Why a Plan to Encourage Search-Engine Competition Failed

Image

What do PrivacyWall, Givero, and Info.com have in common? They’re search engines that compete with Google, and you’ve probably never heard of them. | Illustration by Álvaro Domínguez.

Quick, name a search engine. 

Did you say Google? The internet giant’s name is synonymous with looking for stuff online, but its dominance has drawn scrutiny from regulators who have accused the company of unfairly shutting out its competitors.

When the European Commission came after Google for preinstalling its search engine on devices running its popular Android operating system, the company implemented what seemed like a simple fix: Give users more choices. Now, when consumers in 31 European countries set up a new Android phone, they’re asked to pick one of four search engines, including Google.

That solution may have satisfied European antitrust regulators. But, as Stanford Graduate School of Business economics professor Michael Ostrovsky concludes in a new working paper, it didn’t solve the problem it was meant to fix.

Since 2020, Google has auctioned off the coveted spots on the list of search engines Android consumers may select as their default. As an expert on online auctions, Ostrovsky was intrigued to see this straightforward approach to settling an antitrust issue. But as the results of the first auctions came in, he couldn’t understand why obscure search engines — ever heard of Info.com, PrivacyWall, or Givero? — were popping up on Android screens while more popular alternatives were left off.

In his investigation, Ostrovsky found that a simple feature of the bidding process explained why unpopular search providers were winning Google’s auctions — and how that helped the company maintain its foothold as the world’s most popular search provider.

Bundled Up

Google’s auctions are a response to a common situation in the tech industry: when the maker of a dominant technology platform also offers a product that operates on that platform. “If there’s a competing product — even a superior product — and users want to access that through the platform, the platform provider often ends up prioritizing their own product, whether deliberately or unintentionally,” Ostrovsky explains.

Bundling products together is usually legal, but the practice of “tying” — requiring consumers to purchase one product in order to buy another — can violate antitrust laws. In 1998, in the highest-profile case involving these issues, the U.S. Department of Justice accused Microsoft of squeezing out its “browser war” rivals like Netscape by packaging Internet Explorer with the Windows operating system. (While a federal judge ruled that Microsoft should be broken up, the decision was overturned and the case was settled in 2004.)

For a more current example of bundling, you may not have to look further than your smartphone or tablet. Apple, for example, preinstalls dozens of apps such as Safari, iTunes, and iCloud on its iPhones and iPads. “It’s everywhere,” Ostrovsky says. “Apple, Android, Windows. They all bundle.”

That reality leads to an important question: Are there ways to level the playing field for competing products without drastic solutions like those proposed by the judge in the “browser war” case?

Enter the choice screen — a menu that lets users choose between the manufacturer’s app and its competitors’ offerings. While negotiating with the U.S. government in the ’90s, Microsoft offered to adopt choice screens as a way to make it easier for consumers to consider and install other browsers. That idea didn’t take off in the U.S., but it was accepted in Europe. Between 2010 and 2014, Windows buyers in the European Union were offered a choice of browsers that included Microsoft’s Internet Explorer and several alternatives.

In 2018, the European Commission fined Google a record $5 billion for tying its search engine and its Chrome browser to the Android operating system. The company subsequently agreed to use choice screens to offer Android users more search options.

“But that raised the question of who gets to decide the search engines on the choice screen,” Ostrovsky says. After all, a lot of money was riding on securing a place on that list: The more users a search engine gets, the greater its revenue from advertising. To allocate those spots (and to compensate itself for offering this prime real estate to its competitors), Google set up a system of quarterly country-by-country auctions where the three highest bidders appeared on choice screens alongside Google’s own search engine.

Ostrovsky was intrigued by this approach, as he believes such auctions represent a scalpel-like alternative to the hatchet of antitrust penalties like full-on breakups. “The regulatory toolbox doesn’t have that many things in it. And the tools we do have tend to be heavy-handed, such as ‘Let’s break Facebook into three parts,’ which companies will of course fight tooth and nail,” he says. “This is a more lightweight and elegant solution than breaking a company up.”

But he soon found the choice-screen solution’s results were surprising.

Searching for Answers

“When Google announced the results of the choice-screen auctions, I found them very weird,” Ostrovsky says.

While some of the winners were well-known providers like Bing and DuckDuckGo, others were ones he — and the European acquaintances he spoke with — had never heard of, such as the aforementioned Info.com and PrivacyWall, which won the majority of the auctions.

To figure out what had happened, Ostrovsky did “a bit of forensic research” to model how the auctions function. He found that a seemingly minor feature of the bidding process has a significant effect on who wins: Rather than being asked to bid for each appearance on the Android choice screen, search engines bid on what they’ll pay Google each time someone installs their app.

That gives less popular — but better monetizing — search engines an incentive to place high bids, even though fewer users will ultimately pick them. As Ostrovsky explains, “Say search engines A and B are bidding, and search engine A monetizes users really aggressively so they’re willing to pay $10 per install to gain those users and related revenue. Search engine B may be much more popular, but may not monetize users as well because they emphasize privacy or donate a fraction of their ad revenue to charity, for example. So B may only bid $5 per install. And search engine A will win the auction.”

As Ostrovsky’s model predicted, many search engines that were much less likely to be installed wound up winning the Android auctions. By the third round of the auctions, popular search engines like DuckDuckGo and Ecosia were almost completely absent from the lists of winners. Meanwhile, Info.com won choice-screen spots in all 31 countries in all three rounds, despite being installed fewer than 100,000 times worldwide. (By comparison, Google has more than 5 billion installs.) The end result: The Android choice screens don’t offer much of a real choice, so the vast majority of users are likely to pick Google, making the remedy largely ineffective.

Competing Interests

The solution, Ostrovsky says, is not to throw out such auctions altogether but to tweak them so they’re more efficient.

For example, Google could simply auction off spots on its choice screens on a per-appearance basis, so that search engines pay for the right to appear on the choice screen, which would make the auctions more attractive to the popular search engines that users are more likely to install.

Alternatively, Ostrovsky proposes that the design could take a page from the auctions search engines already use to sell ads. “Originally, in ad auctions, advertisers used to bid for spots in ad rankings on a per-click basis, but search engines realized advertisers willing to pay a lot per click may get fewer clicks, which meant lower overall revenue,” he explains. “So they moved to using both what advertisers would pay per click and some measure of the expected number of clicks to decide ad rankings.”

However, Ostrovsky notes that the Android scenario is unlike the ad sales scenario in that Google benefits when its more serious competitors fail to win places on the choice screen. “In this case, the auctioneer is OK with having obscure search engines win, because the auctioneer’s own search engine is chosen more often as a result,” he says. “That’s not to say Google is deliberately doing this — just as with ad auctions, it will take time to converge to the best design, and one shouldn’t expect it right away on the first attempt. In fact, Google and the European authorities should be commended for trying new and creative solutions. But it’s important to keep in mind that the incentives for these auctions are not fully aligned.”

In the end, Ostrovsky thinks the best solution is somewhere between the extremes of locking in one default option and overwhelming users with choice screens for every possible option. “Forcing auctions for every item would make buying a phone the most painful thing ever — a choice screen for keyboards, a choice screen for weather apps, and so on,” he says. “But if platforms are going to run such auctions for the most important choices, they should design and run effective ones that actually do what they’re intended to do.”