(By You Yunting) Since the advent of the commercial internet, traditional media outlets has developed two principal revenue models centered on user clicks: ad revenue and paid subscriptions. However, the emergence of AI-powered search, particularly Retrieval-Augmented Generation (RAG), enables users to obtain answers directly without clicking through to source websites. This structural shift has triggered widespread litigation against RAG-based AI platforms, ranging from copyright infringement claims to antitrust actions. For many incumbents, litigation has become the default recourse.
1. RAG Technology Undermines the Ecosystem of Traditional Media Websites
In their capacity as copyright holders, traditional media argue that, unlike conventional large language models, RAG-based search results eliminate users’ need to visit source websites. Where there is no page visits, there is no ad revenue and paid subscriptions, cutting off the economic lifeline for content creators. From their perspective, AI search is not innovation but systematic misappropriation of original content, destroying the creative ecosystem.
Central to this disruption is the RAG model. Conventional large language models, such as earlier iterations of OpenAI’s systems, are trained on massive datasets to internalize statistical patterns and thereafter generate new texts based on probabilities, seldom reproducing entire articles from their training materials. Under the RAG model, the AI search, upon receiving a user’s query, conducts real time web retrieval, locates relevant content and then integrates it into the generated response. If the AI platform implement no output controls, the resulting text may closely mirror the retrieved documents, potentially constituting an actionable act of reproduction under copyright law.
Even where the AI platform employs safeguards to render the output sufficiently transformative, falling outside the scope of an actionable act of reproduction, actually the AI-generated answer page can still substitute for the source websites producing original content. In most instances, users obtain the desired information without ever visiting the source websites, materially impairing the commercial ecosystem of the content-producing websites.
2. CNN Sues the “Barbarian” Perplexity
Perplexity is a generative artificial intelligence (“GenAI”) provider founded by an entrepreneur of Indian descent. Its relatively weak foundational model appears to have driven an aggressive approach to RAG deployment, conflicting with established legal rules and industry norms. In May 2026, CNN initiated litigation against Perplexity in the United States District Court for the Southern District of New York. According to the complaint, describing Perplexity’s business practices as unfettered expansion is fully justified.
Perplexity employs its own crawler to scrape CNN’s news articles, images, and videos on a massive scale, creating a comprehensive index. When users submit queries to Perplexity, the system retrieves full texts of relevant CNN articles, feeds the contents into the model’s context window and generates replies largely copied verbatim from CNN’s original articles, even on a paragraph-by-paragraph basis.
CNN’s complaint includes extensive side-by-side comparison exhibits: CNN’s original text on the left and Perplexity’s output on the right, with substantial portions highlighted in red to demonstrate near-verbatim correspondence between the two. For example, in response to the query “which position did Rubio drop?”, Perplexity’s paid Pro version reproduced verbatim excerpts of a CNN article. In other words, users access CNN’s copyrighted content by paying Perplexity, without any compensation accruing to CNN itself. Perplexity’s Comet browser even circumvents CNN’s paywall restrictions and delivers the complete text of subscriber-exclusive articles verbatim to non-subscribers.
The complaint further identifies two categories of malicious technical practices by Perplexity. First, Perplexity’s crawlers, in disregard of robots.txt rules, blatantly scraped content blocked from automated access, including paywalled news articles. Second, Cloudflare detected that once intercepted, Perplexity’s crawlers disguised themselves as ordinary browsers, such as mimicking a Chrome user on a Mac, and sent requests from undisclosed IP ranges.
3. Even “Polite” Google Got Sued
Compared with Perplexity, Google has operated with greater compliance. Its early operational philosophy was to redirect users from the search engine to source websites, fostering a comparatively balanced referral ecosystem. Google presented only two to three lines of brief text snippets, requiring users to click through to the source website for substantive content. This structure enabled media outlets to monetize through ad revenue and paid subscriptions attributable to Google referral traffic.
But in the AI era, Google has transitioned toward an answer-engine model that retains users within its own interface. In recent years, it has aggressively deployed AI search features capable of delivering AI-generated responses immediately upon users’ queries. Positioned above organic results, these AI-generated answers have materially diminished referral traffic to the results from search recommendation—click-through rates for top-ranked publishers have declined markedly, and a significant volume of publishers have concluded without any outbound clicks.
That said, Google’s AI search remains more compliant than Perplexity’s. Google respects robots.txt rules, refraining from indexing or summarizing content from websites that have opted out of crawling. Google’s AI search employs a modified RAG model: its output extracts fragments from multiple sources, with inline citations linking back to each source. The user experience largely maintains the traditional search framework of citations and links. Moreover, Google avoids revealing full paywalled content in AI summaries, instead guiding users to subscribe or log in on source websites.
Nevertheless, Google’s original search ecosystem has still been disrupted by AI summaries, triggering an inevitable and material decline in visitor traffic to traditional media websites. As the overall pool of referral traffic shrinks, someone must be held responsible. In September 2025, Penske Media Corporation, the owner of iconic media brands including Billboard, Rolling Stone, Variety, and The Hollywood Reporter, sued Google in the U.S. District Court for the District of Columbia, relying not on copyright law, but rather antitrust statutes.
The central theory of Penske’s complaint is as follows: media outlets originally consented to Google’s crawling their website content for the expectation of referral traffic. Google, however, unilaterally repurposed such content for its AI search products. Although Google offers an opt out mechanism through nosnippet tags, exercising this opt-out imposes disproportionate punitive consequences—Google will suppress snippets for both AI search and organic search results, causing the websites to completely lose their exposure and recommendation on Google and be effectively expelled from the Google search ecosystem. For media outlets, exclusion from Google search is commercially fatal, eliminating the vast majority of its inbound traffic.
This scenario presents a classic antitrust concern: a monopolist using its dominant market position to force trading partners to accept unfair terms. Accordingly, Penske had to invoke antitrust litigation as its primary remedy. According to the complaint, it claimed three relevant markets: (1) the general search market, where Google has already been found by U.S. courts to have a monopoly in other cases; (2) the market for online digital publishing traffic distribution, upon which media outlets depend for referral traffic; and (3) the AI search market, where Google is trying to extend its original search dominance into the AI era.
Penske argues that Google has abused its dominant position in the search market through tying arrangements (linking search index access to AI content licensing) and self-preferencing (prioritizing its own AI snippets over links to original websites in search results), thereby intentionally building barriers that rivals cannot overcome and undermining the previously balanced online content ecosystem.
Every technological advance boosts efficiency, and RAG-powered search is no exception, making users skip clicks for instant answers. Yet it disrupts the existing content creation ecosystem and unreasonably harms the interests of content creators, shrinking original content output in the long run. Lawsuits therefore serve as a necessary instrument to re-examine both the RAG model and the value distribution mechanisms behind neutral technologies.
It is necessary to observe a comparative shortcoming in China. U.S. judicial proceedings first addresses whether unauthorized training amounts to copyright infringement before turning to the legal characterization of the RAG model itself. By contrast, Chinese courts have adopted a markedly prudent and restrained posture toward training-related disputes, apparently to protect the emerging AI industry. Consequently, Chinese intellectual property practitioners like myself have no choice but to refer to U.S. cases for research.
Disclaimer of Bridge IP Law Commentary
Short Link: