People used to take a question to Yandex, see ten blue links, and pick one. Now, more and more often, they ask a neural network and get a ready answer — with no list of links. If your site is mentioned in that answer, you win. If not, it's as if you don't exist, even when you're at the top of regular search.
Optimization for this new way of searching is called AEO — Answer Engine Optimization, optimization for answer engines. It's not a replacement for SEO, but a new layer of it.
- Goal: get into the text of the model's answer
- Success: you were cited as a source
- Judge: the language model
Why this matters already
The share of queries going to neural networks and search engines' AI answers instead of classic results is growing fast — by various estimates it's already around 15% and heading toward 30–40% in the coming years. Some of these people won't see a single blue link: they'll read the answer and leave. The only way to stay in the game is to end up inside the answer itself.
The good news: while competitors think only about the top of Yandex, spots in AI answers are freer. It's easier for a young site to get in here than into the overcrowded classic results.
How a neural network chooses whom to cite
The model answers a question based on what it found online (or what it was trained on). For it to take information from your page specifically, that information has to be:
- Explicit. A fact written as a direct sentence, not hidden between the lines. "The average order value in niche X is 1,500 rubles" the model will take. "Prices vary for everyone, it depends on a lot" — no.
- Structured. Headings, lists, tables, question-answer blocks. The model pulls an answer out of a clear structure more easily than out of a solid wall of text.
- Self-contained. A paragraph answers the question in full, without "as we wrote above." The model often takes one fragment out of context — it has to be understandable on its own.
- Credible. Specifics, numbers, dates, sources. Vague text the model steers clear of, to avoid getting it wrong.
Essentially this is the same thing a live reader in a hurry likes. AEO rarely contradicts sound SEO — it sharpens it.
What to do on the page
- Question-headings, short paragraphs
- A direct fact with a number in the first sentence
- A FAQ block with questions and answers
- Schema.org markup (FAQ, HowTo, Article)
The practical minimum:
- Answer the question in the first paragraph, and unfold the details below. The model and a person in a hurry take the beginning.
- Add a FAQ block — pairs of "specific question → short answer." It's the ideal format for citation.
- Mark up content with Schema.org — FAQPage, HowTo, Article. Markup tells the machine directly where the question, the answer, and the steps are.
- Write facts as statements with numbers. Not "quite a lot," but "about 40%."
- Don't block your site from AI bots. Check that robots.txt allows access for GPTBot, ClaudeBot, PerplexityBot, and others — otherwise the model physically won't see your page.
Check your AI visibility
Telling by eye whether neural networks cite you and how ready your page is for it is hard. We have
a free test: open /free/ai-visibility, give it a link to a page —
the tool assesses how fit it is for AI answers and suggests what to fix.
Common questions
Will AEO replace regular SEO?
How do I know I'm already being cited?
Do I need Schema.org markup if the site is small?
This is the last article of the guide. If you've gone through it in full, you already have the whole picture: from building the core to clusters, pages, and visibility in neural networks. Come back to the wiki table of contents whenever you need to refresh a specific step.