How to get your company cited by ChatGPT and Perplexity: a practical AEO and GEO playbook

A prospect types a question into ChatGPT instead of Google: "best accounting system for a 10-person agency" or "who in Almaty integrates 1C with Kaspi." The model answers with three or four named companies and a short reason for each. Your company ranks on page one of Google for the same query. It isn't in the answer.

That gap is what answer engine optimization (AEO) and generative engine optimization (GEO) are built to close. Neither is a replacement for SEO. Both depend on a different set of mechanics, and getting cited requires understanding those mechanics rather than repackaging a blog strategy with a new label.

Why this is a different problem than ranking on Google

Google's results page still works the way it has for two decades: a list of links, ranked by a mix of relevance and authority signals, with the user doing the reading. ChatGPT and Perplexity compress that step. They read for the user and hand back a short, synthesized answer with, sometimes, a citation attached.

That synthesis step changes what matters:

• The unit of value isn't a ranking position, it's a sentence. A model pulls the one paragraph on your page that answers the question cleanly, not your whole page.
• Citations are selective, not exhaustive. An answer carries only a handful of sources. If your page says the same thing as five competitors but says it less directly, it gets skipped.
• Freshness and retrieval replace some of what links used to do. Perplexity runs its own crawler and index, and ChatGPT has its own search crawler. Neither relies purely on backlink authority the way classic ranking does.

None of this means links and domain authority stop mattering. Both tools still favor pages that already rank well and get referenced elsewhere. A second layer sits on top: can the system actually extract a clean, quotable fact from your page once it finds it?

In Kazakhstan, add Yandex to the picture. According to StatCounter, Google had 68.25% of searches in the country in October 2026 and Yandex 29.61%. Yandex describes its Alice AI answers in Search as built from pages in its search results, with links to the sources. So a page Yandex barely sees won't make it into those answers either.

The crawlers you're optimizing for

A handful of named bots decide whether your content even gets read:

• OAI-SearchBot and GPTBot. OpenAI's crawlers. According to OpenAI's documentation, OAI-SearchBot is for search in ChatGPT, while GPTBot collects content that may be used to train models. The settings are independent: you can allow OAI-SearchBot to appear in answers and disallow GPTBot.
• PerplexityBot. Perplexity's crawler. According to the company's documentation, it surfaces and links websites in Perplexity answers and is not used to train foundation models.
• Google-Extended. A separate robots.txt token Google provides to control Gemini model training and the use of content in Gemini. According to the documentation, it does not affect inclusion in regular Google Search.
• YandexBot. Yandex's regular crawler: without indexing in Yandex there is no Alice AI answer in Search that links to you. Check your site in Yandex Webmaster, not only in Google Search Console.
• CCBot (Common Crawl). An open web archive used to train many models; an indirect path.

Server logs filtered by these user agents tell you something SEO dashboards won't: whether the AI tools are even showing up. If OAI-SearchBot and PerplexityBot never hit your site, nothing downstream matters yet. If they're crawling and you're still not cited, the problem is further down the pipeline.

Make the page readable by a machine first

Most sites write for a person scanning top to bottom, with the key fact buried under three paragraphs of scene-setting. A retrieval model doesn't scan. It needs the answer close to the question.

Three changes do most of the work.

Answer the question in the first sentence of the relevant section. If a section is titled "Pricing," the first sentence under that heading should state a number or a range, not a sentence about your philosophy on fair pricing.

Use schema.org structured data where it genuinely applies. FAQPage, Product, Organization, LocalBusiness and Review markup give crawlers an explicit, machine-parseable version of what's already on the page. This doesn't trick anything into citing you; it removes ambiguity about what the page is claiming.

Keep a clean, current sitemap and don't block useful crawlers in robots.txt. Some sites, trying to block AI training, block OAI-SearchBot and PerplexityBot too with one blanket rule, and drop out of the answers along with it. Decide deliberately: allowing search and blocking training can be set separately.

For Kazakhstan, add a fourth: Russian and Kazakh versions of your pages with correct hreflang. People ask AI in both languages, and the assistant answers from whichever version it found.

A newer, still-informal convention worth knowing about is the llms.txt file, a plain-text summary of a site's key pages, proposed as a lightweight way to help AI systems find the right content fast. Adoption is uneven right now, so treat it as a nice-to-have, not a guaranteed lever.

Write the sentence that gets lifted

Technical plumbing gets you read. Content gets you quoted. The sentences that end up in an AI answer tend to share a shape:

• A specific number with context: "Setup takes 10 business days and includes migration of up to 5,000 records," not "fast and easy setup."
• A named system instead of a vague category: "integrates with 1C and Kaspi," not "integrates with popular systems."
• A direct comparison, stated plainly: "Plan A includes phone support; Plan B does not."

This is also where FAQ pages earn their keep, beyond the schema markup. A well-formed FAQ, real questions a buyer actually asks, answered in two or three sentences each, is close to the exact format a model needs to lift and attribute. Pad it with marketing language and the model has to do extra work to find the fact, which pushes it toward a competitor's cleaner page instead.

One caveat applies directly to this kind of content: if you publish comparison pages, "best of" lists or testimonials to get cited, state your own interest plainly. A comparison secretly written by one of the vendors it ranks first is a trust problem before it's an AEO problem.

Where citations actually come from

Neither ChatGPT nor Perplexity cites company websites exclusively. Third-party sources carry real weight:

A company with a thin review footprint and no presence where its buyers actually ask questions is handing those citation slots to competitors by default, no matter how clean its own website is. For a local business in Kazakhstan, start with the 2GIS listing: the right address, phone and hours, and replies to reviews.

Checking whether it's working

There's no equivalent yet of Google Search Console for AI citations, though third-party trackers have started filling the gap. In the absence of a single dashboard, three manual checks get you most of the way:

1. Run your 10 to 15 highest-intent buyer questions through ChatGPT, Perplexity and Alice AI monthly, in Russian and in Kazakh, and record whether your company appears, with what phrasing, and from which source.
2. Check server logs for OAI-SearchBot, GPTBot and PerplexityBot hits, and watch which pages they request most.
3. Track referral traffic from chatgpt.com and perplexity.ai in your analytics: small volume right now for most companies, but a leading indicator of whether the citations are turning into enquiries.

What we do

We don't sell AEO audits as a packaged product. What we do build, regularly, is the infrastructure this whole approach depends on: usable content management systems, structured data, and API integrations that make a site's facts extractable in the first place. It is part of our website development and of rebuilding existing sites.

One example from our own operation: in our outreach system, before writing anything to a company, a model reads its homepage and up to four internal pages and extracts one or two concrete sentences about what that business actually does. It only works when the page states something specific and plain. Vague copy produces a vague, useless output. That's the same filter an AI crawler applies to your own site when deciding what's worth quoting: specific beats polished every time.

On the content side, we built a casting platform from scratch with its own admin panel and a blog section designed for ongoing SEO publishing. We advised the client to keep publishing several articles a month: the same cadence that keeps content fresh enough for retrieval-based tools to keep pulling from it.

The technical rebuild most sites need before any of this works (structured headings, FAQ markup, a clean sitemap, pages that state facts instead of hedging) fits inside a fixed-price project. We quote it after reviewing the site, the price is fixed in the contract, and there is a six-month warranty. How to work out which kind of site a company needs is covered in a website for your business.

FAQ isn't optional, it's part of the build

Treating the FAQ section as an afterthought defeats the purpose of everything above it. It should be written with the same discipline as the rest of the page: real buyer questions, direct answers, FAQPage markup attached, and no marketing padding between the question and the answer. If the site has an AI agent, the same FAQ becomes part of its knowledge base; we covered how such a chat helps buyers in more B2B buyers now start with an AI chatbot.