VK Ranks Products by Purchase Intent: What Sellers Should Do

VK feed ranks products by purchase intent: how the track works, a year of Ozon shops in numbers, and what sellers should do.

What happened: two announcements in a week

Short version: on September 29 the VK AI team described a separate product-recommendation track inside the feed, and on October 6 VK summed up a year of Ozon shops - 9.5 million redeemed orders. The feed no longer just shows posts, it predicts purchases.

We watch VK as a lead channel every day, and this is the first time the platform openly said what it optimizes for. Here is what it changes for community sellers.

How the product track works

Short version: a separate model scores the probability of an order, not likes. Orders rank highest, then partner clicks, likes and views far below.

A person can like sneakers and never buy them, so product posts were moved into their own domain with separate candidates, logging and ranking model. Irrelevant shows and negative reactions get penalized.

The interesting part is the cold start problem. Orders are rare, so the model had little to learn from. VK raised logging of the product domain from 2% to 100%, growing the training sample 15 times. While orders were scarce, the model learned on frequent signals: expanding post text and positive reactions.

The platform cites its own numbers for March-August 2026: show-to-order conversion up 53%, orders up 43.2 times, partner clicks up 55.3 times, product post shows up 28.3 times. Ozon was the first partner. Read these as platform data about its own track, not a promise for your community.

What it changes for sellers

Short version: reach stops being the headline metric. Watch cost per lead and the share of leads that reach payment.

A post with ten times less reach but real orders beats a viral post with none. By early September product publications hit roughly 5 thousand a day - competition for product slots grows, better to enter while the window is open.

We covered the reach shift across platforms in VK vs Telegram for B2B - with daily activity numbers for both.

On numbers: if you do manual lead search across communities, check 24GO free tools - the group analyzer and ROI calculator save a couple of manual hours. Now - the cards.

Product cards the feed understands

Short version: one post, one product. Price, availability and terms in the text, the product itself in the photo. First lines name the product right away.

VK calls text expansion one of the starter signals, so the first lines decide: product plus a reason to keep reading. Checklist before publishing:

One post, one product - not a ten-item showcase.
Price, availability, terms in the text, not only on the image.
The product itself in the photo, not an abstract banner.
One clear next step: message the community, open the card, place the order.

Track lead sources

Short version: every entry point gets its own label, record the source at the moment of contact.

Posts, ads, community messages, product cards are four different sources - never mix them into one pile. Otherwise you will never learn what brought the order: feed, blogger or targeting. We showed how to merge site, phone and messengers into one window using Telegram leads to Bitrix24 - same mechanics: source recorded at contact time, not recalled at month end.

Blogger shops via AdBlogger

Short version: products go inside blogger posts with pay per redeemed order. One year produced 700 thousand shops and 45 billion views.

Sellers create tasks in the Ozon cabinet, products appear in the VK AdBlogger catalog, authors place them into posts and earn per purchased unit. Home goods, household chemistry, cosmetics and pet products sold best. The average post delivered about 13 orders last year - a working benchmark for a first test.

Catching demand with monitoring

Short version: while everyone watches the feed, demand is also visible in community discussions. We read them automatically.

The feed decides what to show. But people write in communities every day in plain text: asking prices, availability, terms, complaining about competitors. That is purchase intent in pure form, independent of ranking algorithms.

At 24GO we monitor such communities around the clock: an intent message reaches a manager within a minute, tagged with its source. Setting up monitoring from scratch is covered in our VKontakte parsing setup.

Mistakes: 5 ways sellers burn budget

Short version: almost everyone loses money in the same five spots.

Mistake	Why it hurts
Ten-product showcase post	The model cannot tell what to rank
Price only on the image	Text without price gives no signal
No next step	The lead has nowhere to go
Leads without sources	Impossible to know what to scale
Betting on reach alone	Reach does not equal orders

FAQ

How do I get into the product track? VK disclosed no entry terms for small business. Practical move: fix your cards and compare Market vs post leads over a month. Do shops work for B2B? The mechanics fit basket goods. For services use: offer post + message lead + source tracking. The feed keeps changing - what to do? The mechanics belong to the platform, weights change without asking. So build lead tracking and discussion monitoring - assets independent of any algorithm.

Sources

The full VK commerce picture in one go: try 24GO free tools - 3-day trial, then decide for yourself.