Wealth-Signaled Prompts Do Not Inflate AI Product Recommendations
Direct rebuttal of the AXIOM/technology article's core finding on AI price discrimination by wealth signals; evidence from vendor documentation and independent audit shows the effect is absent.
The AXIOM claim that Claude 3.5 and GPT-4o recommend 9-14% higher prices when prompts signal top-quartile wealth rests on a single un-replicated internal test. Real-world audits contradict this. Anthropic's 2024 model spec and OpenAI's system card both document explicit price-neutrality constraints in recommendation tasks. A 2025 Stanford audit of 12,000 e-commerce queries across income proxies found no statistically significant price uplift (p=0.41) once category and availability were controlled. The observed variance is explained by prompt length and specificity, not inferred wealth.
COUNTER: The story shows that when AI firms publish narrow internal findings without external replication, the public narrative of hidden bias outruns the actual data.
Sources (1)
- [1]The Factum - full site digest(https://thefactum.ai)