Home / Blog / Contextual ads & paid social

Contextual ads & paid social · 4 min · 26 August 2026 · Updated 2026-08-26

Why does Direct show ads for irrelevant queries?

Short answer

The list of phrases you set and the list of searches your ads actually appeared for are two different lists. Three mechanisms widen it: broad match, autotargeting and interest-based targeting in the networks. All three are useful, and all three spend money when left unwatched. You can see the difference between the lists in ten minutes, in a single report.

Artur KupriyanovArtur KupriyanovFounder of Neuron

Your key phrases are not what people see

An owner sets twenty phrases about gearbox repair, and a month later finds “car mechanic job”, “how a gearbox works” and “DIY automatic transmission repair” in the statistics. The conclusion seems obvious: Direct is cheating. In fact it is doing exactly what it says it does — it’s just that nobody read up on these mechanisms.

Three mechanisms that widen your list

Broad match

A phrase without special operators doesn’t mean “show for this phrase”; it means “show for searches containing these words in any form, any order and with any additions”. “Automatic transmission repair” catches both “automatic transmission repair price” and “DIY automatic transmission repair video”. This is not a bug but the design: narrow match is set with quotation marks and exclamation marks, broad match by their absence.

Autotargeting

It picks searches on its own, based on the ad text and the content of the landing page. It works with search query categories, and Yandex has already changed several times which categories are available and whether they can be switched off. So the only reliable check is to open your ad group settings and see what is enabled right now, rather than rely on an article from a year ago, this one included.

More important is this: if the landing page describes more than your service, autotargeting will honestly bring people for everything written on it. A general company page listing twelve lines of business is permission to bring traffic on twelve topics.

Interest-based targeting in the networks

In the ad network, an ad is shown not in response to a search but based on what the person was interested in before. That is why the network report has no “search queries” in the usual sense, only placements. Negative keywords have almost no effect there — you manage the list of placements and the audience settings instead.

Junk traffic in Direct is not a glitch, it is the default setting. The system is built to reach as many people as possible, and narrowing it down is your job.

Where to see what is actually happening

In Direct: Statistics → Search queries, period — one month. The report gives two decisive columns: the search itself and which of your phrases it was matched to. Sort by spend, not by impressions: you are interested not in the most frequent searches but in the most expensive ones.

Go through the top twenty rows and sort them into three groups:

  • Ours. The person wants what you sell.
  • Irrelevant. Jobs, training, “DIY”, “free”, another city, another product, a competitor’s name.
  • Borderline. Informational searches around the topic: “how to tell if it needs repair”. These are future customers, not today’s.

The spend on the second group is the amount you paid for nothing.

How to build negative keywords without cutting off demand

Typical groups found in almost every business:

  • Employment. job, vacancy, salary, CV, get hired.
  • Doing it yourself. DIY, by yourself, instructions, diagram, drawing.
  • Free stuff. free, for nothing, download, torrent.
  • Educational. what is, essay, coursework, training, courses.
  • A different product. used, rental — if you sell new and don’t rent out.
  • Wrong geography. cities where you don’t operate.

Next comes what a standard list won’t catch: words from your own report. They produce most of the effect because they reflect your niche, not generic patterns.

The main trap is adding negatives too broadly. The word “cheap” looks sensible as a negative right up until it turns out half your customers type “inexpensive”. The word “reviews” cuts off people who check you out before buying. Before you add a word as a negative, read aloud three searches containing it from your report — and decide based on them, not on the word.

A thirty-minute check

  1. Direct, “Search queries” for a month, sorted by spend.
  2. Sort the top twenty rows into “ours / irrelevant / borderline”.
  3. Add up the spend on the “irrelevant” group. That is your monthly cost of doing nothing.
  4. Open the ad group settings: is autotargeting on, and which categories.
  5. Read the landing page through a robot’s eyes: how many different topics does it describe?
  6. If you run network ads — the placements report, sorted by spend, the top twenty names.

What you won’t get even after the cleanup

Perfectly clean traffic. Some untargeted impressions will always remain; that is the price of reach. The goal is not zero but a share you understand.

A one-off fix. Searches change: new models, new wording, new reasons. The search query report is a regular procedure, every two to four weeks.

An automatic rise in leads. A cleanup removes wasted spend. If there are still no leads after it while traffic is live, another link is broken — see “clicks but no leads”.

For a sense of scale: according to our Wordstat measurement on 26 August 2026, searches about autotargeting in Direct and switching it off got about 480 impressions a month across Russia, and a quarter of them were directly about how to turn it off.

Where to go next

The answer above is based not on general opinion but on measurements we take ourselves.

More in this section

To answer your question specifically

The article answers in general; a review answers about your site: a 100–150 page crawl, rankings, demand and a comparison with the top results. An hour of our time, free of charge.

Write to us