AI Copilots in Real Estate: Evidence from China

We study how an intermediary-facing AI copilot affects market outcomes in two-sided, intermediated markets, using the rollout of an AI copilot for real estate agents on a large Chinese resale housing platform. Combining a randomized field experiment with an exposure difference-in-differences design, we find that copilot access accelerates listing sales, reduces time-on-market on both sides, raises transaction prices, and improves post-transaction ratings. Because the copilot is exposed through distinct buyer- and seller-facing in-app interactions, the rollout allows us to organize the estimates in an own- and cross-side response matrix that maps buyer- and seller-side exposure to outcomes across the transaction. The matrix reveals two patterns: efficiency gains are strongly two-sided with economically meaningful cross-side effects, while price effects load primarily on buyer-side exposure. A stylized search-and-match framework interprets price and time-to-trade jointly and maps the price effect into realized-value and continuation-value or outside-option channels, helping discipline the mechanism. A back-of-the-envelope accounting exercise shows that buyer surplus would break even with an unmeasured buyer-value component of about 1% of transaction value; buyer feedback moves in a direction consistent with improved perceived transaction value.