In “Algorithm Transparency and Search Manipulation: Steering vs. Persuasion” (with Raphael Boleslavsky and Mehdi Shadmehr) we study a familiar feature of digital markets that is easy to overlook. Platforms do not merely decide which products are available. They decide which product consumers encounter first. That choice matters because prominence affects attention, continuation search, and ultimately which seller is considered. But rankings can do something else as well. When consumers understand how a platform’s private information affects its ranking, prominence becomes a signal. A ranking can therefore both steer consumers by changing their search path and persuade them by changing what they believe.

We introduce a model featuring a platform that, due to its vast consumer data, privately knows which of two products is the better match for a consumer but earns more from selling one of them. Some consumers buy the prominent product immediately, while others can continue searching at a cost. When the ranking algorithm is opaque, the platform’s preferred equilibrium simply places the more profitable product first. Transparency of the ranking algorithm changes the problem. Once consumers understand the state-contingent ranking rule, the platform can commit to making the ranking informative. It may use positive sorting, placing the profitable product first as favorable news and persuading consumers to stop searching. It may also instead use negative sorting, placing the alternative first in a way that makes consumers sufficiently optimistic about the profitable product to continue searching for it.

Positive sorting benefits both consumers with lower and higher search costs. Negative sorting harms consumers who never search and buy the prominent product, while those able to search may gain or lose depending on the strength of search and information frictions. A further result complicates the usual policy intuition that broader transparency must be better. If the platform can issue a transparent recommendation separately from the ranking, it no longer needs to sacrifice prominence in order to communicate. It keeps the profitable product first and moves persuasion into the recommendation layer. Platform profit rises, but the consumer gains created by a transparent ranking disappear. Transparency is therefore not only about how much a platform explains. Its architecture matters.

The broader implication is relevant for platform managers, economists, and regulators. A ranking is both a message and an allocation of attention. Transparency can discipline a platform when its claims must be embodied in the same prominence decision that shapes consumers’ search costs. That discipline weakens when recommendations become a separate, costless messaging layer. One possible policy principle is therefore recommendation–display consistency. When a platform recommends a product on the basis of information unavailable to consumers, it should also make that product correspondingly easy to inspect and purchase. In less formal terms, platforms may need to put their money where their mouth is.