How to be visible in AI assistant answers
A growing share of people ask an assistant rather than a search engine. What decides whether a model mentions your brand — and what you can do about it.
Search is changing faster than at any point in the past twenty years. A growing share of people no longer type a query into a search engine but ask an assistant — and get the answer directly, without clicking through to any website.
For brands that means something uncomfortable. Traffic can fall even while the brand is more visible than before. And conversely: a brand a model names in its answer gains visibility that will never appear in its analytics.
What a model needs in order to mention you
A language model does not pick sources by search ranking. It picks texts with something to take — concrete claims, numbers, definitions, verifiable facts. Vague marketing copy does not get cited, because there is nothing in it to cite.
The difference is visible in two sentences about the same service:
We deliver innovative solutions tailored to your needs.
Managing performance campaigns is charged as a monthly retainer from CZK 15,000; one-off projects are quoted as a fixed price against the brief.
A model can do nothing with the first sentence. The second answers a question somebody actually asked.
Structure is half the work
A text with an honest heading hierarchy and the answer placed directly beneath the question reads better for humans and machines alike. The question-and-answer format is the most effective here, because the pair is already separated and the model does not have to infer it.
Structured data from schema.org belongs with this. It describes what a page is
about in a way nobody has to derive from prose. FAQPage for frequently asked
questions and DefinedTerm for a glossary are two formats worth deploying
almost every time.
The llms.txt file
A convention is emerging of placing an llms.txt file at a site’s root — a
structured overview of the content with short descriptions of individual pages.
It is not an enforced standard, and it is not yet clear how much models use it.
The cost of producing one is negligible, though, especially when generated automatically from the same data as the site. It is worth trying.
What to do now
The best preparation for AI-mediated search happens to be good content marketing as well: write concretely, answer real questions, give numbers and do not round them into meaninglessness.
Brands that were already doing this have a head start. The rest have a reason to begin.