Some legal cases age like wine. Others age like browser tabs left open too long.
Brazil’s Google News inquiry belongs firmly in the latter category. On April 3, Brazil’s Administrative Council for Economic Defense (CADE) announced that its Tribunal had unanimously decided to send a seven-year-old administrative inquiry concerning Google’s use of journalistic content—whether for indexing information, displaying “snippets,” or, more recently, generating “AI Overviews”—back to the agency’s General Superintendence (SG/CADE). This time, the Tribunal instructed SG/CADE to open a formal administrative proceeding and conduct a deeper investigation.
The shift followed the vote of acting President Diogo Thomson de Andrade (the “Thomson vote”) and a concurring opinion (voto-vista) by Commissioner Camila Cabral Pires-Alves (the “Pires-Alves vote”). The rapporteur, former Commissioner Gustavo Augusto Freitas de Lima, had originally voted to dismiss the matter, but later amended his vote to join the concurring opinion.
A unanimous vote may sound decisive. In this case, it is almost the opposite. Rather than close a stale inquiry launched in 2019, whose original factual premises have changed beyond recognition, CADE chose to keep the matter alive while substituting a new theory of harm and, in some respects, a new set of facts. There are several reasons to view that move with concern and skepticism.
A New Case in an Old File
A brief procedural recap is in order.
The inquiry traces back to another well-known Brazilian case: Google Scraping. CADE did not open the matter in 2019 through SG/CADE acting ex officio—that is, on its own initiative. Instead, the Tribunal itself launched the inquiry while adjudicating the Google Scraping proceeding.
The trigger was Globo/G1’s 2019 response to a request for information, or RFI. Issued by Brazil’s largest media conglomerate, the RFI asked whether Google had used Globo’s content on its news platform. Globo responded:
As for the period between 2011 and 2012, we know that G1 content appeared in search results, but we are unable to determine at this point [in 2019] how Google’s news platform worked at the time, nor what content was displayed. As for the subsequent period [2013 onwards], the answer is yes—the platform provided links, headlines, and photos of G1 content. However, we cannot say with certainty exactly when this content began to be displayed.
Globo’s central allegation was straightforward: Google News and Google Search displayed headlines and snippets derived from third-party news content without compensation. The claim was framed as both exclusionary and exploitative conduct. In antitrust terms, an exclusionary claim alleges that a firm harmed rivals or shut them out of the market; an exploitative claim alleges that a firm used its market power to impose unfair terms. In response, SG/CADE opened the inquiry through Order No. 4/2019.
After four years of investigation, CADE’s Department of Economic Studies (DEE/CADE) issued Technical Note 24/2023, concluding that the evidence failed to support any of the proposed theories of competitive harm. The findings were unequivocal: no incentive to foreclose rivals, no essential-facility concerns, and—crucially—a positive net flow of traffic from Google to publishers, with “insufficient evidence that snippets led consumers to stop accessing the links” (Thomson vote, ¶36). SG/CADE reached the same conclusion in Technical Note 70/2024, recommending dismissal for lack of evidence of an antitrust violation—or, put differently, lack of competitive harm (Thomson vote, ¶63).
Ordinarily, that would have ended the matter.
Instead, on March 28, 2025, Commissioner Camila Cabral Pires-Alves issued Decision No. 9/2025, proposing that the Tribunal revisit the case through a procedural mechanism known as avocação. Under Brazilian competition-law procedure, a commissioner who disagrees with SG/CADE’s recommendation may ask the Tribunal to substitute its own judgment for that of the investigative authority. In practice, avocação allows the Tribunal to review and potentially overturn SG/CADE decisions, including dismissals in conduct cases and unconditional approvals in merger reviews.
The Tribunal accepted Pires-Alves’ proposal and randomly assigned the case to Freitas de Lima as rapporteur (Thomson vote ¶¶66, 69). At CADE’s 249th Ordinary Judgment Session on June 11, 2025, Freitas de Lima voted to affirm SG/CADE’s recommendation of dismissal.
His reasoning tracked the agency’s earlier findings. Snippets, he concluded, “generate additional traffic for the portals, effectively functioning as a form of free publicity” (Thomson vote ¶80). A leveraging theory failed because Google “does not operate directly” in the downstream market for news portals (Thomson vote ¶81). The practice was justified under the rule of reason because indexing is “a central element of the business model of search engines, facilitating users’ rapid access to information and generating gains in efficiency and utility” (Thomson vote ¶82). Finally, economic dependence by publishers, standing alone, does not constitute a competition-law violation (Thomson vote ¶86).
By that point, DEE/CADE, SG/CADE, and the assigned rapporteur had all independently reached the same conclusion.
The case nevertheless took another turn.
At that same session, Acting President Diogo Thomson de Andrade filed a request for further consideration (Thomson vote ¶94) and launched what he described as a “supplementary investigation” (Section 2 of his vote). He issued additional decisions, sought further information from Google, sent RFIs to seven publisher associations (Thomson vote ¶99, Table 4), and explicitly invited “civil society in general to submit technical and factual contributions” (Thomson vote ¶101).
The effort generated 10 additional submissions from Brazilian media outlets—including Zero Hora, A Gazeta, and Folha de S.Paulo—as well as advocacy organizations such as the Open Markets Institute’s Center for Journalism & Liberty, Article 19 Brazil, Reporters Without Borders, Sleeping Giants Brasil, and Foxglove (Thomson vote ¶102, Table 5). Many of these organizations are sophisticated advocacy groups with established positions on platform regulation.
The timing matters.
Google launched AI Overviews in Brazil only in August 2024 (Thomson vote ¶154), after SG/CADE had already prepared Technical Note 70/2024 and recommended dismissal. The original 2019 inquiry therefore could not have generated evidence concerning AI Overviews. In substantive terms, Thomson’s supplementary investigation looks less like a continuation of the original case than a new investigation into a new product feature, attached to an old file whose original premises—scraping, snippets, and headlines—had already failed on the evidence.
Thomson is candid about the institutional rationale behind this approach. He writes that the supplementary investigation “was not guided solely by the aim of strengthening the evidence in the strict sense” (Thomson vote ¶111). Rather, it reflected “a broader institutional understanding of the need to increase the openness of competition policy to the involvement of civil society and of actors who, although not directly competing in the market under scrutiny, represent diffuse interests.” Later, he contrasts this approach with analysis driven primarily by technical specialists, arguing that doing otherwise would create “an artificial barrier to the expression of diverse perspectives” and risk “a kind of closure of the marketplace of ideas” (¶¶117-118).
That rationale raises important concerns.
Treating SG/CADE’s dismissal recommendation as the starting point for a year-long, advocacy-driven supplementary investigation risks transforming the agency from a competition-law enforcer into a stakeholder-management forum. There is, of course, a place for broad debate about competition policy. Competition-law proceedings serve a narrower purpose. They assess specific harms under defined legal standards, using evidence-based tools intended to reduce error costs—the risk that an agency either condemns lawful conduct or misses genuinely harmful conduct.
CADE should not hesitate to pursue cases when the evidence demonstrates harm to competition. But here, the evidence assembled by the agency’s own economic experts points in the opposite direction. Against that backdrop, the unusual duration and procedural trajectory of this case raise legitimate questions about the fairness and neutrality of the process.
The Return of the Phantom Menace
The most consequential—and ambitious—move in the case is substantive. Thomson acknowledges that CADE’s technical bodies have thoroughly examined the traditional theories of harm. In his view, the case deserves a second life through a different lens: exploitative abuse.
As Thomson explains:
Thus, the theories of harm already explored by the DEE/Cade – increased costs for rivals, leveraging of a dominant position, scraping of journalistic content that allows access to information without redirecting to publishers, and tying – as well as by the SG/Cade – predatory innovation and blocking of essential inputs, retention of traffic with a view to increasing advertising revenue, and self-preferencing in the event of traffic diversion or retention for one’s own benefit – and subsequently analysed in the opinion of the Reporting Commissioner, have, in my view, already received sufficient analytical treatment in the course of the proceedings and in previous decisions.
There remains, however, a need for a more detailed examination of the exploitative aspect of the conduct, even though this has been addressed in part in the aforementioned statements. For this reason, I shall focus here on the aspect of abuse of a dominant position that I consider relevant, specifically with regard to alleged exploitative abuse (or abuse of economic dependence), whether or not associated with elements of exclusionary abuse. (emphasis in original) (Thomson vote, ¶¶ 502-503).
Thomson reaches this conclusion through a particular reading of Article 36 of Brazil’s Competition Law (Law No. 12,529/2011). Having acknowledged that the conventional abuse theories failed on the evidence, he argues that Brazilian competition law nevertheless reaches the conduct through what both he and Pires-Alves call the “typological openness” of Article 36—that is, the statute’s non-exhaustive list of anticompetitive practices.
Pires-Alves expressly embraces this approach. “I concur with Commissioner Diogo [Thomson]: it seems more appropriate to frame the theory of harm, at the forefront, as exploitative” (Pires-Alves vote ¶10). She notes that “the trajectory of exploitative abuses in [Brazil] was marked more by institutional caution than by conceptual rejection” (Pires-Alves vote ¶11, citing Barbosa & Kastrup, 2021) and argues that Article 36’s typological openness permits scrutiny of “unilaterally imposed unfair conditions on dependent partners, including when the extraction of value does not manifest in the form of direct monetary price” (Pires-Alves vote ¶11).
Pires-Alves also acknowledges that the exclusionary case has largely failed. In her view, that failure does not doom the exploitative theory:
The current absence of a stronger showing of exclusion does not neutralize the exploitative theory of harm. It is preferable to acknowledge that the issue admits a primarily exploitative reading, without ruling out possible exclusionary effects, and that the current state of the investigation does not yet point to a definitive conclusion in this regard. (Pires-Alves vote ¶14).
That argument faces a significant obstacle: exploitative-abuse cases are competition law’s phantom menace. They are theoretically available in many jurisdictions, but rarely enforced in practice.
European Union law permits exploitative-abuse claims under Article 102(a) of the Treaty on the Functioning of the European Union (TFEU). Yet the European Commission’s own enforcement guidance prioritizes exclusionary conduct, and exploitative-abuse cases remain uncommon. U.S. antitrust law has gone further. Since the U.S. Supreme Court’s 2004 decision in Verizon Communications Inc. v. Law Offices of Curtis V. Trinko, LLP, monopoly pricing itself has generally been treated as lawful absent anticompetitive conduct. As the Court explained:
The mere possession of monopoly power, and the concomitant charging of monopoly prices, is not only not unlawful; it is an important element of the free-market system. The opportunity to charge monopoly prices—at least for a short period—is what attracts “business acumen” in the first place; it induces risk taking that produces innovation and economic growth. To safeguard the incentive to innovate, the possession of monopoly power will not be found unlawful unless it is accompanied by an element of anticompetitive conduct.
In Justice Antonin Scalia’s view—and, we would argue, the correct one—the prospect of earning monopoly profits is a feature, not a bug, of a competitive economy. The possibility of charging supracompetitive prices rewards risk-taking, investment, and innovation. Brazilian law reflects a similar principle. Article 36, §1 of the Brazilian Competition Law provides that “[a]chieving dominance in a market by natural process and by being the most efficient economic agent in relation to competitors does not characterize” an antitrust violation. Antitrust enforcement should therefore focus on supracompetitive prices achieved through anticompetitive conduct, not those resulting from superior efficiency, innovation, or business acumen.
Even setting that debate aside, exploitative abuse is notoriously difficult to define and prove. Drawing a principled line between a “reasonable” price and an “excessive” one—or determining the “fair” allocation of economic benefits—is no easy task. As Massimo Motta and Alexandre de Streel observe, “the proof of an excessive price, or in other words the search for the competitive price, may be like a quest for the Holy Grail.”
For that reason, even scholars who support the use of antitrust law against excessive pricing generally regard it as a remedy of last resort. Motta and De Streel argue:
It is important that in those guidelines [referencing the European Commission’s Guidelines on Article 82, now Article 102 TFEU], the Commission commits itself to limit the use of its broad legal power and make explicit that excessive pricing actions should be an option of last resort for antitrust authorities, and that they should be used only when other routes fail. (emphasis added and footnotes omitted) (Motta & De Streel, p. 42).
The Thomson-Pires-Alves approach points Brazilian law in the opposite direction.
As Giuseppe Colangelo has explained, the “free-riding” narrative underlying many claims by newspapers and other media organizations against digital platforms “serves chiefly as a rhetorical tool, strategically used to legitimize a mandated transfer of revenue that policymakers appear ready to pursue even in the absence of clear evidence that platform-based business models have caused economic harm to press publishers.”
Once that theory is accepted—and once Article 36’s “typological openness,” combined with notions of “structural dependence,” becomes a substitute for traditional dominance-and-foreclosure analysis—the doctrine is unlikely to stop with Google. Any platform with allegedly dependent counterparties, a category broad enough to encompass countless ordinary commercial relationships, could become a target for exploitative-abuse claims untethered from measurable competitive effects.
As Craig Conrath has explained:
[I]t is necessary to distinguish between contract power and monopoly power, between post-contractual opportunism and abuse of dominant position. … Therefore, whenever there is a claim of monopolistic practice by a party in a contractual relationship with the person accused of a monopolistic practice, it is important not to ask the question, “Does this party now have any other choices?” This is a question of contract law. Instead, one should ask the relevant antimonopoly law question: When the contract was made, did this party have other choices? (Practical Handbook Of Antimonopoly Law Enforcement For An Economy In Transition, § 5, at 5 (1995)).
That distinction matters here. The mere fact that a business has become economically dependent on a successful platform does not transform a contractual dispute into an antitrust violation.
It is also worth remembering what platforms such as Google News and other content aggregators actually do. They collect, organize, and direct users to relevant information, reducing search costs—the time and effort consumers would otherwise spend finding news and other content on their own. That benefit should not vanish from the analysis simply because some publishers dislike the commercial terms of the relationship.
Dominance Without a Market
Even if one accepts the direction this case has taken, a threshold question remains: In what market, exactly, is Google supposed to be dominant for purposes of an AI Overviews investigation?
There are at least three possible answers, and the case looks weak under all of them.
One possibility is that AI Overviews are simply a new feature within Google Search. On that view, the relevant market remains general search, where Google’s share in Brazil has exceeded 90% for years, albeit with some recent erosion. If so, the AI-related concerns are merely an extension of the original snippets-and-scraping investigation.
The problem is that this framing ignores the competitive threat that generative artificial intelligence poses to Google itself. In the remedies phase of United States v. Google LLC, after finding that Google possessed monopoly power in general search, Judge Amit Mehta declined to impose structural remedies, such as divestiture of Chrome or Android. A key reason was the emergence of generative AI. Mehta concluded that AI had “changed the course of this case” and posed a meaningful threat to Google’s position. Against that backdrop, it is difficult to reconcile a theory that treats Google’s efforts to compete in AI as evidence of deepening monopoly power.
A second possibility is that AI Overviews compete in a distinct market for generative-AI assistants and conversational search tools. Users who seek information through ChatGPT, Claude, Perplexity, Grok, or Gemini are often making functionally similar choices.
In that market, Google is not dominant. It is a challenger. Gemini has gained traction, but it still trails ChatGPT and faces strong competition from Anthropic, Perplexity, xAI, and others. More importantly, the market itself remains highly dynamic. Google, Microsoft, OpenAI, Anthropic, Perplexity, and other firms are rapidly experimenting with how AI-driven search works, how sources are cited, how referral traffic is generated, and how these products are monetized.
Treating any particular implementation as the basis for an antitrust proceeding risks freezing a temporary design choice into law. In markets evolving this quickly, false positives are especially costly. Regulators may deter features that consumers value, publishers benefit from, and competitors do not challenge in practice.
Recent empirical evidence reinforces this point. In “The Emerging Market for Intelligence” (2025), Aydin Demirer, Andrey Fradkin, Steven Tadelis, and Yiming Peng document extraordinary turnover in generative-AI markets. Drawing on usage data from OpenRouter and Microsoft Azure, they find that the number of available large-language models grew from 253 to 651 in 2025 alone, while the number of model creators nearly doubled and the number of inference providers more than tripled. Intelligence-adjusted prices at the technological frontier fell by roughly 1,000x since 2023. Leadership shifted repeatedly among Anthropic, Google, and xAI models throughout the year, and “the top 10 models today accounted for just 20% of market share four months ago and did not even exist ten months ago.” Their conclusion is straightforward: “No single model dominates across use cases.”
That is not what an entrenched market looks like.
A third possibility is a broader market definition encompassing search engines, AI assistants, social media, email, and other channels through which publishers reach readers. From the perspective of news organizations, the relevant question is not whether Google exists in isolation, but whether publishers have meaningful alternatives for distribution and audience acquisition.
The Thomson vote attempts to avoid this inquiry through a dependence-based framework:
In my view, the core of the case does not lie in precisely and exhaustively determining whether the relevant market should be defined, with mathematical precision, as general search, thematic news search, digital content distribution, or any other intermediate formulation. The core issue is to determine whether, regardless of the formal market definition adopted, publishers operate in a position of structural dependence on Google for purposes of discovery, distribution, monetization, and the management of their own economic risk.
This emphasis on dependence, rather than definitional formalism, is consistent with Padilla’s (2024) approach. If the test for exploitative abuse is to focus on the harm caused by dominance, it would make no sense to require, as an absolute prerequisite, an exhaustive mapping of all digital attention channels. The legally relevant point is another: whether the dominant firm is capable of imposing terms that the trading partner would not accept in a minimally competitive environment. (Thomson vote, ¶¶591-592)
That approach raises serious concerns.
For one thing, Brazilian law still requires dominance to establish abuse of dominance. Unlike jurisdictions such as Germany or Japan, Brazil has no standalone doctrine of abuse of economic dependence. More fundamentally, the reasoning risks becoming circular. The challenged contractual arrangement becomes evidence both of the firm’s alleged market power and of the harm that power supposedly produces. Competition law ordinarily requires separate proof of market power, improper conduct, and competitive harm.
The Case for Closing the Case
Commissioner Gustavo Augusto Freitas de Lima’s approach would likely have produced the better outcome.
As noted above, Freitas de Lima—the case’s original rapporteur—initially voted to uphold SG/CADE’s recommendation of dismissal, relying heavily on the DEE’s technical analysis, which found no evidence of competitive harm arising from the challenged conduct. Following Thomson’s supplementary investigation, Freitas de Lima modified his position in one important respect. He agreed that AI Overviews might warrant scrutiny, but as a distinct investigation separate from—albeit related to—the original Google News matter. He explained:
§ 133. Accordingly, following the analysis conducted based on the submissions provided and taking into account the study prepared by the DEE, I find that there is no evidence of a material violation of the economic order with respect to the conduct originally investigated, which consisted of the collection, indexing, and display of journalistic content through snippets and related mechanisms in the search engine.
§ 134. I emphasize, however, that the use of artificial intelligence tools integrated into the search engine may imply a significant functional difference in the conduct, with potential competitive impacts distinct from those observed in the traditional search model. Considering the evidence gathered, and without delving into the merits of the matter, I believe there is sufficient evidence to justify the initiation of a new investigation to specifically address the innovations introduced by Google’s Artificial Intelligence Overview (AI Overview).
§ 135. In light of the foregoing, I vote to dismiss the present Administrative Inquiry regarding the original conduct, pursuant to Article 144, §3, I, of RICADE [CADE’s Internal Regulation] and Article 67, §2, I, of Law No. 12,1529/11. As for the conduct related to the use of artificial intelligence in the search ecosystem (AI Overview), I concur with the proposal contained in the dissenting opinion of Commissioner Diogo Thomson, under the terms outlined therein, understanding that the case should be returned to the SG for the opening of a new specific investigation.
If CADE genuinely believes that Google’s deployment of AI Overviews in Brazil raises competition concerns, then the procedurally sound response is the one Freitas de Lima outlined—not keeping a 2019 inquiry on procedural life support.
CADE should close the original inquiry. The factual premises that gave rise to it have been overtaken by technological change, and the evidentiary record assembled over seven years supports dismissal. Closing the case would not prevent future enforcement. It would simply require the agency to articulate a new theory of harm, define the relevant market, and build a new evidentiary record tailored to the conduct actually under scrutiny.
CADE should also resist the temptation to embrace a broad exploitative-abuse doctrine. Even in jurisdictions that recognize such claims, exploitative abuse remains a tool of last resort. It is difficult to define, difficult to prove, and even more difficult to remedy in a predictable and administrable way.
AI Overviews may or may not raise genuine competition concerns. That question deserves serious analysis. But if CADE wants to investigate the future of search, it should start a new case—not keep rewriting an old one.
