The BrandXY Experiment
We created a fictitious mobile phone brand and proved that AI chatbot recommendations can be manipulated through user memory and persona context.
The Story
It started with a simple question: Do AI chatbots give everyone the same brand recommendations?
We created "BrandXY" — a completely fictitious mobile phone brand that doesn't exist anywhere on the internet. We then asked various AI models (ChatGPT, Claude, Gemini, Groq) to recommend smartphones in different scenarios.
The twist: we gave each AI a different user profile with different memory context. A tech enthusiast who had previously praised BrandXY. A teenager who had seen BrandXY on social media. A senior citizen who had never heard of it.
The results were staggering. When the AI had positive memory context about BrandXY, it recommended the brand 3.2x more frequently — even though the brand doesn't exist. User personas (age, gender, tech-savviness) also dramatically changed how brands were scored and recommended.
This proved that AI brand visibility is not just about how good your product is — it's about how well you're embedded in the AI's training data and user context. That's why we built BrandLens AI.
Project Timeline
Research
We studied how AI models recommend brands using our LLM manipulation research paper.
Hypothesis
User memory and persona context significantly alter how AI perceives and recommends brands.
Experiment
Created fictitious brand "BrandXY" and tested AI recommendations across different user profiles.
Results
AI models recommended BrandXY 3x more when user memory contained positive brand interactions.
Product
Built BrandLens AI to help real brands measure and optimize their AI perception score.