How much does one answer tell us?
The study will compare repeated answers to the same unaided mattress question. Each observation starts in a fresh conversation. No brand list or desired shortlist format is shown to the model.
What we plan to measure
- First-choice recommendation share, separate from any brand mention.
- Explicit ranks, brand diversity, and cases without a clear recommendation.
- Within-run agreement and 95% sampling intervals.
- Modeled leader-identification risk at samples of 1, 5, 10, 25, and 50.
What this study will not establish
A single collection window cannot measure stability across days. API proxies cannot reproduce every consumer app’s routing, memory, web access, or personalization. Six equally weighted systems would not represent consumer usage share.
Simulation will estimate risk under stated assumptions and reviewed classifications. It will not reveal a known, universal “true” leading mattress brand.
Follow-up questions, not yet approved
“What mattress brand would you recommend for a side sleeper?”
“What’s the best mattress for around $2,000?”
These conditions stay separate from the unaided baseline. They will test whether purchase context changes the recommendation distribution.
Results await evidence and publication review. No mattress findings, brand ranking, or benchmark score has been published.
Review the methodology →