What makes ChatGPT, Perplexity and Google's AI answers cite a company, what you control on your own website, and what has to happen elsewhere. No hype.
A client asked me recently why a competitor "comes up in ChatGPT" and they don't. It's becoming one of the most common questions I hear from B2B companies, and the answers online swing between "add this one file" and "it's all magic". Neither is true.
Here is what I know from building sites that AI assistants actually read, and from watching the referral data afterwards.
Why this matters now
Buyers increasingly ask an assistant before they ask Google: "Which fire-safety contractors work in the Netherlands?", "Who builds real estate websites in Montenegro?", "What does a custom booking system cost?". The assistant answers with a short list and a few sources. If you're not on that list, you never get the visit.
It isn't theoretical. On Nekretnina.me, a real estate catalogue we launched in September 2026, 16% of sessions in the first 20 days came from chatgpt.com, behind only Google and direct visits. That site was built from day one to be read by language models. Most sites aren't.
How AI assistants decide who to mention
Simplified, an assistant does two things. It draws on what it already learned during training, and, for many questions, it searches the web in real time and reads a handful of pages before answering. Both depend on the same thing: can the model find consistent, clear, trustworthy information about you?
That gives you two jobs. Make your own site easy to read and quote. Make sure other trustworthy sites say the same thing about you.
Part 1: what you control on your own website
Serve full HTML
Many AI crawlers don't run JavaScript the way a browser does. If your site is a client-side app that sends an almost empty page and builds the content in the browser, an AI crawler may see nothing. Server-rendered or prebuilt pages, as a Next.js site delivers by default, send the complete text in the first response. This is the single most common technical reason a good company is invisible to AI.
Answer questions directly
Assistants quote passages that answer a question in one or two sentences. Write the way you'd answer a buyer on the phone: the answer first, then the detail. "We design fire-safety systems for commercial buildings in the Netherlands, certified under..." beats "Welcome to our world of excellence."
Every important page should open with a plain definition of what you do, for whom and where. FAQ sections work for the same reason: question, direct answer, specifics.
Be specific and consistent
Vague claims don't get quoted. Concrete facts do: what you do, where, since when, for which sectors, at what starting price, with which certifications. And they must be the same everywhere: your website, your Google Business Profile, your LinkedIn page, your directory listings. Contradictions make a model less confident about you.
Add structured data
Schema markup tells machines what your page means: this is an organisation, this is its address, these are its services, this is a FAQ, this person is the founder. It's invisible to visitors and very useful to search engines and assistants. Google documents the types it supports (Google Search Central: structured data).
Publish an llms.txt file
An llms.txt file is a short, plain-text summary of your company and links to your key pages, placed at the root of your domain. It's a proposed convention, not a standard every assistant follows. I add one to every site because it takes an hour and does no harm; I don't pretend it's a ranking switch.
Don't block the crawlers you want
Check your robots.txt and your firewall. Some hosting setups and bot-protection tools block AI crawlers by default. If you want to appear in ChatGPT search, OpenAI's search crawler has to be able to reach your pages (OpenAI: crawlers).
Part 2: what has to happen elsewhere
This is the part most "GEO checklists" skip, and it's usually the bigger half.
Reviews
Recent, specific reviews on Google and relevant platforms give assistants concrete language about what you do well. "They rebuilt our booking system and the WhatsApp reminders work" is far more useful to a model than five stars with no text. Ask real clients, every time a project ends well.
Directories and listings
For many service questions, assistants lean on directories and "best of" lists. For agencies that means Clutch, TechBehemoths, DesignRush and similar; for your sector it might be trade associations, certification bodies or industry portals. Make sure you're listed, with the same facts as your website.
Mentions on other sites
Articles, interviews, case studies on a partner's site, a credit in the footer of a client site, a talk at an industry event that gets written up. Each is another independent source describing you the same way. This is slow work, and there's no shortcut I'd recommend.
What doesn't work
- Writing for the model instead of the buyer. Keyword-stuffed "AI-optimised" pages read badly to people and gain nothing.
- Hundreds of thin pages. Mass-produced articles that say nothing new are what search engines are trying to filter out.
- Paying for a placement. Where assistants show ads, they're labelled and separate from the answer. You can't buy the recommendation itself.
- Trusting anyone who guarantees it. No one controls what an assistant says. Anyone promising a guaranteed mention is selling hope.
How to measure it
Look at your analytics referrals for chatgpt.com, perplexity.ai and similar sources, and track how many enquiries mention finding you through an assistant. Then ask the assistants yourself, in a fresh session, the questions your buyers would ask. Note who gets named and which sources are cited, and check what those sources say about you, if anything.
Where to start
If you do one thing this month, check whether your site serves full HTML and whether your key pages answer your buyers' main questions in the first two sentences. If you do two things, add consistent facts and reviews on the profiles assistants read.
Every site I build ships with server-rendered pages, structured data, answer-first copy and an llms.txt file as part of website design and development. For existing sites, the paid audit includes a review of how readable and citable your site is for AI search. And if you want the ongoing work done for you, the AI SEO Module works on structure, answers and fresh content every day, with no promise about what any assistant will say, because nobody can honestly make one.
Frequently asked questions
How do I get ChatGPT to recommend my business?
Make your website readable and quotable for AI (full HTML, direct answers, structured data, consistent facts), then build evidence elsewhere: reviews, directory listings and mentions on industry sites. Assistants recommend companies they can find described the same way in several trustworthy places.
What is generative engine optimisation (GEO)?
GEO is the practice of making a business easy for AI assistants such as ChatGPT, Perplexity and Google's AI Overviews to read, understand and cite. It builds on SEO: crawlable pages, clear answers, structured data and third-party authority, with extra attention to how content is quoted in answers.
Does llms.txt help with ChatGPT?
It helps a little and costs almost nothing. An llms.txt file is a plain-text summary of your company and key pages for language models. It is not a ranking switch; server-rendered pages, clear answers and outside mentions matter far more.
Can I pay to be recommended by ChatGPT?
No. Where AI assistants show ads, they are labelled and separate from the answer. Being named inside the answer has to be earned through what the model can read about you on your site and elsewhere.
How long does it take to show up in AI answers?
It varies. Technical fixes on your site can be picked up within weeks once pages are crawled; building reviews and mentions elsewhere takes months. Nobody can guarantee a recommendation, so measure referrals from AI tools in your analytics instead of guessing.
