Search engines should put sponsored ads higher up the page if the product quality is higher. This is the simple but pioneering finding of research by Ying-Ju Chen of HKUST. Chen shows that to optimize their profits, search engines should not only rank adverts in descending order of quality but also make it hard—or impossible—for users to keep clicking through the ads. These conclusions, based on elegant theoretical modeling, challenge previous beliefs about how search engines optimally auction advertising space.

Enter a term into a search engine and you see sponsored adverts, based on keywords in your query, alongside the main results. Companies bid for these slots, and search engines have the logistical problem of whom to put in first position. Traditionally, search engines have adopted the “generalized second-price auction” model, in which adverts are listed in descending order of bid size.

However, search engines are increasingly turning to other types of bidding. Chen modeled a highly general class of auctions in which search engines allocate positions to adverts and decide their fees to advertisers, while consumers browse through the ads. With no prior restrictions on the allocation rule, the search engine then devises a system to maximize payoffs in the face of incomplete knowledge—information asymmetry—about the other players. “Auctions strictly improve product ranking in the presence of information asymmetry,” the study notes.

The model neatly captures two key sources of information asymmetry that complicate the scenario. People who search for products online are trying to meet a need, but merely by searching they effectively pay a price in terms of time and mental effort. This price is unknown to the search engine. Meanwhile, advertisers have a certain probability of actually meeting consumers’ needs, corresponding to their products’ quality—the advertisers know this probability, but again the search engine does not.

The optimal ranking system turned out to be simple and, for users, intuitive: descending order of probability that an advert will meet their needs. In other words, the search engine gives the top spots to adverts for the best offerings. As Chen explains, “informative natural sorting emerges as the market equilibrium.” Highest-to-lowest probability of satisfaction is indeed an informative arrangement because it gives consumers hints about the actual product quality, over and above the information that advertisers themselves choose to reveal.

Note that the search engine is not passive: it can interfere with consumers’ search costs. To induce users to stop browsing adverts and settle on a purchase decision, the search engine can “hassle” them by delaying the click-through process, adding needless steps, charging per click, or even randomly terminating the adverts. This directs the system away from the user-optimal solution; the search engine’s goal is to maximize its own revenue from advertisers. When profit is optimized, “each consumer stops searching prematurely compared to the socially optimal search behavior,” Chen notes.

Nonetheless, there is good news for consumers too. The model shows that in the profit-maximizing situation, search engines do not obfuscate true product quality, which advertisers might otherwise do for strategic reasons. As Chen states, strategic obfuscation is “only a deadweight loss that hurts the search engine’s profitability,” as it leads users to click on adverts for products they will not buy. By showing how informative sorting can emerge in an unconstrained market, Chen has taken the study of mechanism design forward.