and a space; neither a comma nor running the words together works.

Repeating the same word. One trainer says Ozon sums the hits, so the more places a relevant word appears the higher the card ranks for it — the exact inverse of the Wildberries rule. An agency owner takes the opposite view: over-stuffing leads to pessimisation, the common complaint of sellers who tell him they improved their SEO and lost sales. The one shared point is not to repeat an irrelevant word.

Holiday keywords. One seller disproves these on screen: a card listing 8 March, 14 February and New Year in its description shows not one holiday-related query in its report, while the top result for "gift for 14 February" is a Makita drill that mentions no holiday at all. The real damage is elsewhere: irrelevant traffic looks and leaves, conversion falls, and since clicks are paid for, the ad cost inflates.

Constantly reworking the SEO. A seller of four years argues against it: each pass through moderation with new keys recalculates the weight and changes the card's composite score. His routine: build it, run it a week, clean it from the report, leave it alone. The seller above tests key sets deliberately instead — one pricing the cost of change, the other the value of learning.

The sharpest objection. One seller says SEO's influence is overstated: sales held up on his own product with one sentence of description, keyword swaps with his students never produced a result, and he puts the description's share of total influence at no more than 1%; advertising decides position, and SEO is not even in his top five factors. The seller on the other side climbed from position 220 to third in a year — but his list does not rest on words alone: a 100/100 content rating, 10 reviews minimum, a high product rating, FBO delivery speed and stock in the buyer's region, conversion lifted from 8-9% to 18.6%, and advertising itself. Switch the ads off, he adds, and organic traffic falls too.

In short

• Text is not the whole of ranking: two sources put query relevance at 20-40%.
• The product name is the field everyone agrees on; only ~70-75 characters show on desktop, ~45 on mobile.
• Filling the characteristics is cheap ground; one source recommends 95% fill depth.
• Whether the description and rich content are indexed is openly disputed.
• Ozon has never commented on the keywords field; opinions run from useless to important.
• Irrelevant words do measurable harm: conversion drops, click spend is wasted.
• Measure indexing from the panel: query report, impressions, position, organic-versus-paid split.

Commbot keeps this field separation in mind when generating card texts and attributes in bulk.