Artificial Intelligence Clones and Matching Markets

AI clones, equipped with users' histories and preferences, can evaluate far more alternatives than users themselves, but may represent their preferences imperfectly. Liang (2026) formalizes this fidelity--coverage tradeoff for a single user searching a large candidate pool and shows that, in sufficiently high dimension, two exact in-person evaluations outperform arbitrarily broad noisy AI-driven search. We extend this framework to a capacity-constrained matching market. Each user either exactly evaluates a random menu of $\ell$ options or delegates search to an AI that noisily ranks the full market. A centralized clearinghouse then clears these rankings subject to unit capacities and a common priority order. Introducing congestion through a matching constraint alters Liang's findings. In a balanced market, in-person search must cover $\Theta(\sqrt{k/\log k})$ alternatives to outperform AI-driven search in average welfare; at every fixed imbalanced demand--supply ratio, the threshold is $\Theta(\log k)$, where $k$ is the number of dimensions. Exact evaluation improves match quality, whereas sparse in-person menus leave feasible capacity undiscovered and users unmatched. AI's distinctive operational value is therefore coverage, despite noisier evaluations.