“Which web agencies in Noord-Brabant build a web shop with iDEAL?” Five years ago that question went to Google and produced a list. Today it increasingly goes to ChatGPT and produces three names, with a sentence of explanation each. Whether your company is one of those three depends on things largely within your own control.
This article describes how ChatGPT and similar assistants arrive at their answers, what you can do to be named, how to check it, and where the limits are.
How ChatGPT finds companies
There are two routes, and they call for different work.
Route 1: the model searches live. For questions about current or local topics, ChatGPT runs a search through Bing, reads a number of results and summarises them. Perplexity and Google’s AI mode work similarly. Almost the same rules apply here as for classic findability: if you are not in the first results on Bing or Google, you are not read and therefore not named.
Route 2: the model answers from memory. For general questions the model answers from what it read during training. That memory is a snapshot of the public web from months ago. Here, mentions in many independent places count: Wikipedia, industry directories, news articles, forums, review sites. A company that exists only on its own website barely exists for this memory.
Most business questions run through route 1. That is good news, because a well-maintained website gives you direct influence there.
What makes a model decide to name you
Tests with hundreds of questions keep producing the same picture. Companies that get named have four things in common.
A page that answers the question literally
When someone asks about web shops with iDEAL in Brabant, the page that says “we build web shops with iDEAL, Mollie and Adyen for businesses in Noord-Brabant” beats the page that says “we realise digital growth”. The model is looking for text it can take over almost word for word.
Concrete, verifiable facts
Prices, delivery times, technologies, customer counts, location. A model likes citing a fact and avoids vague claims, because it is trained not to make things up. Whoever publishes no facts gives the model nothing to cite.
The same information in several places
Your company name, address, services and specialisms must tell the same story on your site, in the Chamber of Commerce register, on LinkedIn, in Google Business Profile and in reviews. Contradictions make a source unreliable, and a model would rather leave out an unreliable source than risk an error.
Recognition from outside
Reviews on independent platforms, mentions in local media, a guest article in a trade publication, a case study with a partner. Each is a voice saying that you exist and do what you claim.
What you can do this month
Write your service pages as answers. Open every page with what you do, for whom, where and at what price. Use the questions customers ask as headings and answer them directly underneath.
Publish your prices, or at least ranges. “From €3,950” is quotable. “On request” is not, and the competitor who does name a figure gets named in your place.
Align your company details. Check your Google Business Profile, LinkedIn page, Chamber of Commerce listing and the footer of your site. Same name, same address, same description in the same words.
Add structured data. Organization schema with logo, address and service area. Service schema for your services with prices. FAQPage schema for your frequently asked questions. This is the language in which a machine reads what you are without guessing.
Let the crawlers in. Check that your robots.txt does not block the crawlers of OpenAI (GPTBot, OAI-SearchBot), Google (Google-Extended) and Perplexity. Some security plugins do so by default, and then you are invisible without knowing it.
Collect reviews where the model reads them. Google Business Profile and the large review platforms weigh more than a quotes carousel on your own site.
How to check whether it works
Make a list of ten questions a customer would ask, in that customer’s words. Put them to ChatGPT, Gemini and Perplexity monthly in a fresh conversation, logged out where possible, and note whether you are named, in which position, and which source the model cites. After three months you will see a trend. Also compare which competitors do get named and look at their pages: usually it is immediately clear why.
In your web analytics these assistants appear more and more often as referring sources. That traffic is small but valuable, because the visitor has already received a recommendation before arriving.
What not to do
No hidden text “for the AI”. No fake reviews. No pages that exist only to feed a model. Language models are trained to recognise manipulation, and the platforms share those signals with their search index. What a human should not see, a machine should not see either.
Do not expect a miracle cure either. For most questions, ordinary findability still decides whether you are read. Whoever sits on page three of Google does not suddenly come out on top in ChatGPT.
Frequently asked questions
Can I pay to be mentioned in ChatGPT?
Not at present in the way you buy ads in Google. Advertising models are being tested, but the mentions in the answers themselves are organic. That is exactly why it pays to start early.
How long before ChatGPT names me?
Through live search it can happen within weeks, once your pages rank well in Bing and Google. Through the trained memory it can take months, because you have to wait for a new training round that has picked up your mentions.
Does publishing in English help?
Not for Dutch customers. The models answer in the language of the question and look for sources in that language. A Dutch question is answered with Dutch sources. Publish in the language your customer asks in, and add English if you also work internationally.
What if ChatGPT says something wrong about my company?
Find the source of the error; usually it is an outdated page or an incorrect directory listing. Correct that source and make sure your own site shows the right information prominently and with structured data. The model follows the best-documented version of the truth.
Where to start
Put five customer questions to ChatGPT tonight. If you are named, see which source the model uses and strengthen it. If you are not, read the pages of those who are: the difference is almost always clarity, figures and consistency, not budget.
We build sites written and structured this way, and include this check in our SEO service. Send us the five questions your customers ask and we will tell you where you stand today.
