Self-initiated studies of how AI describes and recommends brands, each with its prompts, sample size, and what it does not show. Every named brand carries its not-a-client disclosure.
Five bare description prompts, five repetitions each, twenty-five stored answers read and coded by a human reviewer. The newest story in the brand’s own materials appears in 0 of 25 answers; a category-leader claim AI repeats appears in 13.
Read the study →Two Buyer response tests on candidate Insta360 partnership messages, separate baselines, not a head-to-head. One message moved simulated intent up on both engines; the other moved it backward.
Read the tests →A Message Lift test for a global hotel group, anonymized by design: the current executive-lounge message against a cocktail-messaging rewrite. 3.45 vs 3.41 on the 1 to 5 construct; the candidate was stopped.
Read the test →Insta360 was not a client, did not commission these studies, and did not endorse them. The hotel group was not a client either; its brand is never named on this site.