GEO Fact and Fiction.

AI engines find and name a business by reading the open web the moment someone asks, then assembling an answer from the clearest and most corroborated sources they find. No single file or schema controls it. What earns a mention is being described plainly, structured well, and saying the same thing across many places at once.

That is good news. It means AI visibility is something you can architect, not a lottery you enter and hope. The businesses that show up are rarely the loudest. They are the clearest. Once you can tell the fact from the fiction about how these engines actually work, the next move gets obvious.

What actually decides whether AI names your business

Start with how much of this is already real. Google's AI Overviews reach more than two billion people a month, up from 1.5 billion only weeks earlier, and ChatGPT serves over 800 million weekly active users. When that many people ask an engine to recommend something in your category, the answer it returns becomes a discovery surface in its own right.

Here is the mechanic underneath. When someone asks a modern engine about a category, it does not simply recite what it memorized during training. ChatGPT, Gemini, Google AI Overviews, Copilot, and Claude all retrieve live information from the web at the moment of the question, read the strongest sources, and synthesize them into one composed answer. Your business gets named when your description is clear enough to lift and consistent enough to trust. So the real question is not whether to play. It is which moves genuinely matter, and which are folklore.

Does an llms.txt file work? The honest answer

One idea has traveled fast: add a file called llms.txt to your site, write a tidy summary of who you are, and AI engines will read it. It is an elegant proposal. The honest picture in 2026 is that the major engines do not use it yet.

Ahrefs looked at roughly 38,000 sites that publish a valid llms.txt and found that 97 percent of them received zero requests for the file. Google has been direct about it. Search advocate John Mueller put it plainly: "AFAIK none of the AI services have said they're using LLMs.TXT," and compared it to the old keywords meta tag, a signal engines politely set aside. Adoption sits near one in ten sites, which tells you the conversation is louder than the usage.

This is not a reason for frustration. It is a reason for focus. The minutes you might spend hand-tuning a file no crawler reads are better invested in making the pages crawlers actually do read clearer. llms.txt has a genuine home in agent-to-agent and developer-documentation use, where a tool wants a compact map of a site. As a lever for a business hoping to be recommended by name, it is not the one doing the work.

Do FAQ pages still work? Read the fine print

FAQs are having an identity moment. In May 2026, Google stopped showing FAQ rich results in search, and it is retiring the related reports and tests through the summer. It is easy to read that as "FAQs are over." The detail that matters points the other way.

Google confirmed that FAQPage structured data is still valid and still parsed. The visible search feature went away. The content kept its value. As one widely shared way of putting it goes, the change made something obvious that was always true: the schema was never doing the work, the content always was.

And the content has rarely been more useful. People prompt AI engines in full questions. A page that answers real questions plainly, in the words a buyer would actually use, maps almost exactly onto how someone talks to an assistant. A good FAQ is not a markup trick. It is a clear answer sitting where an engine can find it. That still earns citations, with or without the rich result.

How ChatGPT, Claude, Gemini, AI Overviews, and Copilot differ

Treating "AI search" as one thing is the most common simplification, and it quietly costs visibility. The engines retrieve and cite differently enough that the same business can be named by one and not yet surfaced by another. The brands that do well treat them as related but distinct readers.

A 2025 Yext analysis studied 6.8 million citations across 1.6 million AI responses and found clear personalities. Gemini pulled about 52 percent of its citations from brand-owned websites, rewarding clear, structured pages on your own domain. ChatGPT pulled nearly 49 percent from third-party directories and listings. Same question, different idea of what counts as a trustworthy source.

The engines are not one reader. They are five readers with slightly different reading lists, all arriving at the same shelf. Clarity and consistency are what put you on it.

How GEO is different from SEO

GEO and SEO are cousins, not rivals. Both start from the same place: content the open web can read. They reward different finishes.

Classic search optimization earns a ranked link and a click. Generative engine optimization earns a mention inside a composed answer, where there may be no list of ten blue links at all. The Princeton-led research that named the field tested this directly across 10,000 queries. Adding concrete statistics, citing sources, and including relevant quotations raised a page's visibility in AI answers by up to roughly 40 percent. Old-school keyword stuffing did close to nothing.

Here is the part that reframes the work. In that research, a page's traditional rank position was a weak predictor of whether an engine cited it. A clearly written, well-sourced passage on a page ranked fifth can be lifted ahead of the page ranked first. Structure and substance travel into AI answers in a way that raw ranking does not. That is an opening for any business willing to write clearly, whether or not it already sits at the very top of the results.

How to know what to do next as the ground keeps shifting

The one constant in AI visibility is motion. Citation patterns move month to month. LinkedIn, for one, climbed from around eleventh to fifth among ChatGPT's most-cited domains in a single quarter. Anyone selling a fixed checklist of AI tricks is selling a snapshot of a moving picture.

The way to stay steady is to optimize for principles, not for this week's tactic. Three durable moves hold up no matter which engine is ascendant.

None of this asks you to chase every platform or rewrite your site every time a new acronym trends. It asks you to be clear, to be consistent, and to look at the real picture. That is a far calmer way to grow than reacting to every rumor about what the engines want this week.

The bottom line

The story of GEO is calmer than the hype suggests. The engines are not mysterious, and they are not waiting for a secret file. They read the open web, reward clarity, and trust consistency. Tell a clear, steady story across the places that matter, measure where you land, and let it compound.

The businesses that do this are not chasing every change. They are architecting the kind of presence that every engine arrives at on its own. That is the foundation, and it holds no matter how the acronyms evolve next.

Frequently asked questions

Does an llms.txt file help AI find or recommend my business?

Not in any meaningful way yet. As of 2026, no major AI engine officially reads llms.txt. Ahrefs found that 97 percent of sites publishing the file received zero requests for it, and Google has said it has no plans to support it. The file has real uses for agent-to-agent and developer-documentation purposes, but as a way to get a business recommended by AI, your effort is better spent on clear, well-structured pages that engines already crawl.

Do FAQ pages and FAQ schema still work for AI search?

Yes, as content. Google stopped showing FAQ rich results in search in May 2026, but it still parses FAQPage structured data, and the underlying value was always the content rather than the markup. People prompt AI engines in full questions, so a page that answers real questions plainly maps almost exactly onto how someone asks an assistant. A clear FAQ still earns citations, with or without the rich result.

How do ChatGPT, Claude, Gemini, Google AI Overviews, and Copilot each find and cite businesses?

They retrieve live information from the web and synthesize an answer, but they favor different sources. A Yext study of 6.8 million citations found Gemini pulled about 52 percent of its citations from brand-owned sites, while ChatGPT pulled nearly 49 percent from third-party directories. Google AI Overviews draw on Google's index, Copilot on Bing, ChatGPT and Claude run live web searches, and Gemini grounds in Google Search. Being described consistently across both your own site and trusted third-party places helps every engine reach the same conclusion.

How is GEO different from SEO?

SEO earns a ranked link and a click. GEO, or generative engine optimization, earns a mention inside an AI-composed answer where there may be no list of links at all. Princeton-led research found that adding statistics, citing sources, and including quotations raised AI visibility by up to about 40 percent, while keyword stuffing did almost nothing. A page's traditional rank was a weak predictor of whether AI cited it, so clear, well-sourced writing can be lifted into an answer even from a lower ranking position.

How can a business know what to do next for AI visibility?

Optimize for principles rather than this week's tactic, because citation patterns shift constantly. Three durable moves hold up: write the clearest possible description of what you do, keep that description consistent across your site and the third-party places people discuss your category, and measure where the major engines actually describe and cite you so your next step is based on real data rather than rumor.