What changes with AI search
On classic Google, ranking seventh still brings some clicks. In a ChatGPT or Gemini answer, you are either cited or you do not exist. The user asks a full question, such as which agency suits a mining company in Antofagasta, and gets an answer built from fragments of a few sources.
So the work changes focus: having a page that ranks is no longer enough. You need paragraphs that answer a question on their own, with verifiable data, and a company the model can identify without ambiguity.
What we measure on our own site
The figures above are the visits GA4 recorded from ChatGPT and Gemini over the last 90 days. It is a small volume next to Google, and we say so. What matters is quality: 24 of the sessions from ChatGPT had engagement, two out of three, because the user arrives with a recommendation already made.
One honest caveat: GA4 only sees people who click the link. The times a model mentions the company without the user clicking are not recorded by any public tool.
First, be readable
ChatGPT searches the web with its own crawler, OAI-SearchBot, and relies on the Bing index. Gemini answers from Google's index. The starting point is technical: allow those crawlers in robots.txt, verify the site in Bing Webmaster Tools, notify every change with IndexNow and serve content in HTML, not hidden behind JavaScript.
The llms.txt file adds order and costs little, but as of 2026 no major engine has confirmed using it to choose sources. We publish it as a complement, not as a strategy.
Then, be citable
A model cites what it can extract without context: a question as a heading, the answer in the first two lines and a figure with source and date. A generic marketing paragraph is not cited because it answers nothing.
On our service pages every block answers a real buying question, and every figure states the platform and the period. The same discipline that builds trust on Google is, in AI search, the difference between being a source or not.
Chile and Argentina are not answered the same way
Questions carry geography. A company serving Santiago and Mendoza needs pages that say clearly where it works, with cities, regions and specific industries, and organisation structured data that is consistent across the site and external profiles.
Entity consistency weighs more than it seems: same name, same description and same links on the site, Google Business Profile, LinkedIn and directories. If each profile tells a different story, the model does not know whom to recommend.
What weighs outside your site
Models lean on what others say about you: press mentions, comparisons, industry directories and reviews. An agency, a winery or an industrial supplier that only talks about itself on its own website is less likely to be cited than one that third parties mention.
That is why GEO does not end in the code. It includes media relations, presence in sector listings and real client reviews, with names.
Frequently asked questions
Can you pay to appear in ChatGPT?
ChatGPT's organic answers cannot be bought. What you work on is making the site crawlable, its answers citable and the company mentioned by third parties.
How do I know if ChatGPT recommends me?
In GA4 you can see visits arriving from chatgpt.com and gemini.google.com as a referral source. Mentions without a click are not recorded, so it is best to complement it with periodic question tests.
Does GEO replace SEO?
No. Gemini uses Google's index and ChatGPT uses Bing's. Without a base of technical SEO and solid content, there is nothing to cite you from.
Does the llms.txt file help?
It is a low-cost complement. As of 2026 no major engine has confirmed using it as a signal, so it does not replace citable content or external mentions.
The technical step by step, with robots.txt, Bing and IndexNow, is in the guide GEO: how to get AI to cite your brand.
Source: Google Analytics 4 for estudiofuria.com, last 90 days, checked on 9 Nov 2026. Mentions without a click cannot be measured with public tools.
