ReleaseInsight · On RadarHabr AI→ original

Google's AI Answers in Search: From 18% to 88% of Queries — It All Depends on the Sample

The figure of "80% of queries with a Google AI answer" is a property of the sample, not of the market. Semrush, on 400 commercial keywords in the US, counted 88%, seoClarity 30%, and Pew, using real people's searches, about 18%. There is no overall figure for the Russian-language internet (Runet): Yandex does not publish the share of Alice AI answers, so you will have to measure it yourself.

AI-processed from Habr AI; edited by Hamidun News
Google's AI Answers in Search: From 18% to 88% of Queries — It All Depends on the Sample
Source: Habr AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceHabr AI ↗
Editor-in-Chief
Key metric #1Long informational queries get an AI answer 53% of the time, short ones 8%
Key metric #2Clicks on a regular link beneath an AI answer fall from 15% to 8%
Key metric #3Measure the share of Alice AI on dozens of your own questions, broken down by type

If you have heard that Google shows an AI answer above 80% of queries, there is no need to rush into overhauling your SEO. That figure holds for one specific panel, not for search as a whole. Let's look at where the spread comes from and what it means for projects on the Runet.

One Search Engine, Three Different Figures

Semrush took a fixed panel of 400 commercial keywords (banks, software, online stores) and, every month from September 2025 to August 2026, looked at US desktop results. The share of keywords with AI Overviews grew from 45.75% to 88%. Weighted by search frequency, 97% of searches in the panel have an AI answer: the 12% of keywords without one account for less than 3% of demand.

seoClarity, on its own keyword set, counted 30% of US desktop queries for September 2025. Pew Research looked at live searches: 900 American adults made 68,879 queries in March 2025, and about 18% of them had an AI summary. The spread is almost fivefold, yet the search engine is the same everywhere. Only the set of queries changes.

The main factor is the type and length of the query. In Pew's data, 8% of queries of one or two words got an AI summary, versus 53% of queries of 10 words or more. According to SE Ranking, the more frequent the query, the less often an AI answer appears. In seoClarity's data, transactional queries get one 12.54% of the time, local ones 0.01%. A single overall average simply blends all of this into mush.

What About Yandex and Alice AI

Yandex does not publish what share of queries show Alice AI. The Webmaster help page says only that an answer appears when the algorithms deem it useful, and that it supplements the results rather than replacing them. The only figure from the company itself dates from April 2026: updated models increased the share of queries with a quick answer by a factor of 1.5. The base is not stated, and a 1.5x increase from 2% versus from 30% are completely different situations.

Indirect measurements do not provide an overall share either. Topvisor, on one premium e-commerce project (2,909 queries over a year and a half), calculated how often an AI answer cites a site with an organic ranking: 28.2% for question queries and 21.4% for the rest. That is a share of citations, not a share of queries with an AI answer.

Organic results are not being devalued, though. According to a report by Ashmanov and Partners, from August 2025 to January 2026 the share of sites outside Yandex's top 10 in Alice's answers fell from 40% to 10%. A Yandex representative says a quick answer is built on roughly ten sources.

How Many Clicks Are Lost

Open figures exist only for Google. According to Pew, when an AI summary was present, users clicked a regular link in 8% of visits, versus 15% without one, almost twice as often. Links inside the summary itself got just 1%. A study based on Wikipedia data estimated the drop in English Wikipedia traffic at roughly 5%, while Kommersant writes of losses of up to 30% of organic traffic for informational sites but does not provide its methodology. There is no publicly available measurement of losses specifically under Alice AI.

How to Measure It Yourself

The author suggests a simple procedure using your own data:

  1. Take several dozen queries for which your site already gets impressions in Yandex Webmaster, and add the questions customers ask in your niche.
  2. Break them down by type: how to do something, how much it costs, what to choose, buy, near me.
  3. Enter each question in Yandex search and note whether an Alice AI answer appears above the results.
  4. Calculate the share for each type separately, not an overall average.
  5. For questions with an AI answer, check whether your site is among the sources.

Practical Takeaway

Do not transfer Semrush's 88% to your own market, and do not transfer a measurement from one niche to a neighboring one. The share of AI answers across several dozen of your own questions will tell you more than any Google report: it is taken from where your traffic comes from. If an AI answer appears above only a small part of these questions, there is no reason to restructure your Yandex organic work because of other people's figures. If above a large part, it is worth seriously working on getting into the answer's sources, starting with informational and long questions, where it appears most often.

Story

Key Questions Answered

What are the key details on One Search Engine, Three Different Figures?

Semrush took a fixed panel of 400 commercial keywords (banks, software, online stores) and, every month from September 2025 to August 2026, looked at US desktop results. The share of keywords with AI Overviews grew from 45.75% to 88%. Weighted by search frequency, 97% of searches in the panel have an AI answer: the 12% of keywords without one account for less than 3% of demand. seoClarity, on its own keyword set, counted 30% of US desktop queries for September 2025.

What About Yandex and Alice AI?

Yandex does not publish what share of queries show Alice AI. The Webmaster help page says only that an answer appears when the algorithms deem it useful, and that it supplements the results rather than replacing them. The only figure from the company itself dates from April 2026: updated models increased the share of queries with a quick answer by a factor of 1.5. The base is not stated, and a 1.5x increase from 2% versus from 30% are completely different situations.

How Many Clicks Are Lost?

Open figures exist only for Google. According to Pew, when an AI summary was present, users clicked a regular link in 8% of visits, versus 15% without one, almost twice as often. Links inside the summary itself got just 1%. A study based on Wikipedia data estimated the drop in English Wikipedia traffic at roughly 5%, while Kommersant writes of losses of up to 30% of organic traffic for informational sites but does not provide its methodology.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…