EN
Ihor Nikolenko- How an Assistant Decides Who to Cite
- Step 1: Become Visible to Everything That Searches on Your Behalf
- Step 2: Write So There's Something to Quote
- Step 3: Be the Same Entity Everywhere
- How GEO Differs from Regular SEO
- Step 4: Measure Honestly
- Where to Start If You're Short on Time
- The Short Version
- Frequently Asked Questions
More and more often, a customer walks in saying "ChatGPT recommended you." Or they don't walk in at all — because it recommended a competitor instead. Search no longer ends with ten blue links: a person asks an assistant a question, gets a ready-made answer with two or three sources, and looks no further. If you're not among those sources, you don't exist for that person.
This shift is happening right now, and it has one nice property: the rules aren't set in stone yet. Big brands have an edge in traditional search thanks to years of accumulated backlinks, but assistants look at something different — whether your page can be quoted as an answer. This is where a small or midsize business can outmaneuver bigger players, if it understands exactly how a model decides who to name.
I've already written separately about the llms.txt file. This article covers the other side of GEO (Generative Engine Optimization): how assistants find pages, why some get cited and others get ignored, and what to change on your site to land in an answer. No magic, just mechanisms you can verify.
How an Assistant Decides Who to Cite
A language model, on its own, doesn't "know" anything about your site in real time. When you ask ChatGPT, Claude, Perplexity, or Copilot about something specific — prices, companies, recent events — the assistant first runs a search, gets a list of pages, reads excerpts from them, and only then composes an answer, citing sources. It's the same principle as regular search, just with a different final step: synthesis instead of a list.
The first practical takeaway follows from that: to get cited, you first have to be found. Search in ChatGPT and Copilot relies on the Bing index, Perplexity and Anthropic run their own crawlers, and Google uses its own index for AI Overviews and AI Mode. A site that isn't in the Bing index is invisible to some assistants, no matter how well it ranks in Google.
The second takeaway: out of a dozen pages it finds, the model picks the ones from which it's easiest to pull a clear answer. A page where the answer to "how much does it cost" is buried in the fifth paragraph, after the company history, loses to a page where the price and terms sit right under a heading asking the same question. A model doesn't read between the lines — it looks for the lines.
Step 1: Become Visible to Everything That Searches on Your Behalf
Start with Bing Webmaster Tools. It's free, setup is similar to Google Search Console, and you can import data from it directly. Submit your sitemap and check the indexing of your key pages — in Bing specifically, not just Google.
Next is IndexNow, an open protocol that lets a site notify search engines about new and changed pages itself, instead of waiting for a crawler to come back around. It's supported by Bing, Yandex, Naver, Seznam, and others; Google hasn't joined the protocol yet. For a site that updates content weekly, that's the difference between "we were seen today" and "we were seen eventually."
Now check your robots.txt. Assistant crawlers have their own names: GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot from Perplexity, Google-Extended from Google for Gemini. A lot of off-the-shelf robots.txt templates, and even some CDN settings, block them "just in case." If you want to show up in answers, these bots need to see at least your public pages. Deciding who to let in is your call, but it should be a deliberate choice, not something inherited from a template.
And a separate note on contact information: check what your site looks like without JavaScript. Some protection services inject your phone number and email via a script, which leaves the contact page empty for a crawler. An assistant that can't see how to reach you won't recommend you to anyone.
Step 2: Write So There's Something to Quote
Assistants like three things, and all three are human-friendly too.
Answer first. Under every question-style heading, the first sentence should be a direct answer. Explanation, nuance, and examples come after. It's a format a model can lift wholesale, and a person can read in a second.
Specifics instead of adjectives. "Reliable and high-quality" can't be quoted. "Same-day callout, a 12-month workmanship guarantee under contract" can. Every verifiable detail on the page is a potential citation.
Headings that match how the question is phrased. People ask assistants in conversational language: "what's the difference between X and Y," "what's better for…," "how much does… cost." If your H2 sounds the same, your odds of landing in the answer go up. A "Questions and Answers" section on service pages is the simplest way to get headings like that.
Comparison tables, step-by-step lists, term definitions — anything with clear structure gets pulled out readily by the model. Long, padded paragraphs get skipped, just like a human would skip them.
Step 3: Be the Same Entity Everywhere
A model doesn't build its picture of you from one page — it draws on everything out there: your website, your Google listing, social profiles, directories, media mentions, reviews. When your company name, address, phone number, and description of what you do match everywhere, the model is confident it's dealing with one entity, and names it more readily. When the website has one name, the listing has another, and the directory has an old address, that confidence drops.
Structured data is a way to tell the machine directly who you are. Organization markup with your name, logo, contacts, and links to your profiles; Product for goods; Article for posts; FAQPage for answers; BreadcrumbList for navigation. This isn't about "rich snippets in Google" — Google has stopped showing many of those long ago. It's about giving an assistant's crawler your facts in a format it doesn't have to guess at.
Mentions on third-party resources — industry directories, publications, partner pages, podcasts — work as corroboration. Not because they pass "authority" the way links do in classic SEO, but because a model that sees you in several independent places treats you as a real player in the topic.
How GEO Differs from Regular SEO
It's not a replacement or an opposite — more a shift in emphasis. Here's how I explain it to clients:
| Question | Classic SEO | GEO |
|---|---|---|
| What counts as a result | A ranking position and a click | A mention in the answer with a source link |
| Who's the competition | Ten sites on page one | The two or three sources the model chose to cite |
| What matters most | Relevance, links, behavior | Clarity of the answer, verifiable facts, entity consistency |
| What's the unit of optimization | A page targeting a keyword | A snippet targeting a specific question |
| Where to get indexed | Google and Bing, plus assistant crawlers | |
| How to measure | Rankings, clicks, conversions | Assistant referral traffic, mentions in answers, "an AI recommended us" leads |
The good news: everything that makes a page easier for an assistant also makes it better for a human and for Google. None of the effort goes to waste.
Step 4: Measure Honestly
It's easy to fool yourself here. "I asked ChatGPT and it named us" isn't a measurement, because the answer will be different next time. Here's what you can actually track:
- Referral traffic from assistant domains in your analytics: chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai, and others as traffic sources. Set up a dedicated segment for them and watch which pages they open.
- A "How did you hear about us" field on your form or in your first-call script. "An AI assistant recommended you" should be on the list — otherwise you'll never know the channel is working.
- A regular check on a fixed set of questions. Once a month, ask assistants the same ten questions from your niche and record whether you show up, and who shows up instead of you. It's not a precise metric, but the trend is visible.
An indicator should show fact, not intent. If there's no referral traffic from assistants, it isn't working yet — and the fix is to go back to the steps above, not to reach for new theories.
Where to Start If You're Short on Time
If you have one evening, do three things: connect Bing Webmaster Tools and submit your sitemap, check robots.txt for blocked assistant bots, and rewrite one "Questions and Answers" section on your main service page in the "answer-first sentence" format. That alone changes your visibility.
If you're building or migrating a site, it's easier to use a platform where this is built in from the start: structured data, a sitemap, a file for assistants, IndexNow support, contacts in plain HTML. That's how the NEO platform is built — not as an "SEO module," but as a way to stop thinking about the basics every single time. And if you want to figure out exactly where assistants are already answering questions in your niche, and whose answers they're using, that's one of the things I dig into during a consultation with a 30-day plan.
The Short Version
- Assistants search first and synthesize second: to get cited, you need to be found — in Bing too.
- Bing Webmaster Tools, IndexNow, and a robots.txt open to assistant crawlers are the technical foundation of GEO.
- A model quotes snippets with a direct answer, specific details, and question-style headings.
- Consistent name, address, phone number, and description across every platform builds a model's confidence in you as an entity.
- Schema.org structured data is a way to hand facts to a machine without it having to guess.
- Measure referral traffic from assistant domains and add an "AI recommended us" option to your form.
- Everything good for an assistant is good for a person and for Google — the effort isn't duplicated.
Frequently Asked Questions
Should I block AI bots so they don't "steal" my content?
This is a trade-off. Blocking protects your text from being used to train models, but it also cuts you out of answer citations. For most businesses selling products and services, visibility is worth more than the text itself.
Do I need to write separate pages "for AI"?
No. Separate pages for machines just leads to duplication and lower quality. Pages properly written for people are already suitable for assistants.
How quickly does this show results?
Technical steps — indexing in Bing, IndexNow — pay off fast, often after the very next crawl. Actually appearing in answers depends on competition in your topic and how well your pages truly answer the question; that's measured in weeks and months, not days.
Does GEO replace regular SEO?
No. Assistants build their answers on the same underlying indexes. A site without a solid classic-SEO foundation won't make it into the answers either.
