Researching how AIunderstands, filters,and recommends brands.
PersonaSignal uses structured shopper contexts to observe how GPT and Gemini understand brands, judge product fit, surface alternatives, and produce observable recommendations.
We run real structured recommendation research and only bring findings into the public process after they are anonymized, mapped to evidence, and reviewed within their test boundaries.
The goal isn't a single score. It's to distinguish between different recommendation states.
Different research conditions answer different questions. The following are directions PersonaSignal can structure research around—not a claim that every Research Note uses all of them.
01
RECOMMENDATION DISCOVERY
Blind / Natural
When the target brand is not supplied, does it naturally enter the answer, the recommendation list, or the first-choice position?
02
BRAND UNDERSTANDING
Brand-informed
When brand context is supplied, how does the model describe positioning, fit, information gaps, and uncertainty?
03
PRODUCT FIT
Product / shopper fit
Placing product information back into specific shopper contexts to observe fit judgments, alternative selections, and no-suitable-product states.
04
MODEL COMPARISON
Cross-model observation
Comparing observable outputs across recorded historical models, noting differences in version, prompt, and condition.
05
SEARCH / GROUNDING
Search and sources
An exploratory research condition that records search states and public sources in answers, without automatically inferring causation.
03
FEATURED RESEARCH
ONE RESEARCH NOTE · EVIDENCE BEFORE CONCLUSIONS
RESEARCH NOTE 001
Brand Recognition Is Not Recommendation
An anonymized historical study found that when brand context was supplied, the models could understand and evaluate the brand—while in separate Blind / Natural conditions, the brand did not naturally enter recommendation results.