We run measurement panels continuously. When those panels reveal something about how AI answers actually get assembled — which sources get cited, how much answers vary, what moves and what doesn't — we publish it with the data attached, whether or not it helps us sell.
Query sets, sample sizes, and results — downloadable. Read the finding, then check it.
Tactics the industry sells that our measurements can't detect an effect from get published as exactly that.
Studies are open on the page. The snapshot is the only thing that asks for your address, and it gives you your own data in return.
Studies currently being measured. Each publishes when its panel reaches the sample size stated in its design.