Neural answers: ads.

What happened

In August 2026 Kommersant and Sostav published the first figures on advertising inside Yandex neural answers. Neural results are a block in which the search engine, instead of showing ten links, writes the answer to the query itself. According to the agency Digital Budget, in the second quarter of 2026 the share of neural results in ad impressions of the ten largest financial advertisers grew from 17% to 30%. This is a share of ad impressions, not a share of queries that Yandex answers with a neural network: retellings often confuse the two.

Advertising appeared inside this block, and it was distributed unevenly: 57% of impressions went to Sber, 27% to Yandex’s own services, 9% to T-Bank. All the other advertisers in the country were left with less than 7%. Three companies control more than 93% of impressions in the fastest-growing search format.

This is not a conspiracy or a closed club. Ads for the neural answer are selected by the ordinary Yandex Direct auction: what is shown is what is relevant to the query and wins on bid. The big players win with broad keyword coverage and budgets that let them cover hundreds of thousands of queries a day. That is why the block is home to banking products, not dental clinics and car services.

Why it matters right now

Earlier we looked at the zero click as a problem of the future: search answers by itself, links slide off the edge of the screen. Now that problem has numbers. According to an Impuls.Guru study published in June 2026, 90.2% of users use neural networks in one way or another, 42.8% start a complex search with them, and 58.9% use AI when choosing goods and services. Half - 50.9% - do not double-check the answer in any other source.

At the same time Yandex’s own share of the Russian search market has grown to 73.34%. That is, the platform that increasingly answers instead of websites has simultaneously become even bigger. For a business that gets leads from search this is a shift in two directions at once: part of the organic traffic is eaten by the neural answer, and the ad block inside it is occupied by other people’s money.

The estimated effect is modest so far: PR Partner expects this kind of promotion to add 1-5% to banks’ search traffic. For a bank with millions of queries that is thousands of conversions, for a local business - a handful. But for large advertisers the share of neural results in impressions almost doubled in a single quarter, and the format is gaining weight fast.

How the placement works

Trigger. A neural answer does not appear on every query. Most often it is questions that require an explanation or comparison: “which deposit is more profitable”, “how does an implant differ from a bridge”, “how to choose a kitchen”. Short transactional queries like “buy kitchen Moscow” get a detailed answer less often.

Ad selection. The ad is taken from ordinary Yandex Direct campaigns by relevance to the query and bid. There is no separate “neural account”, and no separate checkbox either. If your campaign covers the questions Yandex answers itself, you are already taking part in this auction - you are simply losing it on budget.

Answer sources. The text of the neural answer itself is assembled from pages the model considers authoritative on the topic. This is where the second entrance appears - not paid, but content-based. A site that is cited in the answer gets a visit even without advertising.

What small and medium-sized business should do

Don’t fight the banks for generic queries. Winning the auction on “how to choose a deposit” against Sber is impossible, and there is no need to try. The competitive field for a medium-sized business is narrow and local questions where there are no big advertisers: “implants with periodontitis St Petersburg”, “how long does screed take to dry in a new build”. It is exactly on such queries that a neural answer exists while the advertising in it is more often empty.

Questions in the semantics. Classic keyword cores collect “buy”, “price”, “order”. A core for neural results is questions: “how”, “why”, “what is the difference”, “which is better”. Collect them in a separate campaign and send them to pages that actually answer, not to a generic home page.

Answer pages. The model cites a page if it has a direct answer in the first paragraphs, a structure with subheadings, specific figures and a clear author. It is the same principle as for ordinary SEO, only stricter: the neural network does not retell filler, and it does not cite someone else’s text rephrased.

A metric for neural traffic. Visits from the neural answer are not yet shown as a separate source in Yandex Metrica, but they are visible by landing page: question pages start receiving search visits that were not there before. Set up a segment for such pages and watch how they convert - the figures differ from ordinary search traffic.

What not to do

The first reaction of many advertisers is to raise bids sharply in order to “get into neural”. That way you simply overpay for ordinary impressions: there is no separate bid for the neural block. The second mistake is to generate hundreds of “question” articles with a neural network in the hope the model will cite them. The model is trained on the same material and does not treat a retelling of itself as a source. The third is to wait for the format to “settle down”. In the second quarter alone the share of neural results in large advertisers’ impressions grew from 17 to 30%, and no equilibrium is in sight yet.

If there are no leads already, the place to start is not neural results but basic diagnostics: why there is a site but no leads. A neural answer amplifies a working funnel, but it does not fix a broken one.

In brief

• For the ten largest financial advertisers, neural results already deliver 30% of ad impressions, up from 17% at the start of the quarter.
• Advertising in it is selected by the ordinary Yandex Direct auction; 93% of impressions belong to Sber, Yandex and T-Bank.
• The place for medium-sized business is narrow and local questions where there are no big advertisers.
• The second entrance is free: pages the model cites as a source.
• The gain so far is 1-5% of traffic, but the format’s share of impressions almost doubled in a quarter.

Frequently asked questions

Can I buy advertising specifically in the neural answer?

No. There is no separate placement; ads get there from ordinary Yandex Direct campaigns by relevance and bid.

Can Yandex Metrica show that a visit came from a neural answer?

There is no separate source yet. It can be assessed indirectly - by landing pages with question queries.

Which matters more: advertising or getting into the answer’s sources?

For small business - the sources. The auction on mass queries is won by big budgets, while citation depends on page quality, not money.

Should I rewrite the whole site for neural networks?

No. A few pages that directly answer the questions of clients in your niche are enough. The rest is ordinary SEO and a working offer.