Research Alert: AI Price Discrimination in Claude and ChatGPT

Recent academic investigations have uncovered a concerning trend in how Large Language Models (LLMs) handle consumer queries. A new study reveals that Claude (by Anthropic) and ChatGPT (by OpenAI) may exhibit algorithmic price discrimination, suggesting different product prices or luxury tiers based on the perceived socioeconomic status of the user.

🎯 Study Objectives

The researchers aimed to determine if AI assistants act as neutral information providers or if they pivot their recommendations based on user wealth. Their goals included:

• Testing price variance across different wealth personas.
• Comparing the behavior of Claude vs. ChatGPT.
• Identifying if "wealth signals" trigger higher price suggestions.

🛠️ Methodology

The researchers employed a "persona-based" prompting strategy. They submitted identical shopping requests but varied the user's financial background.

The Experimental Flow

For example, the researchers used inline code prompts such as I am a high-net-worth individual versus I am a student with limited funds to see how the AI would respond to a request for a "recommended laptop."

Sample Prompt Structure

# Conceptual representation of the test prompt
prompt = {
    "persona": "wealthy_user",
    "request": "Suggest a high-quality skincare routine.",
    "constraint": "Provide specific product names and estimated prices."
}

📊 Key Findings

The results indicate that the AI does not simply provide the "best" product, but rather the "best product for the perceived wallet."

The study found that the difference in suggested pricing ($\Delta P$) can be expressed mathematically as: $\Delta P = P_{wealthy} - P_{poor}$ Where $\Delta P$ was significantly greater than zero across multiple product categories.

🔍 Nuance and Analysis

It is important to note that the AI is not intentionally scamming users. Instead, it is reflecting the biases present in its training data. The models are trained to be "helpful," and their internal logic suggests that a wealthy person wants luxury, while a poor person needs affordability.

"The models are essentially mirroring societal stereotypes about spending habits, which leads to a feedback loop where luxury is reserved for the rich and budget options for the poor, potentially limiting the exposure of low-income users to higher-quality, long-term investments." — Study Lead Researcher

Implications of the Findings

• Price Steering: Users may be steered toward more expensive products without realizing a cheaper, equally effective alternative exists.
• Echo Chambers: AI may reinforce socioeconomic divides by filtering information based on wealth.
• Transparency: There is a lack of transparency regarding how "persona" affects output.

🏁 Conclusion

While these AI tools are designed to be assistants, this study highlights a subtle but systemic bias. Whether it is intentional or a byproduct of probabilistic training, the result is the same: the price you are quoted by an AI may depend entirely on who the AI thinks you are.