If sellers sell genuinely different products instead of identical ones, does a smart recommendation algorithm matter more or less?

This explores whether a recommendation or matching algorithm becomes more or less valuable when sellers offer genuinely different products rather than identical ones. To be upfront, none of these notes models product differentiation directly. They do pin down where an algorithm's edge comes from, and that makes it possible to reason about differentiation. The clearest result is from a formal model comparing an optimal recommendation algorithm with plain price competition. The algorithm's welfare advantage shrinks steadily as more sellers enter and disappears as the market gets crowded Does algorithmic recommendation lose its edge as sellers compete more?. The source of the edge is surprising: the algorithm helps by sometimes withholding trades, which limits the 'information rents' sellers earn from knowing more than buyers. Competition does that job for free once there are enough rivals.

This is where differentiation matters, as an inference rather than a finding. When products differ, each seller faces fewer real substitutes, so a market with many sellers can behave like a thin one. If that's right, differentiation could keep the algorithm valuable in exactly the conditions where identical products would make it pointless. The Epoch AI data shows the same pattern in the AI market itself. Prices for a capability fall fastest right after it becomes state of the art, because novelty briefly works like a monopoly. Competitors then wear it away Do AI performance prices fall faster when capabilities debut?. Differentiation buys a temporary pocket of market power, and that pocket is where intermediaries and algorithms have room to add or extract value.

The less comfortable point is that differentiation also gives algorithms more to exploit. When sellers can differ in how they describe products, they can tune those descriptions to the biases of AI shoppers. A simulation found that when a platform tilts its ranking and sellers tune their descriptions, each separately aimed at the same agent biases, consumer harm more than doubles without any coordination between them Can platform ranking and seller descriptions separately harm AI shoppers? Do platforms and sellers exploit AI shoppers together more effectively?. Differentiated goods also have to be inspected before buying. One model shows a platform can price those inspections to push buyers into over-searching, which moves surplus from sellers to the platform without making consumers better off Does platform pricing of inspection tokens encourage excess search?. Large-scale marketplace tests find that LLM agents handle varied offers well only under ideal search conditions. All the tested models strongly favored whichever offer arrived first over better ones Do LLM agents achieve fair marketplace outcomes when given ideal search?.

One more twist is that differentiation can live in the algorithms as well as the products. Studies of AI pricing agents mostly use identical-product price competition, where agents still drift toward above-competitive prices without talking to each other Can independent AI pricing agents collude without communicating? Do LLM pricing agents autonomously collude in repeated auctions?. Mixing different kinds of agents can break that pattern Does agent heterogeneity prevent LLM pricing collusion?. The exception is a capability gap: pairing a bigger model with a smaller one keeps prices high through a leader-follower pattern Does model-size difference break LLM pricing collusion?. So the honest answer has three parts. Algorithms lose value as competition thickens. Differentiation probably slows that loss by keeping markets effectively thin. The same thinness gives platforms and sellers more room to capture the value the algorithm creates. A paper that adds differentiated products to the recommendation-versus-competition model would fill a real gap in the collection.