The Antitrust Case Against AI Overviews
By Madhavi Singh - Edited By Shriya Srikanth
Madhavi Singh is the Deputy Director of the Thurman Arnold Project and a Resident Fellow at the Information Society Project at Yale Law School. She was previously a Research Associate at the National University of Singapore and a Visiting Lecturer at the National Law School of India University. She holds law degrees from Harvard (LL.M.), Oxford (BCL), and the National Law School of India University (B.A. LL.B.).
Introduction
AI-powered answer engines (in the form of AI chatbots and other generative features in search) are fast replacing search engines as the gateway to the internet. Google understands these winds of change and is trying to avoid becoming redundant in this rapidly changing market by aggressively building a flurry of AI products like AI Overviews, AI Mode, and Gemini. It is also dedicating a large (and increasing) percentage of the search engine results page to AI Overviews. Google’s acquisition of DeepMind, its pseudo-acquisition of Character AI, and its large investments in Anthropic and many others are also part of its bigger strategy to co-opt creative destruction rather than succumb to it.
Although other Big Tech companies are also racing to expand their AI portfolios, what sets Google apart is that it has already been found to have engaged in illegal monopolization of the search market.
[[1]] We are now witnessing it leveraging this illegally maintained monopoly position to gain a foothold in and monopolize the market for AI-powered answer engines too. This essay argues that such acts of leveraging constitute an antitrust violation. Antitrust enforcement should strive to prevent this second sin, which would compound Google’s first and facilitate the continuity of its monopolization strategies.
The offense of monopolization under section 2 of the Sherman Act requires proof of two elements: (1) the defendant must possess monopoly power in the relevant market; and (2) there must be willful acquisition or maintenance of monopoly power through exclusionary conduct.[[2]] In the recent Google search antitrust case, the District Court for the District of Columbia provided a detailed analysis establishing that Google holds monopoly power in the relevant market for general search services.[[3]] As the court has comprehensively addressed this issue, the article does not examine it further.
The second element of proving exclusionary conduct under a rule of reason analysis consists of a four-step burden-shifting framework: first, the plaintiff bears the burden to show anti-competitive effect; second, the burden shifts to the defendant to prove the existence of pro-competitive justifications; third, the burden shifts back to the plaintiff to demonstrate that these procompetitive objectives could be achieved through less restrictive means; and finally, the court balances the conduct’s procompetitive and anticompetitive effects.[[4]] Google’s exclusionary conduct consists of denying publishers and content creators the 3 Cs (consent, credit and compensation). Google scrapes websites without consent or compensation to train its AI models, and publishers have no recourse to resist such behaviour, given Google’s criticality in ensuring the discovery and distribution of their content. It then uses this non-consensually scraped data to generate AI-powered responses, which sometimes fail to attribute or misattribute sources. Finally, it prominently placesthese AI-generated answers at the top of the search results page, assigning them an ever-expanding percentage of the screen.
Anti-competitive effects
The anti-competitive effects of this exclusionary conduct are fast becoming apparent. The number of zero-click searches has increased significantly as more queries are being answered through AI Overviews. According to one study, about 80% of search users rely on AI summaries at least 40% of the time, and about 60% of searches result in zero clicks. Although AI summaries have started including some citations to sources, click-through rates have still drastically declined. The growing popularity of AI Overviews and chatbots as a mainstream way of answering queries has been accompanied by a concomitant decline in website traffic (by almost 34.5%) and revenue for publishers. This shift from a search-driven web to an AI-driven web has made it extremely difficult for creators to monetize their content and threatens to break the internet. Publishers who were already enfeebled in an ad-driven ecosystem where Google and Facebook controlled discovery, distribution and monetization of content, risk being further crippled by the proliferation of AI Overviews.
Pro-competitive justifications
Undoubtedly, Google will offer some pro-competitive justifications for its exclusionary conduct. For example, it could try to justify the integration of AI Mode and AI Overview as a simple product design change aimed at improving the functionality and value of its search function by adding new features. Notably, Microsoft in its antitrust case advanced a similar argument that the various acts to integrate Internet Explorer in the Windows Operating System were a design choice or product improvement.[[5]] However, the court rejected this argument, noting that a monopolist’s design choices are not per se lawful and would still be subject to antitrust scrutiny. Further, Google could argue that the non-consensual scraping of publisher data (some categories of which have recently been deemed to be “fair use” under copyright law)[[6]] also has pro-competitive justifications, since such data scraping is the sine qua non for training generative AI models, which in turn improve consumer experience and increase consumer welfare.
Proof of any pro-competitive effects would then have to be assessed against the benchmark of whether these could be achieved through less restrictive means, and a final step of balancing the pro and anti-competitive effects.[[7]] All of these, however, are fact-intensive exercises requiring large swathes of empirical and other evidence and therefore, could only get a fair shake at trial. At this stage, though the anti-competitive effects of Google’s exclusionary conduct are abundantly clear and therefore, the threshold for a regulator to investigate this matter has been reached.
Tying
Another theory of harm which could encapsulate Google’s exclusionary conduct is tying. Tying cases can be brought under section 1 or section 2 of the Sherman Act (or under section 3 of the Clayton Act, in case of physical goods).[[8]] This must be accompanied by the caveat that tying law in the US is a wreck and needs to be reconceptualizedto reflect the realities of the digital age. Indeed, the tying claim in both Microsoft and Epic cases were unsuccessful, even though the same conduct was found illegal under the monopolization claim.[[9]] Nevertheless, from that jurisprudential wreckage, we can extract the following elements of a typical tying claim: (1) the defendant tied together the sale of two distinct products or services; (2) the sale of the tying product is conditioned on the sale of the tied product; (3) the seller has appreciable economic power in the tying product market; and (4) the tying affects a not insubstantial volume of commerce in the tied product market.[[10]] The bundling of AI-powered answer engines (such as AI Mode and AI Overview) with Google’s search engine satisfies all four elements. Evidence of the final two elements, namely, market power and anticompetitive effects, has been discussed above.
Distinct products
To ascertain the distinctiveness of the products, courts apply a consumer-demand test to assess whether there is sufficient consumer demand to make it efficient for a firm to offer the two products separately.[[11]] In this case, the tying product i.e., general search services has been offered as a standalone product for a long time as exemplified by platforms like Google, Bing and DuckDuckGo. Similarly, there are several providers of the tied product i.e., AI-powered answer engines like ChatGPT, Perplexity, and Claude, that provide these as a standalone product. Indeed, Google itself offers Gemini as a standalone product, in addition to integrating it into Google search through AI Overview and AI Mode. Additionally, the court in the Google Search case considered whether generative AI (or AI-powered answer engines) serve as substitutes for general search. It concluded that general search services constitute a distinct relevant market, noting that generative AI has not yet eliminated its need.[[12]] Thus, the two are distinct products which satisfy the first prong of the tying test.
Conditioning/ forcing
The second prong of tying, which requires that the sale of the tying product is conditioned on the sale of the tied product, is trickier. It is clear that Google bundles both products together; that is, Google’s search feature automatically produces an AI Overview. The AI Mode button is also available by default on the Google search page (just below the search bar alongside the Images, Videos, News, and other buttons) in the countries in which the feature has been rolled out. Unfortunately, some courts have required proof of “forcing” or “coercion” for a tying arrangement to be deemed illegal.[[13]] Courts vary in their standards for establishing coercion: some consider proof of market power in the tying market sufficient;[[14]] others examine whether the bundle was the buyer’s “only viable economic option;”[[15]] and still others assess voluntariness based on whether the buyer had knowledge of and accepted the tie.[[16]]
There are still unanswered questions about whether modern manifestations of tying in the form of default placement, product integration or design nudges are sufficient to establish conditional sale/ coercion or whether evidence of something more forceful is required. For instance, Google could argue that there is no coercion (narrowly construed) because users could simply scroll past the AI Overview on the results page or choose not to use the AI Mode. This would be a very narrow construction of the requirement, though, because there is ample evidence to indicate that even without express coercion, subtle behavioral manipulations like nudges, default placements, etc. have the same effect. As discussed above, even though users could scroll past the AI Overview, its premium placement at the top of the results page invariably results in lower click-through rates for all websites. Fortunately, as witnessed in the Google search case, courts have started becoming sophisticated in their analysis of subtle behavioural manipulation and accounting for concepts like default bias, status quo bias, stickiness, etc. in assessing the effects of exclusionary conduct.[[17]]
Other jurisdictions like the EU have adapted their law to accommodate this: in the EU, if consumers do not have the choice to obtain the tying product without the tied product, then that is sufficient to show conditioning or coercion.[[18]] Additionally, instances of technological tying and bundling through product design and integration have become so pervasive that antitrust law must eventually confront how tying doctrine should be adapted to address these developments. The time is therefore ripe to align tying law with technological realities, and this case presents an ideal opportunity to do so.
Echoes of Microsoft
Although it has become somewhat of a cliché to invoke Microsoft as a historical lesson for every platform antitrust case, there are striking similarities between that historical moment and Google’s attempts here. Firstly, there are factual similarities between the two cases. The Microsoft case stemmed from the integration of Internet Explorer into the Windows operating system which Microsoft argued was a product design choice that improved functionality and enhanced user experience. Similarly, here, one of the causes for concern (apart from non-consensual data scraping, lack of attribution and compensation) is that Google has integrated AI Overview and AI Mode into its search home page, which Google would surely try to defend on similar grounds.
Secondly, and more importantly, both cases involve a monopoly’s attempt to extend its control from one generation of information distribution to the next. As the famous Gates’ memo (unearthed in the prolonged antitrust trial) outlined, Microsoft in the 1990s was at an inflection point. The true power of the Windows OS was that it was the platform on which all apps were installed, and it served as the site for all user activity (like listening to music, watching movies, and most productivity tasks). With the rise of the internet, Windows OS was about to be dethroned by browsers, as users switched from using apps locally on their PC to simply executing all tasks directly on the internet through their browsers. Losing its status as the gatekeeper of the dominant ecosystem in which all tasks are executed is any monopoly’s nightmare. Google is now faced with the same threat. Google and its suite of websites (including YouTube) are amongst the most-visited websites in the world. Today, Google search is the starting point for most user queries (nearly 93.57%) and is considered the gateway to the internet. However, AI-powered answer engines are expected to dethrone Google search soon, becoming the next gateway for information access. Just as the Gates’ memo revealed Microsoft’s perception of the internet as an urgent existential threat and the company’s intention to quell and conquer even this new ecosystem, Google’s declaration of Code Red with the release of ChatGPT should give us a sense of déjà vu.
The invocation of the Microsoft antitrust case and its parallels to the present moment is meant to underscore the urgency of bringing prompt antitrust action against Google. This is not merely another instance of a dominant firm leveraging its monopoly to gain power in an adjacent market; what is at stake is the future architecture of the internet and the next generation of technology. The Justice Department’s antitrust suit against Microsoft at that seminal moment has been credited with unleashing innovation and fostering a vibrant and competitive (at least temporarily) ecosystem. Similarly, timely action against Google is essential to ensure that the emerging gateway technologies are not captured by the platform monopolies of the previous era. That Google is a repeat offender only heightens the concern; entrusting the future of the internet to such a company is a profoundly troubling prospect.[[19]]
Conclusion
Modern antitrust scholarship about Big Tech often bemoans the failure of enforcement to act swiftly to prevent rather than remediate monopolization and associated loss of innovation. The remedies phase of the Google search trial has also exposed the difficulty in restoring competition that is once lost. With AI, we are once again at a crossroads where antitrust could play a role in preventing monopolization of emerging technology. Several senators have also urged the Department of Justice and the Federal Trade Commission to investigate the competition risks posed by generative AI features.
Notably, private actors who are the most directly affected have started taking action too. Cloudflare, one of the biggest content delivery networks in the world (responsible for routing roughly 16% of global internet traffic), recently announced that it would block AI crawlers by default and use a new ‘pay per crawl’ model. Penske Media, the publisher of Rolling Stone, The Hollywood Reporter, and several other major outlets, recently filed an antitrust lawsuit against Google, alleging that it unlawfully used and republished content from their websites to train its generative AI programs — conduct that, they argue, amounts to illegal monopolization. Chegg, an online education company, has also brought a suitclaiming that Google’s AI Overviews have caused a drop in traffic and revenue. In the EU too, a group of independent publishers have brought a competition complaint and demanded an interim measure to prevent irreparable harm from Google’s AI Overviews.
However, given the scale and significance of the problem, private actions alone are insufficient. Regulatory intervention is urgently needed. Even if a case were to take years to reach trial and produce a ruling on liability and remedies, the mere initiation of an investigation could exert meaningful pressure on Google to curb its monopoly-maintenance strategies or even produce a settlement. A substantial body of modern antitrust scholarship, coupled with the recent flexing of regulatory muscle in Big Tech cases, has been preparing us for precisely this moment: when antitrust enforcement might finally arrive in time for the party, rather than showing up several years late, after most of the damage is done.
[[1]] United States v. Google LLC, 747 F.Supp.3d 1 (D.D.C. 2024) [Google search case].
[[2]] United States v. Grinnell Corp., 384 U.S. 563 (1966).
[[3]] United States v. Google LLC, 747 F.Supp.3d 1 (D.D.C. 2024).
[[4]] United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001).
[[5]] United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001).
[[6]] Bartz v. Anthropic, 3:24-cv-05417, (N.D. Cal. Jun 23, 2025); Richard Kadrey v. Meta Platforms, 3:23-cv-03417, (N.D. Cal. Jun 25, 2025).
[[7]] United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001).
[[8]] Int’l Bus. Machines Corp. v. United States, 298 U.S. 131 (1936).
[[9]] United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001); Epic Ganes v. Apple, Inc., 67 F.4th 946 (9th Cir. 2023).
[[10]] Jefferson Parish Hosp. Dist. No 2 v. Hyde, 466 U.S. 2 (1984).
[[11]] Eastman Kodak Co. v. Image Tech. Servs., Inc., 504 U.S. 451 (1992); Jefferson Parish Hosp. Dist. No. 2 v. Hyde, 466 U.S. 2 (1984).
[[12]] United States v. Google LLC, 747 F.Supp.3d 1 (D.D.C. 2024).
[[13]] Jefferson Par. Hosp. Dist. No. 2 v. Hyde, 466 U.S. 2, 12 (1984).
[[14]] Tic-X-Press, Inc. v. Omni Promotions Co. of Ga., 815 F.2d 1407 (11th Cir. 1987); Kentucky Fried Chicken Corp. v. Diversified Packaging Corp., 549 F.2d 368 (5th Cir. 1977).
[[15]] Amerinet, Inc. v. Xerox Corp., 972 F.2d 1483 (8th Cir. 1992).
[[16]] Suburban Propane v. Proctor Gas, Inc., 953 F.2d 780 (2d Cir. 1992).
[[17]] United States v. Google LLC, 747 F.Supp.3d 1 (D.D.C. 2024).
[[18]] Case T-201/04, Microsoft Corp. v. Comm'n, 2007 E.C.R. 11-3601 (Ct. First Instance).
[[19]] United States v. Google LLC, 747 F.Supp.3d 1 (D.D.C. 2024); United States v. Google LLC, 2025 WL 1132012 (E.D. Va 2025).