Where do I start with GEO if nothing is done yet?

With the index, not with trendy files. Check how many of your pages Bing, Yandex and Google see - it's three minutes in their consoles. Neural nets cite what sits in search: Copilot assembles answers from the Bing index, Alice works on top of Yandex results. No pages in the index - nothing to cite, and no markup will change that.

Useful, but not primary. I tested it on myself: on a fresh site llms.txt was in place from month one, and citations were zero - because there was no base of pages and links. The file helps the machine quickly grasp who you are, what your services and numbers are. Add it once there's something to describe, and don't expect a standalone effect from it.

How do I know GEO is working?

Two free native reports. Bing Webmaster Tools - AI Performance: when and for which phrases the site landed in Copilot's sources. Yandex Webmaster - 'Site visibility in Alice AI': Share of Voice and the specific queries where you're in the answer's sources. Plus a manual check once a month: ask ChatGPT and Perplexity about your topic and see who they name.

How is GEO content different from regular SEO content?

It has to be easy for a machine to retell. Specifics instead of promises: prices, timelines, case numbers right in the text and headings. My service pages with the price in the title land in Alice's sources. Plus FAQ blocks with direct answers and short factual paragraphs that can be pulled out of context without losing meaning.

Can I buy GEO as a service and get citations fast?

If someone sells you 'AI citations' that bypass the index and links - it doesn't work. Citations grow from the same base as regular search, and at the same speed. Standalone 'premium GEO' tools are needed less often than they're sold; the native Bing and Yandex reports are free.

GEO checklist: how to get cited by AI in 2026

I’ve written about GEO twice already: a numbers-based breakdown of two projects and Yandex’s new Alice AI report. Both articles end with the same conclusion: there is no magic, there is a system. This article is that system, assembled into a single checklist. Print it, walk the items, close the gaps.

Every item is tested on my projects: a cleaning site I run marketing for got cited by Copilot 40 times in six weeks, and it lands in Alice’s sources in 90% of the cases where it’s visible in the results. The reverse experience is mine too: a fresh site with llms.txt but no base - zero citations.

In short

GEO stands on four blocks: base → machine readability → content → measurement.

Order matters: markup without an index doesn’t work, I tested that on myself.

Copilot cites from the Bing index, Alice works on top of Yandex results. No rankings - no citations.

Measure for free: Bing AI Performance + ‘Site visibility in Alice AI’ in Yandex Webmaster.

Block 1. The base: index and rankings

A neural net doesn’t crawl the web at the moment it answers - it assembles sources from the search index. So the first three items are plain SEO, and you can’t skip them.

☐ 1. Pages in the index of all three search engines. Not just Yandex. Copilot pulls sources from Bing - and Bing loses sites easily: it once kept 1-3 of my 77 pages with zero crawl errors. Check the consoles: Yandex Webmaster, Google Search Console, Bing Webmaster Tools. Three minutes, and you instantly see which AIs you even exist for.

☐ 2. Rankings for your queries. Alice takes pages that rank high into its sources - Yandex says so straight in the report’s help docs. In the sources of its answers I see the same pages that sit in the regular top. No rankings even for niche queries - start there, not with GEO.

☐ 3. Pages for real questions. Not “about us” but pages built for demand: service + geo, service + niche, a specific client question. On the cleaning project, the programmatic service pages land in Alice’s sources - each built for its own query.

Block 2. Machine readability

The machine has to understand who you are without guessing. This is tuning, not construction: done once and updated as things change.

☐ 4. AI crawlers aren’t blocked. Check robots.txt: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended must have access. For a services site AI visibility means leads; blocking crawlers makes sense for media that sell content, not for you.

☐ 5. Schema markup. The minimum: Person or Organization with identifiers, Service for services, FAQPage for questions, BreadcrumbList. The machine cross-checks entities - give it structure, not just text.

☐ 6. llms.txt. A file in the site root with facts: who, what you do, niches, prices, cases with numbers, contacts. An honest caveat from my own experience: on its own it doesn’t produce citations - on the fresh site the file was in place and the effect was zero until the base appeared. But once the base is there, it helps the machine gather the facts fast and without errors.

☐ 7. Clean HTML. Content must be in the page source, not loaded by scripts: most AI crawlers don’t execute JS. A one-minute check: open the page source (Ctrl+U) and search for your key text. Found it - you’re fine.

Block 3. Content built to be retold

A neural net doesn’t quote - it retells. The content that wins is the content that survives retelling without losses.

☐ 8. Specifics in the text and headings. Prices, timelines, case numbers. My pages with the price right in the title land in Alice’s sources. “Market leader with an individual approach” is something a machine can’t retell - there are no facts in it.

☐ 9. FAQ blocks with direct answers. The question as the client asks it, the answer in 2-4 sentences with specifics. That’s schema (FAQPage) and ready-made passages for a generative answer at the same time.

☐ 10. Answer blocks. Short self-contained paragraphs that can be pulled out of context: who you are, what you do, for whom, for how much. If a paragraph makes sense without the rest of the page, the machine will take it into the answer.

Block 4. Measurement

☐ 11. Native reports once a month. Bing Webmaster Tools → AI Performance: Copilot citations by phrase. Yandex Webmaster → ‘Performance’ → ‘Site visibility in Alice AI’: Share of Voice, queries, a ‘Sites’ tab with competitors. Both are free. How to read the Alice report - I broke that down separately, including the honest caveats about the calculation base.

☐ 12. A manual check of your topic. Ask ChatGPT, Perplexity and Alice: “recommend a contractor for X”, “how much does Y cost”. Write down who gets named and who gets linked. The answers shift with the index - so monthly, not once.

What’s not in the checklist - on purpose

Paid “GEO tools.” Everything measurable here is free. If someone sells “AI citations at a premium” that bypass the index, it doesn’t work: citations grow from the same base as search.

Separate “content for neural nets.” I don’t have a single special page “for Alice”. What gets cited are ordinary service pages made for people and demand.

Fast timelines. GEO inherits SEO’s speed: index first, then rankings, then citations. On the cleaning project, between the systematic work and the 40 Copilot citations lie months of base, not weeks of magic.

How to use this

Walk the blocks in order. If block 1 fails, blocks 2-3 give you nothing - I tested that on my own site. If the base is there and the citations aren’t, look at block 2: most often the machine simply can’t parse who you are. If that’s fine too, strengthen block 3 on the pages already in the top: they’re one step away from the sources.

Want me to walk this checklist for you - a one-off audit is exactly that. Quick questions - Telegram @dimik90.