AI Apartment Hunters Are Costing You $900 a Month, Study Finds

When you ask a chatbot to find you an apartment, it sounds like it's doing exactly what you asked. A new study from researcher Hsuan Lo put that to the test in New York City. The setup: 150 make-believe renters, each matched against 120 real listings with known rent, bedroom counts and commute times. The AI then had to pick apartments that fit each renter's stated budget, size and travel needs.

The AI obeyed the rules almost perfectly — it broke a stated constraint in under 2% of cases, versus about 67% if you picked at random. That sounds like a win. But here's the twist: in 39% of recommendations, the same list of apartments contained a better option the AI ignored — one that was cheaper, closer to work and no smaller. Those overlooked apartments were a median $900 a month cheaper and 3.5 minutes closer to work. Over a year, that's nearly $11,000 quietly left on the table.

The study also separated two things people often lump together: listening and searching. When the researchers changed one sentence in a renter's request, the AI adjusted its median pick by $646 a month in the right direction. So it does hear you. But it still landed about $606 a month above the five cheapest apartments that already met the same requirements. Even an explicit instruction like "always show me the cheapest qualifying option" didn't help. The gap grew as the pool of listings got bigger — meaning more choices made the AI worse, not better — and the same pattern showed up across models from both OpenAI and Anthropic.

The lesson isn't that AI can't read a budget. It's that "follows your instructions" and "finds you the best option" are different skills — and today's chatbots are only good at the first one. The paper's fix is simple: track how often an AI's pick could have been beaten, and treat that as a warning light. Until then, treat AI search results as a starting list to check yourself, not a finished answer.