Research
AI visibility research and data
This is where Llamat publishes what it measures itself: which tools ChatGPT and Gemini name when asked real search phrases, which AI crawlers the most popular websites block in robots.txt, and how often people in the US search Google for GEO, AEO and AI visibility. Every study gives its date, sample and method, links its data as a CSV file and says how often we repeat it: the AI answers every month, search demand every quarter, and the robots.txt scan is a single snapshot from 3 October 2026.
What we measure and why
Our free AI visibility checker shows how ChatGPT, Gemini and Google AI Overviews answer about one brand. These studies look wider: at a whole product category, at the largest sites on the web and at what people search for. We run them so that the numbers we quote on this site can be traced to files we keep, and so that anyone can count them again from the CSV.
- Which AI visibility tools do ChatGPT and Gemini recommend? Repeated every month. The first edition was measured on 3 October 2026: 11 real search phrases from Google Suggest (US, English), each asked word for word three times on logged-out ChatGPT and three times on logged-out Gemini, 66 answers, all of them successful. Semrush was named in 27 of 33 ChatGPT answers and 30 of 33 Gemini answers. Llamat was named in none of the 66. CSV.
- Which websites block AI crawlers? One snapshot, taken on 3 October 2026. We read robots.txt on the Tranco top 10,000 and could tell the rules for 6,528 sites: 17.2% fully block at least one AI crawler and 11.7% fully block OpenAI's GPTBot, but 443 of those 766 sites (58%) do not fully block OAI-SearchBot, the crawler behind ChatGPT search. CSV.
- Generative engine optimization statistics: refreshed every quarter. US Google Ads search volumes for 169 groups of phrases about GEO, AEO, AI visibility and LLM SEO, month by month from September 2025 to August 2026, pulled on 18 September 2026: together about 75,000 searches a month. CSV.
How we keep and publish the data
Raw data is kept. Each call that reached ChatGPT or Gemini is saved as it ran, with the answer text and its sources, failed answers included. Calls that our own API rate limit rejected before they reached an engine are not measurements; on 3 October 2026 we removed 124 such records from the raw file, and the collector no longer writes them. Apart from that, the files are not edited by hand. The tables on our research pages are computed from those files by scripts. In the answer datasets, tool names are found by matching names in the answer text and checked by hand, not judged by a language model.
Every study links a CSV file you can download: the tools and cited domains of each answer, one row per domain for the robots.txt scan, and every number from the tables for search demand. If you quote a figure, please cite the study page and the date of the measurement.
The AI answers are collected the same way as for our free report: logged-out sessions, ChatGPT through our own proxy farm with Czech exit IP addresses and Gemini with its country and language set to the market's. Our methodology describes the collection in full, together with what measurements like these cannot tell you.