Research on
AI search visibility.
Open, openly-licensed research on how brands surface across AI search engines — measured on Ranqo's production data across the five major engines.
How we measure
AI visibility.
Every figure in these papers comes out of the same tracking pipeline our customers run, across ChatGPT, Gemini, Perplexity, Claude, and Grok.
- Production telemetry, not a benchmark
- The data is the same pipeline that answers our customers' dashboards. Nothing is staged for the paper, and no prompt is written to make a point. That means the numbers move with the real web, and it also means we inherit its messiness rather than a clean lab distribution.
- The prompts never name the brand
- Asking an engine about a company guarantees it talks about that company, which measures nothing. Every tracked prompt describes the problem a buyer has, so a mention is the engine choosing to recommend the brand unprompted.
- Every answer is read, not just matched
- Each response is parsed for whether the brand appears, where it ranks among the vendors named, how it is characterized, and every URL cited as evidence. Citations are resolved to their real destinations before they are counted.
- Grounded on live search, every run
- Queries run with web search enabled, so answers reflect what the engines can retrieve that day rather than what they memorised at training time. Runs repeat on a schedule, which is what makes a trend distinguishable from a single noisy answer.
Published in full.
Including the parts that undercut us.
Every paper here is posted to arXiv under CC BY 4.0 and can be read, quoted, and criticised without asking us. We publish the preprint because a vendor making claims about a market it sells into should be checkable, and because the alternative, a statistic in a blog post with no method behind it, is exactly the kind of source this field has too much of already.
The limits are stated in the papers rather than buried. The brands measured are the brands that use Ranqo, which is not a random sample of the web. Engine coverage is uneven, because some engines are only tracked on higher plans and therefore carry a smaller and differently-composed set of brands. Generative answers vary between identical runs, so any single observation is weaker than the trend it sits in. Where a result is sensitive to one of those limits, we say so next to the result.
What we do not publish is the underlying per-brand telemetry. It belongs to the brands being measured, and releasing it would expose their competitive position to make our paper more convincing. Aggregates, method, and every figure needed to argue with the conclusion are in the paper itself.
Ready to see what AI
says about you?
Start monitoring your AI visibility today. No credit card required.