What does ChatGPT say when someone asks about your business?

According to the Hungarian AI Index survey of 2026, more than half of Hungarian adults use artificial intelligence: at least once in the month before the survey. Whoever looks for an accountant, a dentist or a developer today has ChatGPT or Gemini at hand, and those give one answer, not ten links. The results page gave you ten chances, the answer gives you one.

This article describes what that changes and what you can do about it — the way I would explain it to one of my clients, without the jargon.

Try it with your own business

You do not need a specialist for this, three questions are enough. Open ChatGPT or any similar service, and ask:

• “Which [your trade] business would you recommend in [your town]?”
• “What do you know about [your business name]?”
• “What should I look out for when ordering [your service]?”

The answer can differ by day and by service, so a single good or bad answer is not worth far-reaching conclusions. The pattern is what matters: whether you appear in the picture at all, and whether the model says about you what you would say yourself.

What has changed in search?

For a long time search gave you addresses: you typed something, you got ten links, and you decided which one to click. The new layer gives you an answer. Above Google's results there is, more and more often, a summary — and ChatGPT is a conversation to begin with, there is no results page at all.

The answer does not come from nowhere. The model reads source pages — websites, articles, databases — and builds a few sentences from them, often naming the source an assertion comes from. Whoever got an answer clicks on less, because someone has already replied to their question.

By 2026 the scale can be stated plainly. ChatGPT is used by 900 million people a week, per OpenAI's February 2026 disclosure; Google's answer summaries are seen by two and a half billion people a month, and the Gemini app is used by 900 million — the latter two announced by Google in May 2026. These are not novelty numbers any more, this is the everyday state of search.

The systems do not work from the same index. Gemini and Google's answer box work from the Google search index — what Google reads well, they read well. Copilot works from Bing's, Claude from Brave's, and ChatGPT and Perplexity partly from what their own crawlers collect. For a business this means there are several gates to fit through, not one — but tidy content fits through all of them.

• Your website if machines can read it
• Other websites competitors, listings
• Articles, reviews mentions of you
• Question “who do you recommend?”
• Language model builds the answer
• One answer a few sources quoted

this is what it reads

What does this mean for a small business?

Two things happen at once. Fewer clicks make it through to websites, because the answer settles many questions on the spot. But whoever is in the answer gets more than the ten blue links ever gave them — the model mentions them by name, as a recommendation.

Meanwhile the users are ahead of the businesses. In Microsoft's 147-country survey published in spring 2026, Hungary — my home market — ranks 19th in the world for AI use, first among the Visegrád countries. Content in Hungarian is scarce, though, and language matters: for local-language questions, in one 2026 study, Google's answer summary cited a local-language source 85% of the time and ChatGPT 70% of the time (Temso). Where such sources are few, the model picks from few.

That is where the gap is. The order that hardened over years in the world of ten blue links is being dealt out again, and the answer goes to whoever the model can find, read, and see named by others.

What gets a business into an AI answer?

It does not hang on a single setting. It hangs on four things that together make a business visible and quotable to the machine.

Getting into the indexes

The answer works from search indexes, and a business none of them knows cannot be quoted. That takes crawlers getting in: the robots.txt file has to let them, and the hosting firewall must not filter them out. They also have to know about your pages — search consoles, the sitemap and change notifications take care of that. The model does not look at your page, it reads it: sliders, animations and text baked into images do not exist for it, clear headings and actual text do.

Sentences that stand on their own

The model does not quote the whole page, it lifts pieces out of it. A sentence is quotable when it is true and clear on its own. “Open Monday to Friday, 8 till 4” — that is quotable. “Our door is always open” — that says nothing to the machine, and not much to the reader either.

Placement matters too. 44% of ChatGPT's citations come from the first third of a page, according to a 2026 analysis of 18,000 citations. And when someone asks who does something, the model prefers to quote a service page with the answer up front: who, what, where, for how much.

Consistent details everywhere

The model recognises a business when its name, address and service read the same everywhere — on the website, in the Google business profile, on social pages. Where the details contradict each other, the machine is unsure, and it leaves unsure sources out.

Mentions elsewhere

The model prefers to state what it sees in several places. A trade article, a genuine review, a listing you appear in — each one is a confirmation that the business exists and is what it says it is. Measurements say this is the strongest signal: in Ahrefs' analysis of 75,000 brands, appearing in AI answers went together with mentions on other sites far more closely than with links.

What the measurements say does not decide it: the data layer under the code (structured data), a file called llms.txt, and Q&A blocks. None showed a measurable effect in the 2026 studies (schema, llms.txt). They do no harm, and this page has them too, but they are not what it comes down to.

What can you do right now?

Part of this work lives in the copy, and anyone can start on it. Another part usually needs a developer.

• Let the website give straight answers to real questions — what you do, for whom, where, what it starts at.
• Check whether you are in the search engines: Google Search Console and Bing Webmaster Tools show for free which of your pages are indexed.
• Name, address and service should match everywhere — website, business profile, social pages.
• Say facts, not slogans — “quality and reliability” tells nobody anything, “we call back within 24 hours” does.
• Let the crawlers in: the robots.txt and the hosting firewall should let search crawlers through — the server log shows whether they do.

Anyone can start on the first four. The fifth usually needs a developer, and it is the work I do — together with the rest — under the name AI visibility.

How I do this in practice

The work starts with the server log: which search and AI crawlers visit the site, what they request, and what they get back. On my own site in September 2026 this showed that over a month ChatGPT's search crawler requested five pages and its training crawler a hundred and thirty — and answers work from the former.

Then come the gates: the search consoles, the sitemap, change notifications, and making sure the firewall does not shut out search crawlers. Then the pages: reshaping the service pages so the answer comes first — who, what, where, for how much, what the proof is, and who it is not for — and lining up the business's details beyond the website, so the machine sees the same thing everywhere. Even so, the larger part of the work is not writing new content, it is putting the existing content in order.

Measuring is best done with the same questions, monthly, in several assistants. In a 2026 study of nearly three thousand runs, the same list of recommendations came back twice less than one time in a hundred, while the share of runs a name appears in stayed stable. So I do not measure position. Of my projects, the case study about a trades business sits closest to this: visibility was the axis of that work.

What cannot be promised: a specific position or a date. The models refresh from time to time, their answers shift. What can: putting in order everything that depends on us, and measuring back now and then — with the same questions you tried at the top of this article.

Questions on this topic

← Back to the home page