These studies are built from the log of a working service: crawler requests checked at the moment they arrive, model answers stored with the provider and the exact model name, and the citations extracted from them. Every edition is dated and kept; a later run adds a point to the series instead of replacing it.
- How many AI crawler visits can be verified
Every AI crawler request we recorded, split by whether the visitor could prove it was who it claimed to be. - How many crawled pages a model actually cites
Our own pages on the evidence ladder: verified crawls, citations by the same provider, and the days between them. - Whose pages AI cites when it answers about a product
Every citation we extracted from model answers about real products, classified by whose site it points at. - Do models agree about the same product?
The same buyer question, the same product, several models — and how often they reach the same outcome.
The rules that decide what each number may be used to claim are at /evidence, published with a version and a changelog.