AI search optimization for websites

What AI search optimization and GEO are

AI search optimization is the work of making sure AI assistants find a site's pages, understand them correctly and cite them in their answers. This covers ChatGPT, Perplexity, Google AI Overviews, Alice and Neuro in Yandex search, and other assistants that answer a question themselves instead of returning a list of links.

People increasingly skip the results page and ask instead: "what should I choose", "how do I do this", "how much does it cost". The assistant reads several pages and builds an answer from fragments. If your site is not among those pages, you are not in the answer either, even with good rankings in regular search.

For us this is not theory. In 2026, SEO Writer's paying customers most often found us first through ChatGPT rather than through search results: a person asked the assistant which service to pick and followed the link in the answer.

The work goes by several names, and they nearly always mean the same thing:

Below is what exactly to do and which part SEO Writer takes over.

How it differs from classic SEO

AI search optimization does not replace classic SEO; it is built on top of it. Assistants that search the web take their sources from search engines or from their own crawler:

• Google AI Overviews follow the same technical requirements as regular search: the page has to be indexed and eligible for a snippet. This is stated in Google's documentation.
• ChatGPT crawls sites with its own OAI-SearchBot and also sends more specific queries to its search partners.
• Neuro and Alice in Yandex search take facts from web sources and link to them, as Yandex described Neuro at launch.

So a page that is not in the Google or Yandex index has almost no chance of getting into AI answers. The difference lies elsewhere: what the page competes for and how success is measured.

The main practical takeaway: a text has to work not only when read top to bottom, but also when cut into pieces. Each section answers its own question and makes sense without its neighbors.

How AI assistants choose sources

When an assistant needs a fresh answer, it rewrites the question into one or more search queries, opens the pages it finds, picks suitable fragments and builds an answer with links. OpenAI states that ChatGPT search may send additional queries that are more specific than the user's question.

So assistants do not pick "the best site"; they pick the best fragments for each of those queries. A page has a better chance of being cited when:

• the answer is in the first sentences of a section, not at the end after an introduction;
• there are specifics: conditions, timelines, prices, steps, a comparison table;
• the section heading matches how people phrase the question;
• it is clear who wrote it, when it was updated and what it relies on;
• the facts about the company are the same on the site and on other platforms.

We have tested this on our own service. When we asked assistants a buyer's questions, such as which Russian SEO article services accept payment with a Russian card, ChatGPT more often named services that had that fact written right on the page. If a fact only lives in the owner's head, the assistant will never learn it.

How to optimize a site for AI search: 8 steps

The order matters: the first two steps are technical, and without them the rest of the work will not pay off. Check your site against each one.

1. 1

   Let AI crawlers in

   Check robots.txt. OAI-SearchBot handles ChatGPT search, GPTBot handles model training. They are separate settings: blocking GPTBot does not remove a site from ChatGPT search, blocking OAI-SearchBot does. Perplexity has its own PerplexityBot. Google AI Overviews and Yandex Neuro need no extra crawlers beyond the regular Googlebot and YandexBot. Make sure important pages carry no nosnippet: Google keeps such pages out of AI Overviews. Watch the groups in robots.txt: on our own site the rules once sat only in the last crawler's group and did not apply to the others.
2. 2

   Serve the text in HTML

   Not every crawler runs JavaScript. If a page's text only appears after scripts run, some assistants will see an empty page. The check is simple: open the page source (Ctrl+U) and look for the first paragraph of text.
3. 3

   Lead with the answer

   The first one or two sentences of a section answer the question in its heading. At the top of a long article, a key takeaways block of 3–5 points helps: it is a ready short answer that is easy to quote.
4. 4

   Structure pages around questions

   Phrase H2 and H3 headings the way people ask. One section, one idea. Use lists for enumerations, tables for comparisons and numbered steps for procedures: an assistant can lift such fragments whole.
5. 5

   Add Schema.org structured data

   Article markup states the author and dates, FAQPage the questions and answers, HowTo the steps, and Organization on the home page who you are and how you differ from companies with similar names. Markup must match the visible text: a question in the markup that is not on the page adds no trust. You can check a page with our free tool.
6. 6

   Show expertise

   An author byline with a page about the author, an update date, links to primary sources, your own data and experience. Expert content with specifics gets cited more readily than a retelling of other articles.
7. 7

   Speed up indexing

   A new page will not show up in answers until it is indexed. Add the site to Google Search Console, Bing Webmaster and Yandex Webmaster, keep the sitemap current, and submit new URLs through IndexNow (supported by Bing and Yandex).
8. 8

   Work on mentions elsewhere

   Assistants cross-check sources. Directories, industry reviews, testimonials and listings with the same facts about you (name, what you do, prices, contacts) make mentions more likely. Different numbers on different platforms work against you.

What to focus on for a store, a service site and a blog

The plan is the same, but the emphasis depends on the kind of project.

Online store

A buyer asks an assistant what to choose and where to buy, and assistants answer such questions with roundups and comparisons. So an online store needs buying guides ("how to choose", "how X differs from Y", "what is best for…") with spec tables and links to categories. Product pages need full specifications in text, not only in images, plus Product markup with price and availability. Delivery, payment, warranty and returns should be described plainly, with numbers and conditions: this is the data an assistant checks first.

Service site

Specifics and trust decide here: what the service includes, what it costs and how long it takes, which city you work in, who does the work. Starting prices and timelines belong on the service page, and clients' common questions in a separate block with answers. An assistant cross-checks sources, and reviews on independent platforms are one of its signs that a company can be trusted.

Blog and expert site

A blog needs answers to questions and first-hand experience: data, examples, mistakes you have seen yourself. An assistant has no use for a retelling of other articles — it already has the originals. Refresh older posts, show the update date and connect articles with internal links: this shows that the site covers the topic in depth.

Common AI search optimization mistakes

• Blocking every AI crawler at once. Sites often block OAI-SearchBot together with GPTBot and drop out of ChatGPT search. If you only want to opt out of model training, block GPTBot.
• Making hundreds of pages, one per question. Thin pages built for a long tail of queries are treated as low value by search algorithms: Google explicitly classes mass-produced content of this kind as spam. One detailed page where every section answers its own question works better.
• Publishing generated content unchecked. A wrong price or deadline quoted by an assistant hurts a brand more than not being mentioned at all. This applies to articles from our service too: read them before they go live.
• Marking up what is not in the text. Questions and ratings in Schema.org that users cannot see on the page break search engine rules.
• Keeping different facts in different places. One price on the site, another in a directory, a third on social media — an assistant will pick any of them or skip you altogether.
• Expecting results from a single change. AI visibility comes from indexing, content quality and mentions together. One llms.txt file or one article will not change it.

How to check your visibility in AI answers

Assistants offer no ready-made reports like Search Console, so AI visibility is measured by hand or with scripts. A working routine:

1. Write down 10–15 questions a buyer asks before purchasing: "what should I choose", "how is X different from Y", "how much does it cost", "where can I order it".
2. Ask them in ChatGPT, Perplexity, Google AI Overviews and, for the Russian market, Yandex's Alice and Neuro. Note whether the site is mentioned, whether there is a link and who is named next to it.
3. Repeat monthly with the same wording: answers shift, and only identical questions can be compared.
4. In your analytics, look at visits from chatgpt.com and perplexity.ai; in server logs, look for OAI-SearchBot, ChatGPT-User and PerplexityBot.

This is how we check SEO Writer too: 15 buyer questions once a month. The very first round showed how much assistants differ: one named the service in almost every answer, another only now and then. Look at each assistant separately rather than at "AI" in general.

What SEO Writer does in every article

SEO Writer writes articles for your site and publishes them. Almost everything from the content part of the plan above happens in every article with no extra setup:

What stays with you: crawler access in robots.txt, the technical state of the site, facts about the company on the site, mentions elsewhere, and a check of each article before publishing. The model writes from the facts it finds, but your company's figures, prices and promises need a human to confirm them.

What no service can promise

Nobody can guarantee a place in AI answers. An assistant's answer depends on the wording of the question, the conversation so far, the region and the day: the same question at different times returns different sources. Only the assistant's developers control that, not an agency and not a service. A promise to "guarantee you a spot in ChatGPT" is a reason to ask what exactly the contract says.

What you can control is what source selection depends on: indexing, the structure and facts on your pages, structured data, mentions. That is also what is worth measuring: whether a page is indexed, which queries it is shown for, and whether the site is mentioned in answers to your test questions.

AI visibility: a service, an agency or in-house

Getting into AI answers takes three kinds of work, and they are easy to split between different people:

• Technical — robots.txt, HTML that does not depend on scripts, speed, indexing. A one-off job for a developer or an SEO specialist.
• Content — articles and pages with answers, structure and markup. This is an ongoing stream, and a service costs less than an agency here: SEO Writer prepares and publishes articles with the right structure on its own.
• Brand mentions — reviews, directories, testimonials, PR. A service cannot help with this: it is work for an agency, a PR specialist or you.

Agencies that do AI visibility usually take on all three, which makes sense for a large brand. A smaller site usually needs a one-off technical setup and a steady flow of articles.