Yandex and Google both now put AI-generated answers above the search results, and a small industry of advice has grown up around them: add an llms.txt file, chop your text into chunks, add "AI markup", rewrite your articles for the models. Some of that advice contradicts documentation the search engines have published themselves. This article sets out what the primary sources say about getting into those answers, what is practice or hypothesis, and where FAQ blocks, answer-first paragraphs and structured data fit. Every quote from the documentation was checked on 15 October 2026.

How AI answers in search are put together

Google describes its mechanism in its guide to generative AI features in Search (developers.google.com, last updated 10 July 2026, checked 15 October 2026). AI Overviews and AI Mode rest on Google's core ranking systems. The model retrieves relevant pages from the Search index, which Google calls retrieval-augmented generation or grounding, reads them and writes an answer with links to its sources. The second technique is query fan-out: the model issues several related queries of its own and gathers results for each. Google's example is a question about a lawn full of weeds, which fans out into queries about herbicides, weeding without chemicals and prevention.

Yandex, the dominant search engine in Russia, documents something similar in its Webmaster help page on how Alice AI answers are formed (checked 15 October 2026). Alice AI analyses the query, splits it into sub-questions where needed, sends those to Search and assembles an answer from blocks of meaning. In the help page's example, the answer to "why is sea water salty" opens with a one- or two-sentence summary, followed by separate paragraphs on where the salt comes from, how it accumulates and how it is distributed. The page stresses that Alice AI relies on sources that rank highly in search and does not reproduce page text verbatim, except for direct quotations.

The naming is confusing. Yandex's help for searchers still calls the feature "Search with Neuro", while the help for site owners talks about Alice AI answers throughout. Yandex's own timeline, on its page about quality factors, runs: quick answers based on YandexGPT in 2023, Alice AI answers in 2025. For a site owner it is the same block above the results.

Both descriptions mean that pages get into AI answers through ordinary search results. Neither engine has a separate queue for "AI-optimised" pages. A page that is not indexed, or sits on page three, will almost certainly not become a source however it is rewritten. Checking indexing therefore stays the first step, and how to do it in both consoles is covered in our guide to checking whether your site is indexed.

What Google's documentation says

The requirements are short. Google's page on AI features and your website (developers.google.com, last updated 10 December 2025, checked 15 October 2026) says there are no additional requirements to appear in AI Overviews or AI Mode. A page has to be indexed and eligible to be shown in Google Search with a snippet. You do not need new machine-readable files, AI text files or special schema.org markup.

The generative AI guide adds a second condition: the site must be included in generative AI features through a Search Console setting called Search generative AI. According to Search Console Help (checked 15 October 2026), the control was rolled out to all websites as of 31 August 2026, and inclusion is the default. If you choose to exclude the site, it stops appearing in AI Overviews, AI Mode and Discover's generative features, and you get no traffic from them.

The same guide has a mythbusting section listing popular advice you can ignore for Google Search:

  • llms.txt files and other "special" markup. Google Search does not use them. Keeping such a file for other services is fine; it neither helps nor harms your visibility in Google.
  • Chunking content. Google's systems understand a page that covers several topics and show the relevant part. Google names no ideal page length.
  • Rewriting content just for AI. The models understand synonyms, so there is no need to chase every long-tail phrasing.
  • Seeking inauthentic mentions on other sites. Generative features depend on both ranking and spam-fighting systems.
  • Overfocusing on structured data. It is not required for generative AI search.

The guide also carries a warning for anyone planning a separate page for every question. Creating pages for every variation of a query, fan-out queries included, primarily to manipulate generative AI responses violates Google's scaled content abuse policy. As for "AEO" and "GEO", the guide's position is that optimising for generative AI search is optimising for search, and therefore still SEO.

What Yandex's documentation says

Yandex makes the same point in its own words. Its page on what affects ranking in Search and inclusion in Alice AI answers (checked 15 October 2026) says that because generative answers are built on search results, no fundamentally new requirements arise. It then lists four quality aspects whose Russian initials spell ЭПОС ("epic"): expertise, usefulness, originality and substance.

Each aspect comes with a self-check question. For originality it asks whether you are creating content of unique value that is hard or impossible to get anywhere else. The page notes that textual uniqueness guarantees nothing by itself: a text rewritten or generated from other people's material can be "unique" and still worthless. For substance, Yandex asks you to avoid filler and observes that not every long text has substance. Where the line sits when you work with language models is covered in our piece on AI-written text and search engines.

A few details from the Alice AI help page change what "ranking" in an answer means:

  • Source links in an answer are not ranked. They follow the meaning and logic of the answer.
  • There can be more links than results on a standard results page, and several of them can point to one site.
  • One page can be linked more than once if it supports several paragraphs.
  • The answer to the same query changes from day to day and even hour to hour, partly because of personalisation.

That makes a single manual check of "are we in it or not" meaningless. The site is in the answer in the morning and gone by the evening, and neither tells you anything about the page. You need statistics over a period, which come later in this article.

FAQ blocks now that the rich result is gone

For years people added FAQ blocks for the rich result: FAQPage markup gave you expandable questions under your listing in Google. That ended in two stages. In September 2023 Google limited the feature to well-known, authoritative government and health websites. On 8 May 2026 its changelog added a deprecation notice saying the feature would no longer appear in Search from 7 May 2026, and in June the documentation was removed entirely (developers.google.com/search/updates, checked 15 October 2026).

So an FAQ block now works only as text, which is no reason to delete it. A good block answers what a reader asks after the main article, in short answers that stand on their own. Yandex describes a similar shape when it explains how Alice AI builds an answer: a short answer first, then blocks of detail. Our hypothesis, which the documentation neither confirms nor rules out, is that a "specific question plus two- or three-sentence answer" pair suits a system that splits queries into sub-questions itself. Test it on your own data; the measurement section below shows how.

How to collect questions so the block does not turn into a list of the obvious:

  1. Take question-form queries from Yandex Wordstat (Yandex's keyword volume tool) and from the Search Console queries report for that page. If the page already gets impressions for "can I…", people really ask it.
  2. Read the People also ask box in Google and Yandex's query suggestions for the main query. They are ready-made phrasings from real people.
  3. Add questions that reach support and sales. In its section on relevance, Yandex advises talking to colleagues who deal with customers.
  4. Drop questions whose answer already appears almost word for word in the body. Repeating a paragraph in the FAQ adds nothing.

Write each answer so it can be read without the article. The first sentence answers the question: yes, no, a number, a name. The second and third give the condition or exception. If an answer needs a paragraph and a table, the question deserves its own section in the body.

Answer-first paragraphs

The second technique recommended for AI answers is called answer-first: a section opens with a direct answer to its heading, and the detail follows. Neither Google nor Yandex requires it, and Google even says you do not need to rewrite content for generative search. Still, it has long been plain good editing, and it matches how Yandex describes an Alice AI answer: short answer, then explanation.

An example shows the difference. A section headed "How much does moving a site to a new domain cost" might open: "A migration touches many aspects, and the cost depends on various factors." Delete that sentence and nothing is lost. The answer-first version: "For a site under 500 pages where URLs stay the same, a move usually takes a specialist two or three working days; sites whose URL structure changes cost more, because every old address needs a redirect." The figures here are illustrative; in your article they should be your own, and checked.

The rules we apply when editing:

  • The first sentence of a section answers its heading. If the heading names a topic, the first sentence gives the section's main conclusion.
  • A paragraph makes sense without the one before it. Openers like "this approach" or "as mentioned above" are better replaced with the name of the thing.
  • Numbers, dates and names appear explicitly. "Recently" and "many experts say" give neither the reader nor a model anything to use.
  • A term is defined where it first appears.

Do not overdo it. Google says there is no need to break content into small pieces. A self-contained paragraph helps the reader. A one-sentence stub under every subheading turns the article into a stack of cards nobody finishes.

Structured data: what it does and does not do

Google's documentation says there is no special markup for AI Overviews or AI Mode, and structured data is not required for generative AI search. Google recommends keeping it as part of ordinary SEO, because it makes pages eligible for rich results. Its list of best practices on the AI features page also says structured data should match the visible text on the page.

Yandex's Schema.org help page (checked 15 October 2026) lists where Yandex uses markup: products and prices, software, recipes, films, academic papers, organisations, images and video. For questions and answers it mentions markup for mobile results that helps pick out the best answer; that applies to pages where users ask and answer questions, such as forums. The page says nothing about markup as a condition for appearing in Alice AI answers.

What this means for an article page with an FAQ block:

  • Organization, breadcrumb and article markup remain useful hygiene. They help search engines understand what the page is and who published it.
  • FAQPage markup can stay if it already exists and matches the visible block. It will no longer produce a rich result in Google, and the documentation gives no reason to add it from scratch for AI answers.
  • Markup that describes content not on the page breaks both engines' rules, AI or no AI.

If your team is about to spend a developer-week on "markup for AI", that week is better spent by an editor rewriting the opening paragraphs of twenty pages that already rank in the top ten.

How to measure visibility in AI answers

Both engines now have their own reports, and they work differently.

Google Search Console has a Generative AI performance report. According to Search Console Help (checked 15 October 2026), it shows how often links to your site appeared in AI Overviews and AI Mode, the trend over time, which pages get the most and fewest impressions, and the split by device and country. As of 31 August 2026 it is available to all websites; if you do not see it, your site has not yet received enough impressions in these features. Search Labs experiments are excluded. In the main Performance report, impressions and clicks from AI Overviews and AI Mode are counted under the "Web" search type along with the rest of organic search, and cannot be separated there.

Yandex Webmaster has a tool called Site visibility in Alice AI (help checked 15 October 2026). It does not count clicks or impressions. Its main metric is Share of Voice: the share of queries where an Alice AI answer mentioned your site, out of all queries Alice AI answered. Only queries where your site already ranks reasonably high are counted. Data covers the last three months and updates weekly. You can also see sample queries where the site is or is not mentioned, and sites on the same topic that appear in answers often.

Web analytics is the third source. Google Analytics 4 has an AI Assistant channel for visits from ChatGPT, Gemini, DeepSeek, Copilot, Grok and similar services. Visits from AI Overviews and AI Mode do not go into it; they stay in organic search (support.google.com, checked 15 October 2026).

For a client report this gives two lines in the visibility section: Share of Voice in Alice AI and impressions in Google's generative features, each with its trend over the quarter. Do not compare one with the other, because they are measured in different ways.

How to keep your content out of AI answers

Some site owners do not want their material summarised. Yandex has a dedicated robot for this. Its help page on preventing indexing (checked 15 October 2026) says to disallow the URLs in robots.txt for the YandexAdditional or YandexAdditionalBot user agent:

User-agent: YandexAdditionalBot
Disallow: /cats/mycat.html

The change takes effect as search data updates, within 2 to 14 days. The main indexing robot keeps working, and the page stays in ordinary results.

Google gives you two options. The first applies to the whole site: in the Search generative AI setting in Search Console you can choose to exclude it. According to the help page, changes usually take effect within a few days, and the setting does not affect AI training, which is controlled separately through the Google-Extended token. The second option works per page: the nosnippet, data-nosnippet, max-snippet and noindex controls limit what Google can show, AI Overviews and AI Mode included. How to set those in a meta tag or an X-Robots-Tag header is covered in our article on robots.txt and noindex.

Google spells out the cost: an excluded site gets no traffic and no impressions from these features. nosnippet also removes your description from ordinary results. Before opting out, look at how many mentions and visits you already get from AI answers. If Share of Voice is zero and the Google report is empty, the decision can wait.

A workflow for existing articles

If the goal is to appear in AI answers more often, here is a sequence that does not depend on unverified tricks:

  1. Make sure the pages are indexed in both engines and not blocked from snippets. Nothing else works without that.
  2. Pick pages that already rank in the top ten for informational queries. Yandex only measures Alice AI visibility for queries where the site ranks high, and Google builds answers from its own results, so these pages pay off fastest.
  3. Rewrite the opening of each section so it answers the heading, and cut introductions that carry no facts.
  4. Build an FAQ block from real queries and customer questions, five to eight per page.
  5. Add what competitors do not have: your own figures, examples and dated test results. This is what Yandex calls originality and Google calls non-commodity content.
  6. After six to eight weeks, compare Share of Voice in Webmaster and impressions in the Search Console report with your starting values. Changes in AI answers also depend on ordinary rankings, so read both metrics together.

This kind of rework fits into the same plan as new content, for example as a batch of twenty pages a quarter. How to spread rework and new pieces across the weeks without missing deadlines is covered in our article on a quarterly content plan. Bear in mind that Yandex and Google weigh the same signals differently, so results in the two engines can diverge. The other places they need separate handling are collected in our piece on Yandex and Google.

Frequently asked questions

Does my site need an llms.txt file to appear in AI Overviews?

No. Google's generative AI guide says Google Search does not use such files and they have no effect on visibility. If other services need the file, you can keep it. Yandex's help does not mention llms.txt.

Does FAQPage markup help a page get into Alice AI or Google answers?

Neither engine's documentation says so. Google says no special markup is needed for generative features, and the FAQ rich result has not appeared in its results since 7 May 2026. The FAQ block is useful as text, and the markup only needs to match it.

How do I find out whether Alice AI mentions my site?

Open the Site visibility in Alice AI tool in Yandex Webmaster. It shows your Share of Voice over the last three months and sample queries. A "not enough data yet" message means the site ranks too low or is mentioned too rarely.

Can I stop Yandex using my site in answers without dropping out of search?

Yes. Disallow the relevant URLs in robots.txt for YandexAdditionalBot or YandexAdditional. The main robot keeps indexing the pages, and the block takes effect within 2 to 14 days.

Where does Search Console show clicks from AI Overviews?

In the main Performance report, under the "Web" search type, together with the rest of organic search. Impressions in generative features are shown separately in the Generative AI performance report, which has been available since 31 August 2026 to every site with enough impressions.

Do I need to rewrite all my articles for AI?

No. Google says you do not need to write specifically for generative search, and Yandex says no fundamentally new requirements arise. What pays off is reworking pages already in the top ten: answers at the start of sections, dated facts and an FAQ built from real queries.