researchcogny's own data

what we measured.

Data Cogny publishes about itself and about the tools marketers use. Each page says how it was made and when. Use it, cite it, check it.

01studies

Verdict ledger
167 of Cogny's own marketing changes and how each one turned out: 70 worked, 78 partial, 19 didn't work.
data as of 2026-10-20
AI visibility index
50 marketing questions asked to ChatGPT, Claude, Gemini and Perplexity: who gets recommended. Cogny came up in 5 of 45 non-brand questions.
run 2026-10-14
The truth ledger
Six months of measured customer outcomes, checked against Search Console, GA4 and order data.
case study

02cogny bench

Our own benchmark of AI models on marketing-analytics problems with traps in the data: score against cost per run, published as each model ships.

Same Task, Same Data, Half the Score: Measuring Context Robustness on Cogny Bench
We wrapped a Cogny Bench task in a production-shaped envelope — a thousand words of workspace context plus an organizational memory that confidently asserts the wrong explanation. Claude Sonnet 5 held 99 → 99. GPT-5.6 Luna fell 99 → 47, GPT-5.6 Terra 98 → 43, and Claude Haiku 4.5 96 → 49 — after computing the disproving evidence themselves. And the failing runs were cheaper than the passing ones.
2026-07-16

03reference

Press and brand assets: cogny.com/press