Hungarian AI visibility: which agency ChatGPT names

10 questions · 100 runs · 345 mentions · 38 entities · measured 26 August 2026

English summary

The full study is published in Hungarian, because it measures the Hungarian question space: the prompts, the answers and the named companies are all Hungarian. Translating the field would misrepresent what was measured.

This page carries the abstract and the headline numbers. The complete study, including every per-question table, is at /hu/research/magyar-ai-lathatosagi-piac.

What we measured

We asked ChatGPT ten Hungarian questions about AI-visibility, AI-SEO and GEO agencies, ten independent times each, logged out and in a private window. For every run we recorded which companies were named.

The metric is the mention: whether a name is spoken at all, in how many runs out of ten. Position inside the answer is not treated as a measure. An AI answer is prose, not a ranked list, and there is no measured position-to-click curve for LLM answers, so any value attached to position would be an underived weight.

The four findings that survive

The form of the question splits the market in two. On evaluative questions ("which is the strongest X agency in Hungary") AVE Studio is named in 41 of 50 runs. On catalogue and advice-seeking questions the same brand is named in 12 of

1. The difference is large and not noise. Half of the 38 entities we recorded exist only in the catalogue and advice questions, so half the visible market is reachable only through a question type where we are barely present.

Being a source is not being named. A separate measurement recorded the avestudio.pro domain as a source in 69 answers, while AVE Studio was named in 53. The overlap between those sets bounds the gap: in at least 16 runs the model read the site, used it, and did not say the name. The bottleneck is the passage-to- entity step, not retrieval.

The top of the market is a tie. Across 100 runs the two most-named companies land one mention apart, and 85% of AVE Studio's mentions share the answer with the other. They are two members of one shortlist rather than rivals for one slot.

The highest-intent question is the emptiest. Asked which of two service categories a company needs, ChatGPT named no supplier in any of the ten runs. It explains and compares; it does not recommend.

Limits

One engine, one day, logged out. n=10 per question is enough for the aggregate question-type effect and not enough for most single comparisons. The study does not measure which company is best; it measures which companies ChatGPT names.

Individuals who appeared in the field are not named: the measurement is about the machine visibility of companies.