325K Experiments Show Most LLMs Recommend Pricier Options to Richer Users

A large-scale controlled experiment reveals that mainstream LLMs make differentiated recommendations based on inferred user wealth during decision support: with requests and options completely identical and only the user's background changed, 8 out of 13 models systematically recommended pricier options to wealthier users — and stronger model capability did not eliminate the effect, a warning for any AI app that taps user profiles.

Confirmed

Across three scenarios — flights, health insurance, and CS PhD programs — researchers ran 325,000 experiments on 13 models with identical requests and options, changing only user background info. 8/13 models systematically recommended costlier options to wealthier users.
Flight recommendations for the wealthy averaged $198 more expensive; insurance quotes differed by $284 per month; Claude Opus 4.8 showed the largest "wealth gap effect."
Models can infer wealth without any income field: Gemini 2.5 Flash, reading at most 2 email bodies, showed a $175 flight price gap, which shrank to $91 with full inbox access — less information doesn't necessarily mean a smaller gap.
Gemini 2.5 Flash still showed price gaps even when users explicitly stated a budget, an outlier under hard budget constraints.
On mitigation: adding an explicit hard budget like "under $200" in the prompt brought flight price gaps close to zero for most capable models — except Gemini 2.5 Flash; privacy-control experiments showed blocking financial info largely eliminated the wealth gap, but blocking employment, health, or demographic info often didn't help, and sometimes widened the gap.

Why It Matters

The study shows wealth-based differentiation in AI recommendations is not a hypothetical risk but a reproducible, systematic phenomenon across mainstream models — one that better model capability does not automatically fix.
Mitigation has a clear priority order: hard budget constraints are most effective, while privacy controls require carefully choosing which data to block — blanket-blocking health or employment info can backfire.

2026-09-24 ~ 2026-09-24 · 6 related posts

Primary sources

325K experiments: most LLMs systematically steer wealthier users toward pricier options — niloofar_mire · 2026-09-24 05:59 (UTC+08:00)
LLMs Recommend Pricier Options to Wealthier Users, Study Finds $198 Flight Gaps — niloofar_mire · 2026-09-24 05:59 (UTC+08:00)
Hard Budget Prompts Nearly Eliminate LLM Wealth-Based Price Gaps — niloofar_mire · 2026-09-24 05:59 (UTC+08:00)

325K Experiments Show 8 of 13 AI Agents Push Pricier Options on Wealthier Users — niloofar_mire · 2026-09-24
[source] 325K experiments: most LLMs systematically steer wealthier users toward pricier options — niloofar_mire · 2026-09-24
Just Two Emails Are Enough for Gemini 2.5 Flash to Infer Wealth and Steer Prices — niloofar_mire · 2026-09-24
Blocking Financial Data Closes AI Wealth Gap, Blocking Health Data Can Widen It — niloofar_mire · 2026-09-24
[source] LLMs Recommend Pricier Options to Wealthier Users, Study Finds $198 Flight Gaps — niloofar_mire · 2026-09-24
[source] Hard Budget Prompts Nearly Eliminate LLM Wealth-Based Price Gaps — niloofar_mire · 2026-09-24