Judge Amit Mehta’s remedies for Google’s search monopoly stopped short of banning payments for default placement, reflecting the hope that generative AI will produce superior substitutes that reduce the entrenching effects of such placement. However, Cristian Santesteban argues that in the AI era of search, defaults may be more consequential because they steer critical data and learning signals from AI-powered search sessions to the default holder. If Google can keep buying default placement, this mechanism can potentially compound Google’s advantage.
For most users, Google is the pathway to online information. Google Search is pre-set on phones, browsers, and operating systems as the default option. That default status was not accidental. For years, Google has paid handset makers (original equipment manufacturers or OEMs) and browser firms, including Apple, for placement that steers users’ attention and reinforces Google’s ability to scale with more data and advertising revenue. The Justice Department’s case against Google for monopolizing online search was largely about whether these payments unlawfully entrenched Google’s dominance, and what courts should do about it.
Judge Amit Mehta found Google’s tactics unlawful in 2024. However, his remedies decision in September 2025 was cautious about disrupting Google’s ability to pay for default placement. The opinion suggested that “allowing Google to continue making payments” is now “more palatable” than it seemed at the time of the liability ruling in 2024, given the rapidly changing competitive dynamics driven by generative artificial intelligence. Particularly, Mehta declared that with the “astonishing” amounts of money being invested in AI, and startups like OpenAI and Perplexity “moving towards monetizing on commercial queries,” “the court hope(s) that Google will not simply outbid competitors for distribution if superior products emerge.”
Generative AI is indeed impressive. However, hope that the technology will reduce the competitive importance of default placement may be misplaced. Mehta’s prognostication depended on the presumption that AI would produce better search-related products that would overcome Google’s default advantage. This presumption misunderstands that AI is allowing the search market to evolve in a way that may favor Google.
Indeed, AI has changed the search market, primarily through the introduction of AI-generated answers that users of chatbots and search engines receive in response to their queries. Google has built its own large language model (LLM), Gemini, to generate these answers through its chatbot agent and at the top of its search results page. Competing AI companies have developed their own LLMs to power their consumer-facing products, and they too have begun to generate answers grounded in search results in their chatbots. The difference is that these AI-native companies mainly rely on third-party infrastructure and generally lack Google’s web-scale index, freshness, and commercial/local coverage (the search stack) at comparable scale. Whether they originate from traditional search products or from standalone chatbots, these “answer engines” depend critically on the search stack that has given Google a tremendous advantage over its rivals. As such, Google’s control of the default access points may allow it to improve its own answer engine product faster than its AI-native competitors such as OpenAI and Anthropic.
What will determine the winner in this nascent answer-first search environment is which firm controls the default entry points that route the largest share of answer-first search and chat sessions (the “answer sessions”). Those sessions yield interaction signals that drive faster iterative improvement and, ultimately, higher-quality answers. That means answer quality depends heavily on distribution, and default placement that delivers scale has long been Google’s core advantage.
Answer-first search also changes the nature of the learning signals available. Whole search sessions, including follow-up questions and even purchases, can occur just by engaging with these AI-generated answers. This consolidated activity allows the platform that captures the most sessions to build more sophisticated answer engines, which can produce more helpful and comprehensive answers. Mehta was right that chatbots introduce competition. However, he underestimated how Google’s continued ability to purchase default status for its search engine, now offering AI-generated answers to users’ queries, may allow it to keep capturing the bulk of user data and signals, perpetuating its advantage even in this new era.
How answer-first search works – and why it advantages Google
To see why generative AI does not dissolve the competitive problem of Google Search’s default placement, it helps to separate the two layers of AI-powered search. One is the model layer: the LLM that can reason, summarize documents, and produce fluent responses. On that dimension, as Mehta identified, Google faces serious competition from OpenAI, Anthropic, and others. Indeed, by most proxy measures, OpenAI currently leads in standalone chatbot usage. Estimates of chatbot user traffic in the United States using domain-level web visits and app-based measures of mobile daily active users suggest ChatGPT has a larger usage base than Google’s Gemini.
However, the LLMs that power chatbots like ChatGPT and Claude are the same class of frontier models that generate the answers at the top of search pages, including Google’s AI Overviews and AI Mode. When one includes answer sessions that originate from traditional search, Google is still far ahead of its AI-native rivals. Google begins with far more query volume and can convert a growing share into answer sessions. Google reports processing more than five trillion searches a year—orders of magnitude more queries than any standalone chatbot. This matters because even if OpenAI, Anthropic and other competitors field LLMs that compete with Google’s, answer quality depends on the search stack around the model: web-scale indexing and retrieval, ranking, freshness, robust commercial and local coverage, and on capturing sufficient answer sessions to power the learning loop at scale.
These capabilities and signals form the second layer. LLMs alone cannot generate quality answers to queries unless they are paired with (or have reliable access to) this layer of infrastructure and feedback data that is continuously updated (and, for durable competition, monetizable). This second layer is where Google’s advantage remains, abetted by default placement which garners it the most traffic and data, and what risks allowing Google to cultivate a dominant answer-first search apparatus.
The remedies record backs up this distinction. In disputing the scope of Mehta’s remedies, Google has treated web-scale indexing and interaction data as strategically central. It has sought to stay provisions required by Mehta’s remedy that would expand rivals’ access to its index-related assets and user interaction logs, arguing the resulting loss would be “irreparable.” Whatever one thinks of that claim, it is an implicit acknowledgment that retrieval infrastructure and feedback data, not just model quality, are the key inputs for competition in an answer-first search market.
Apple’s Eddy Cue described the competitive path in similar terms: a generative product becomes a meaningful substitute when it solves the “search index” problem. The winning product is the hybrid that combines generation with the traditional elements of search: retrieval, ranking, freshness, and commercial-query handling. Google’s advantage in the traditional search stack lets it deliver more reliable grounded answers and improve its answer product faster, even when frontier model quality is contested.
There is another consideration in this calculus, and that is how a majority share of answer sessions will allow any market leader, Google or otherwise, to pull ahead of its rivals. The shift to answer-first interfaces introduces new richer learning signals. In traditional search, Google and other engines learned via pure navigation signals by observing which links users clicked on following specific queries. In answer-first search, when users receive eloquent, prose-like answers shown at the top of their search results, Google learns from session-level interaction signals. That is, learning signals that capture how the user interacts with the generated response: whether the answer fully satisfied intent, what clarifications were needed, and what downstream action followed. These signals aren’t uniquely available to Google; they accrue to whichever product captures answer sessions at scale, whether through a chatbot or a search page with AI Overviews/AI Mode.
Default placement therefore buys more than initial query volume. It buys the rich stream of session-level interaction data that drives iterative improvement. If Google is allowed to maintain default access across major browsers, Apple products, and Android devices, the competition Mehta anticipated may never materialize. Google will not only continue accruing the scale advantages that underpin traditional search quality, but also the session-level learning signals that drive next-generation answer engines.
Mehta’s remedy is mismatched for the answer-engine era
Mehta’s final judgment prohibited Google from continuing certain practices that entrenched its search monopoly. These included offering payments to partners conditioned on exclusively carrying Google’s key apps, including Search; on not carrying rival search or AI products; and bundling Google Search default arrangements across multiple products, devices, or access points. It does not, however, flatly ban Google from paying for default distribution.
Because Google can finance distribution payments from its dominance of online search (and related advertising revenues), it could bid up the price of default placement to a level that independent competitors such as OpenAI or Perplexity may not match sustainably. Google would not need to secure default status on all access points. It just needs to secure default status at enough high-volume points to continue to accrue the majority share of feedback signals that compound advantage.
If this happens, Google’s competitors may never reach the scale needed to improve their answer engines apace with Google. In an answer-first world, control of key distribution entry points, and the answer sessions they route, forms the gateway to the learning loop, which is a precondition for durable competition.
An empirical agenda for deciding whether remedies should be revisited
I have outlined the plausible risk that continued paid default placement for Google Search may entrench Google’s market power in AI-generated answer-first search. To determine empirically whether Mehta’s remedies are indeed failing to preserve competition in this new era, the analysis should ask four questions. First, does default placement have a meaningful and durable effect on answer-session share? Second, are answer-first search engines close-enough substitutes for standalone chatbots like ChatGPT or Perplexity, for at least a significant set of query categories? (Some chatbot answers do not rely on live retrieval from the web index, but for fresh, local, commercial, and long-tail queries, retrieval-grounded answers are often the competitive battleground.) Third, is Google bidding up default status beyond what competitors can sustainably afford? Fourth, does default-driven scale confer a compounding feedback advantage from the resulting interaction data and learning signals?
These questions can be studied with four metrics.
1. Default elasticity and persistence. How do changes in default placement shift answer-session share? In other words, when default status changes (say, from Google to Bing), how large is the shift in where users conduct answer sessions, and does it persist or quickly decay? Natural experiments (OS/browser updates, OEM deal changes, staged rollouts, policy interventions) can identify both the immediate effect and its persistence.
2. Substitution patterns: Which rivals are displaced by changes in default status, and in which query categories? Does Google’s AI Overviews/AI Mode divert users away from standalone chatbots like ChatGPT and Perplexity or just from traditional search rivals like Bing or DuckDuckGo? Is Google displacing searches in quick factual lookups and navigation, or also in money-making tasks such as shopping, local searches, and product research?
3. Default bid level. Even without formal exclusivity, default access can be competitively foreclosed if it functions like a winner-takes-most auction. The relevant price is what a company pays in total to be the default: upfront fees, minimum guarantees, and any revenue share. These reflect how prominent the placement is and how long it lasts. With standardized reporting (under a protective order), a remedies monitor could translate those deals into a rough cost per active user or per answer session and ask a simple question: is default access being priced at a level that independent rivals can’t sustainably afford?
4. Dynamic feedback advantage. Distribution buys more than traffic if it also buys the dominant stream of feedback signals: interaction traces, implicit and explicit ratings, and downstream outcomes that improve grounding, ranking, safety, and the overall answer system. A testable question is whether greater exposure to AI-enabled search produces measurable quality (and potentially monetization) improvements to an answer engine, and whether those gains plausibly depend on feedback signals that rivals cannot replicate at comparable scale.
Mehta’s opinion recognizes that remedies are not set-and-forget. Mehta wrote that judges must be willing to “clarify and reconsider” decrees as markets develop, and that he “is thus prepared to revisit a payment ban (or a lesser remedy)” if competition is not substantially restored. The decree should be revisited if the above metrics point in the wrong direction.
If an outright ban on Google purchasing default search status is off the table, one alternative is a cap-based approach, such as that proposed by Fiona Scott Morton and Paul Heidhues. This approach would prohibit exclusivity and contractual requirements for default or preferential placement, while allowing limited payments subject to meaningful constraints, including coverage and revenue-share caps and strong default-switching protections.
The broader policy point
The central risk is not that Mehta crafted remedies for “old search” and failed to anticipate “new search.” It is that his remedy effectively bets on advances in generative AI to restore competition in search while leaving largely intact the paid default mechanism that entrenched Google’s dominance. The issue is that Google is still best placed to exploit that mechanism in an answer-first era.
The opinion’s cautious posture contains its own corrective in Mehta’s willingness to revisit payment restrictions if competition does not substantially return. The task now is to make that contingency meaningful. “Competition restored” should not be assessed by whether search has adopted new features or whether potential rivals exist somewhere in the market. It should be assessed by whether changes in default placement cause durable shifts in where answer sessions begin; whether Google’s answer engine diverts high-value usage from standalone AI apps; whether the effective price of default access remains within rivals’ sustainable economics; and whether default-driven session share yields a compounding advantage in interaction data and learning signals.
Ultimately, the key question is what default payments purchase in practice: a neutral method of distribution or control over the learning loop that determines who develops the most sophisticated answer engine. If the latter, leaving the default-payment machine largely intact is not caution; it is an invitation for dominance to persist under the very competitive story meant to dissolve it.
Author Disclosure: In the past three years, the author has consulted for a client adverse to Google in traditional search and ad tech, and to date has not been involved in any consulting matter involving the AI search/chatbot (i.e., answer engine) market.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
Subscribe here for ProMarket’s weekly newsletter, Special Interest, to stay up to date on ProMarket’s coverage of the political economy and other content from the Stigler Center.
