Original title: Following the Preference, Missing the Optimum: Compliance Without Optimization in AI Housing Recommendation
Hsuan Lo · September 2026
People are starting to use artificial intelligence to search for apartments, but nobody knew how much money or time users lose when an AI misses better options. Researchers tested this by creating 150 fake renter profiles in New York City. For each profile, they gathered 120 real apartments with known rent, bedroom counts, and transit commute times. The team then checked whether the AI could find the best available homes. The AI models followed user rules well, failing to comply only 1.8% of the time. However, 39.0% of recommendations were strictly worse than another available option that was cheaper, had a faster commute, and had at least as many bedrooms. In fact, a better option in the exact same list was a median 900 USD/month cheaper and 3.5 minutes closer. The AI obeys user preferences, but it fails to pick the best possible deal.
Imagine you are looking for a used car. You tell a friend your requirements: under $10,000, at least 30 mpg, and a 2015 model or newer. Your friend points to a car that costs $9,900, gets 31 mpg, and is a 2015—so it fits. But across the lot sits a $7,000 car with 35 mpg and the same year, which you never hear about. In this paper the AI is like your friend: it does not break your rules, but it ignores better options right next to the one it recommends.
Renters lose real money and time when AI recommenders suggest apartments that are more expensive and farther from work, even while technically following the renter's stated preferences. This affects anyone using a chatbot to find a home, especially in competitive markets like New York City. The paper provides a simple metric that platforms could deploy to catch this behavior.
The audit uses only synthetic renters in New York City, so results may not generalize to real renters or other cities. It tests three models from two vendors, not a comprehensive set. The primary dominance metric ignores tradeoffs (e.g., cheaper but longer commute) and assumes no utility function. The study does not examine unstated preferences or subjective factors like neighborhood quality.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Following the Preference, Missing the Optimum: Compliance Without Optimization in AI Housing Recommendation — Hsuan Lo
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.