What zero-click search is

A zero-click search is a session that ends without a visit to a site. The user entered a query, got the answer directly on the results page or in a conversation with an assistant, and closed the tab.

The phenomenon is not new. It began with the blocks that search engines built into the results long before generative models:

  • instant answers and featured snippets at the top of the results;
  • knowledge panels about companies, people and brands;
  • calculators, currency converters, weather, timetables;
  • local results with the address, phone number and opening hours;
  • the people-also-ask block with expandable answers.

The new stage began with generative answers. AI Overviews, the equivalent generative answers in Yandex — the dominant search engine in Russia and several CIS markets — and search chat assistants now close not simple factual questions but complex ones: comparing options, selecting by criteria, explaining with caveats. A question like that used to cost the user three or four visits to different sites; now it costs zero.

What it changes for e-commerce

It breaks the chain most SEO reporting is built on: position, click, visit, conversion. The position can stay where it was, impressions can even grow, and visits still fall.

Before Now
The goal is the top ten and a click The goal is the answer and being named as a source
The success metric is organic traffic The success metric is share of impressions, mentions, branded demand
CTR indicates snippet quality CTR also indicates that the answer was taken in the results
Content answers the question in detail Content answers precisely up front and adds what cannot be paraphrased
Position value is even across queries Value diverges sharply: informational queries devalue, transactional ones do not

The effect is distributed unevenly across search intent:

  • Informational queries (“how OLED differs from QLED”, “how to choose running shoes”) are the zone of maximum loss. That cluster is usually exactly what an online store’s blog is built on.
  • Navigational queries (“METRO website”, “my account login”) are barely touched: the user is going to a specific place regardless.
  • Transactional queries (“buy a fridge with delivery today”) are only weakly affected: placing an order requires reaching the site. An assistant can suggest options, but the purchase happens in the store.

That leads to a non-obvious but practical conclusion: a fall in blog traffic with stable sales is not necessarily an SEO failure. It can be a redistribution in which the brand still takes part in the decision, but its participation has become invisible in a click report.

⚠

The dangerous reaction to a drop is to start producing more informational articles to “win the traffic back”. That usually increases the number of pages a generative answer paraphrases without a link, and moves neither citability nor revenue.

How to rebuild the metrics

An organic traffic report stops being sufficient. A working set looks roughly like this:

Indicator Where to look What it tells you
Impressions across a query cluster Search Console, Yandex Webmaster Presence in the results independent of clicks
CTR at a fixed position The same reports, split by position How much of the answer is taken before the click
Share of mentions in AI answers A manual audit of 20–50 queries, dedicated tools Real visibility in generative results
Branded queries Search engine reports, keyword tools such as Google Keyword Planner or Yandex Wordstat The indirect trace of mentions with no click
Direct visits Web analytics Part of the “saw it in an answer, came by myself” journey
Conversion of the remaining traffic Web analytics A test of the hypothesis that it is low-intent visits leaving

It is separately useful to track the gap between the informational and the commercial cluster: if impressions grow, blog clicks fall, and product page traffic and conversion hold steady, the picture is healthy. If both fall, the problem is not zero-click.

💡

Take a baseline measurement before you change anything: 30 to 50 priority queries, each with its position, impressions, clicks, CTR and whether the brand is present in the generative answer. Without that reference point you cannot tell the zero-click effect apart from ordinary seasonality or a ranking drop.

What to do with the content

The strategy comes down to two moves at once: be the one cited where the answer will be taken anyway, and provide what cannot be taken.

Being cited:

  1. A direct answer in the first two or three sentences of a section. The model extracts a fragment — a preamble reduces the chance of being in it.
  2. Schema markup. FAQPage, Product, Offer and Article provide machine-readable facts instead of guesses from the layout.
  3. Specifics with numbers and caveats. “Integration from 2 weeks” gets cited; “integration is fast” does not.
  4. A question-and-answer structure. H2 headings that repeat the phrasing of a real query raise the chance of a match.
  5. An explicit source and date. An author, an update date and a link to the primary source are what E-E-A-T rests on.

Providing what cannot be paraphrased in one line:

  • proprietary data and calculations — measurements, internal test results, category-level industry cuts;
  • live prices, availability and delivery times to a specific region;
  • tools: parametric selection, calculators, configurators, model comparisons;
  • genuine reviews and customer photos;
  • the depth people still come for: detailed reviews, edge-case analysis, operating instructions.

Branded demand is the third pillar. If someone remembers a name from an assistant’s answer, their next query will be branded, and that one does bring them to the site. A brand mention in a generative answer therefore has value of its own even when no click followed, and it is worth counting separately from traffic — exactly as GEO does.

Checklist

  1. Split queries into informational, navigational and transactional, and assess the trend of each cluster separately.
  2. Take baseline measurements: positions, impressions, clicks, CTR and presence in AI answers across 30 to 50 queries.
  3. Check whether priority pages have a direct answer in the first paragraph of each section.
  4. Roll out schema markup on product pages, listings and FAQ blocks; verify it matches the visible content.
  5. Make sure the crawlers of generative systems are not blocked in robots.txt, if citation is what you want.
  6. Add a mentions metric and branded demand trends to the reporting.
  7. Revise the content plan: fewer retellings of publicly available facts, more proprietary data and interactive tools.