Google Rank Is Not Enough - ChatGPT Has to Cite You Before the Customer Clicks
Buyers now ask ChatGPT, Perplexity, and Google AI Overviews who to hire. Ranking on Google does not put your name in that answer. What GEO actually is, why a JS-only site is invisible, and what I put on a page so a model can cite it.
A buyer in 2026 often never sees page two. They ask ChatGPT, Perplexity, or a Google AI Overview: who builds a booking site, who takes payments in Telegram, who can fix a slow homepage. They pick from the names the model cites. Ranking tenth on Google with a pretty landing does not put you in that sentence.
Generative Engine Optimization - GEO - is that work: become the source the model quotes, not only the blue link it used to rank. I already wrote what SEO is, why a one-page site beats a chat profile, and why site search closes orders. This is the next layer. The site has to ship as HTML, answer in the first screen, and stay dated enough that a crawler trusts it.
1. What actually changed in 2026
Search did not die. The default surface for a buying question moved. Google still ranks pages. On top of that ranking, an AI Overview now answers a large share of queries in one block. ChatGPT, after it started placing clickable brand links inside answers, stopped being a dead end and started sending referral traffic. Perplexity already behaved like a citation engine. The buyer’s first shortlist is often that block, not ten blue links.
- SEO still matters: if nothing can retrieve your page, nothing can cite it.
- GEO is the extra job: the page must be extractable - a direct answer, facts, a date, a named service - so the model has a sentence to steal.
- A lot of GEO winners sit outside Google’s top ten. Authority for a model is not the same as a blue-link rank.
2. Why a JS-only site is invisible to the model
Most AI crawlers fetch HTML. They do not sit around waiting for your React bundle to paint the prices. If the real copy lives only after JavaScript, the model sees a shell. Cookie walls, “accept marketing” gates, and infinite client-side galleries do the same damage. A WordPress theme that dumps a megabyte of widgets before the H1 is almost as bad.
This is one reason I still start marketing sites in Next.js with server-rendered pages, not a client-only SPA and not a page builder. The crawler gets the same sentence the human gets. Schema markup on top of that is a hint, not a magic trick - FAQ, Article, LocalBusiness - but the sentence has to exist in the HTML first.
3. What I actually put on a page that models can steal
I write the page as if the first eighty words will be quoted out of context. That is usually what happens. Then I give the model handles: question headings, a short list, one number with a date, a named offer. I do not stuff “ChatGPT” into the copy hoping to rank inside ChatGPT. That is cargo cult.
- Answer first: one dense paragraph at the top that a stranger could paste as the whole article.
- H2s as questions people actually type: “Do I need a website if I have Instagram?” not “Our unique value proposition.”
- One original number or a dated claim. Models lift statistics. Empty adjectives they skip.
- Author, date, and a service a crawler can name. “Iryna Vynnychenko, senior web engineer, Next.js and Telegram Mini Apps” is an entity. “Creative digital solutions” is fog.
- Let the relevant bots in. Blocking GPTBot while wanting ChatGPT citations is a contradiction. Training and search crawlers are not the same flag - set them on purpose.
4. What I would not do this month
I would not buy a “get cited in ChatGPT” agency package that rewrites the homepage into keyword soup. I would not invent statistics. I would not hide the offer behind a lead magnet the crawler cannot read. I would not treat GEO as a replacement for a site that still loads in eight seconds - Core Web Vitals still decide whether a human stays after the citation.
Conclusion: own the sentence, not the trick
GEO is not a new CMS. It is a site a crawler can read, a first paragraph a model can lift, and a named person behind a named service. If ChatGPT already answers your category and your name is missing, the gap is usually the page - JavaScript shell, no facts, no date - not a missing plugin. I build those pages in Next.js with schema, answer-first copy, and a form that still works when the AI referral lands. Write via the form with the question customers already type into ChatGPT. We will put a page on the internet that can actually be cited.
Ready to discuss your project?
I'm a senior web engineer specializing in React and Next.js - available for freelance projects worldwide.
Location
Kyiv, Ukraine
Upwork
View ProfileTelegram
Contact meViber
Contact me