Edip Kilinc is the founder of Webwavers. Together with his team, he develops websites, brands, and digital presences for ambitious companies – with over 250 completed projects, and honored with awards including the German Design Award 2026.
There are two sources from which an answer is fed, and they behave completely differently.
The training knowledge. What the model learned during training. This includes everything that was widely available on the web at the time of training. This part cannot be influenced in the short term – it only changes with the next model, and what ends up there is decided over years.
The live search. For current or specific questions, the system searches the web, opens a few pages, and summarizes them. This part works like a very fast research process – and this is exactly where you can do something.
From this follows the most important practical insight: Those who are easy to find in the live search will appear in answers. Not because ChatGPT maintains its own ranking, but because it finds the same sources as a search engine – and then selects those that answer a question cleanly.
There is no registration form, no directory, and no provider who can sell you a spot. Anyone offering this is selling nothing.
There are two sources from which an answer is fed, and they behave completely differently.
The training knowledge. What the model learned during training. This includes everything that was widely available on the web at the time of training. This part cannot be influenced in the short term – it only changes with the next model, and what ends up there is decided over years.
The live search. For current or specific questions, the system searches the web, opens a few pages, and summarizes them. This part works like a very fast research process – and this is exactly where you can do something.
From this follows the most important practical insight: Those who are easy to find in the live search will appear in answers. Not because ChatGPT maintains its own ranking, but because it finds the same sources as a search engine – and then selects those that answer a question cleanly.
There is no registration form, no directory, and no provider who can sell you a spot. Anyone offering this is selling nothing.
For content to end up in an answer, three things must be right.
1. It must be accessible. The tools that systems use to open pages usually do not execute JavaScript. What is only created in the browser does not exist for them. The test: open the source code, search for a sentence from the middle of the text.
2. It must not be blocked. The crawlers of AI providers can be excluded in the robots.txt file. Some hosts and security services do this by default. If you want to be visible, you have to check this – it is a common and unnoticed reason for exclusion.
3. It must answer a question. An "about us" text doesn't answer anything. A text that answers the question of what a B2B website costs with a price range, however, does.
These three points are the admission ticket. Everything else decides whether you get selected.
From observation, four things can be identified that have a measurable effect.
Comparison content. Questions about better or more suitable are the most common pattern. An article that compares two options with criteria and a well-founded recommendation is cited as a source more often than average.
Concrete numbers. Price ranges, durations, quantities. Systems prefer to draw on content that provides a number because the question demands it.
Being listed in third-party overviews. Comparison lists, industry directories, third-party provider overviews. When a system searches for agencies for X, it finds these lists – and adopts whoever is in them. This is often effective more quickly than your own content.
Specialist articles on third-party sites. A guest contribution in an industry medium beats ten of your own articles because it is more likely to be found during research and counts as an independent source.
What does not work: creating content in mass, stuffing keywords, depositing meta-information for AI bots that do not exist.
The most common cause of invisibility is not your own website. It is that the company is not written about anywhere.
A model knows a brand because others have mentioned it – in posts, lists, forums, interviews, directories. These mentions are the raw material. A brand without mentions does not exist in the model, no matter how good your own site is.
This leads to an uncomfortable consequence: part of the work takes place outside of your own website.
Concretely, this means: being included in the provider lists that exist in your field. Being listed in awards and directories. Writing one guest post a year. Appearing on a podcast because a transcript is created from it. Being mentioned by customers.
This is classic public relations with a new purpose. And it is the part that is hardest to cut short.
The measurement is simple and takes two hours.
Formulate twenty questions as a customer would ask them. Not keywords, but sentences: Which agency is suitable for a B2B website in mechanical engineering? How much does a website relaunch cost?
Enter them into three systems, each without a logged-in account or in a new window, so that no personalization plays a role.
Note three things: Does your brand appear? Is your page cited as a source? Which competitors show up?
This results in three metrics: mention rate, citation rate, and your share of all mentioned providers.
The most insightful part is the third one. Look at which sources are cited in the answers that feature a competitor. Usually, it's two or three recurring pages – comparison lists, specialist portals, directories. Being listed there is the most direct way.
Three points of context to set the right expectations.
It takes time. A new page takes weeks to be found and months to be regularly referenced. Mentions on third-party sites work faster, but require lead time to be created.
The numbers remain small. Direct traffic from AI services is in the low double digits per month for most companies. The value lies elsewhere: in the mention itself and in the brand searches that follow from it.
It fluctuates. The same question yields different answers on two different days. Single measurements are worthless; only the trend over months tells a story.
And one sentence of context: Everything that helps here – retrievable content, concrete answers, clear authorship, third-party mentions – also helps in classic search. There is no conflict of interest and no reason to treat them separately.
