Getting found in ChatGPT and Perplexity: how AI visibility works

A business owner needs someone to build an app. A few years ago they would have Googled it and clicked through some links. Now they ask ChatGPT: "Which app builder in Tilburg should I call?" They get a short list of three or four names, with a line of explanation for each. If you are not on that list, they never saw you. There is no second page to click through to.

That, in a nutshell, is what AI visibility is about. You want AI assistants like ChatGPT, Perplexity and the AI answers in Google to know your business, understand what you do, and mention you when it fits. You will also hear a new acronym for it more and more: GEO, generative engine optimization. In this piece we explain how it works, what you can do about it yourself, and what you can't. And since we do this for our own site too, we use baboons.nl as the example.

What AI visibility is, and what GEO adds

An AI assistant pulls its knowledge from two places. Part lives inside the model, learned from huge amounts of web text. The rest it looks up when you ask, by searching the web and reading a few pages. Perplexity does that for almost every question, and ChatGPT does it more and more often.

GEO is about that second part, and a little about the first. You make sure your pages are easy to find, easy to read and easy to cite. An AI assistant reading your site should be able to work out within seconds who you are, where you are based and what you are good at.

Sounds like SEO? It is closely related, and a findable website is still the foundation. The difference is who is reading: not a search engine lining up links, but a model that summarizes your text and turns it into an answer. That model likes clear facts and sources that agree with each other.

First, an honest note: nobody controls the answers

Let's say it right away. You cannot force ChatGPT to mention you. Nobody can, including any agency that promises it. The models keep changing, and how they choose their sources is largely unknown.

What you can do is lower the threshold as far as possible. Make sure there is clear, consistent information about your business out there, in a form a machine can read easily. That raises the odds you end up in the answer. Nothing more, and nothing less.

How we set it up on baboons.nl

Here is our own setup. You can look at all of it yourself.

An llms.txt with a curated table of contents

At baboons.nl/llms.txt there is a short text file written for language models. At the top is one line about who we are: a product studio in Tilburg building apps, websites, web applications and AI systems for SMEs and start-ups. Below that is a list of our most important pages: the service pages, the city pages and every blog post, each with a short description.

To be fair: llms.txt is an emerging convention, not an official standard. There is no guarantee every AI assistant reads it. But it costs very little, and when a model does read it, it gets a clean overview of your site in one go.

A Markdown twin of every key page

Every link in that llms.txt points to a Markdown version of the page. So next to /webapplicatie-laten-maken/ there is also a webapplicatie-laten-maken.md, and every blog post has one as well. It is the same text, but without the menu, footer, buttons and styling. A model doesn't have to dig through the whole page to find the content.

The important detail: we don't write those Markdown files by hand. They are generated on every build from exactly the same source as the page itself. Change a piece of copy, and the .md version changes with it. So they can never drift apart. For now we only do this for the Dutch version of the site.

Structured data on every page

The code of every page carries structured data, in a format called JSON-LD. It is information your visitor doesn't see, but that tells a machine exactly what is on the page. On our site that covers:

• Organization and LocalBusiness: our name, our address in Tilburg, phone number, email and our LinkedIn page.
• Person: the author of the blog posts, with job title and LinkedIn profile.
• BlogPosting: for every post, the title, description, publication date and who wrote it.
• FAQPage: the frequently asked questions at the bottom of a page, as separate question and answer pairs.
• BreadcrumbList: where the page sits in the site, like Home, Blog, this post.

One source for who we are

Our name, address, phone number and LinkedIn live in one place in the code. Both the visible block on the service pages and the structured data read from it. If something changes, you update it once and it is right everywhere. That sounds like a technical detail, but it is the heart of what we call entity consistency below.

A sitemap and an open door

Our sitemap.xml lists every public page, with the Dutch and English versions linked to each other through hreflang. And robots.txt leaves the door open: crawlers are allowed to read the whole site. If you want AI assistants to find you, don't block them by accident.

What you can do yourself: become citable

Technology helps, but in the end an AI assistant quotes text. So the content has to pull its weight too.

Give concrete answers. A model looks for a sentence that answers the question. "A first working version is usually ready in 6 to 8 weeks" is citable. "We deliver fast, flexible custom solutions" is not.

Use clear headings. Put the question your customer asks in a heading, and answer it straight away in the paragraph below. That reads well for people and for machines.

Add an FAQ where it fits. Real questions customers ask you, with short, honest answers. Not questions you invent to squeeze in a keyword.

Be a source others mention. AI assistants take into account what other sites say about you. Mentions in industry listings, articles, interviews and partner pages help. It is slow work, but it counts.

Put a date on your content. A model wants to know whether information is current. A blog post with a clear publication date is more trustworthy than text without one.

That is how we wrote RAG explained, for example: concrete steps, short answers, clear headings.

Entity consistency: the same you, everywhere

An AI assistant sees your business as an entity: a thing with a name, an address, a description and a few properties. It builds that entity from everything it comes across about you. Your website, your LinkedIn page, your Google Business Profile, listings in directories.

When those sources contradict each other, the picture gets blurry. Are you "Baboons" in one place and "Baboons Digital" in another, is there an old address in a directory, do you call yourself a web agency on LinkedIn and a product studio on your site? Then the model has to choose, and sometimes it chooses wrong or skips you altogether.

So go through it once:

• Is your business name exactly the same everywhere?
• Are your address and phone number right in every directory and profile?
• Do you describe what you do in roughly the same sentence everywhere?
• Does your website link to your LinkedIn, and your LinkedIn back to your website?

Not glamorous, but something you do properly once and then keep up to date.

What it is not

GEO is not a trick to fool AI assistants. Hidden text, keyword lists or fake mentions don't work and can hurt you.

It is also not a replacement for SEO. Many AI assistants search through regular search engines. If you don't rank well there, they will find you less easily too.

And be open about your own use of AI. If you run a chatbot or AI-written content on your site, the AI Act asks for transparency. What that means in practice is in our EU AI Act checklist.

Where do you start?

Only have an afternoon? Check your robots.txt, make sure your name and address match everywhere, and give your key pages a clear heading with a concrete answer.

Want to go further, with structured data, an llms.txt and Markdown versions that update along with your site? Then it lives in how your website or web application itself is built. That is how we set it up too. And if you are also considering your own AI assistant on your site, read ChatGPT wrapper or custom first.

Want to look at your site together?

Curious how your site looks to AI assistants? Book a call. We will look at where you stand together and what is worth the effort. Sometimes the answer is that you are already in good shape, and then we will tell you so.

Frequently asked questions

What is the difference between SEO and GEO?

SEO aims for a spot in the search results of, for example, Google. GEO aims to get mentioned in the answers of AI assistants like ChatGPT and Perplexity. The foundation is largely the same: a good, findable site with clear content. GEO puts extra weight on citable answers, structured data and consistent information about your business.

Should I put an llms.txt on my site?

It does no harm and costs little. Just know that it is an emerging convention and not an official standard, so there is no guarantee every AI assistant uses the file. See it as an extra, not as the foundation.

Can I make sure ChatGPT recommends my business?

No, nobody can guarantee that. You can raise the odds by publishing clear, consistent and easy-to-read information about your business, on your own site and in other places.

Should I block AI crawlers or let them in?

That is a choice. If you block them, AI assistants can't read your site and you will be mentioned less often. If you want to be found, let them in and check that your robots.txt isn't holding anything back by accident.