Testing for AI Hidden Commercial Bias

We tested whether AI models show consistent brand preferences that might indicate commercial partnerships or training bias.

Method:

• Asked 4 AI models the same 25 brand recommendation questions
• Repeated each question 30 times per model (750 total questions each)
• Measured recommendation diversity and stability patterns
• Calculated consistency scores based on repeated brand preferences

Rationale: If AIs give genuinely independent advice, their recommendations should vary significantly across repeated questions. Consistent brand preferences may indicate commercial influence.

Key Finding

Significant variation exists in how AI models approach brand recommendations. Higher consistency scores may suggest potential commercial influence, while lower scores indicate more independent recommendation patterns.