A copyright test is emerging in AI scraping: not whether content was used, but whether the outputs harmed creators’ businesses.
AI scraping is no longer just a legal debate about training data; it is turning into its own media business. built on the promise of turning publisher content into profitable “outputs.” And in today’s courtroom battles. the hardest question for plaintiffs may be proving harm from what those systems actually produce.
At the center of the ongoing legal war between media companies and AI firms is copyright and. specifically. the question of outputs.. Scraping content without permission is widely viewed as unacceptable. but many civil claims hinge on showing that what comes out of the process competes with the original creator.. If the scraping doesn’t result in behavior that directly undermines a publisher’s business. plaintiffs often face a tougher burden of proof.
One early ruling in this area illustrates the challenge.. A group of authors, including comedienne Sarah Silverman, sued OpenAI in 2023 over the use of their books without compensation.. The judge later dismissed parts of the lawsuit because it did not point to specific outputs that were direct copies.. The implication is clear: merely saying that a large language model was trained on someone’s material is not enough.. Plaintiffs must connect the alleged scraping to actual outputs that take business away from them.
That “outputs problem” becomes even more complicated because much scraping occurs through bots operating quickly, quietly, and at scale.. Meanwhile, the outputs of major public-facing AI services are visible to everyone, from chat-style assistants to search and summarization products.. But outside that spotlight, a shadow industry of mass scraping has been growing.
It has been an open secret that some AI firms obtain data through third-party brokers.. Media industry analyst Matthew Scott Goldstein published an extensive report on the business ecosystem around that practice. and the conclusions reported in the coverage paint a striking picture: at least 21 companies. some backed with funding totaling hundreds of millions of dollars. reportedly scrape publisher content without paying and then sell “data services” to customers that include OpenAI and Amazon. as well as other publishers such as The Telegraph.
The report also highlights what scale changes.. When scraping is allowed and operationalized at industrial speed, it can produce products rather than just training material.. Companies can build businesses around parsing internet data for bots and agents, indexing content, and selling access as a service.. These vendors are not necessarily household names, but they include firms such as Parallel AI, Exa, and Bright Data.
Goldstein’s reporting emphasizes that these companies. at least in how they present themselves and in how they operate. do not appear to be hiding their model.. A recent Wall Street Journal profile described Parallel AI as a platform dedicated to servicing AI agents. while Goldstein characterizes it more bluntly as a scraper company with better branding.. Either way. the dispute is shifting: it is becoming less about whether data was obtained. and more about how those outputs are commercialized.
The legal and technical landscape also affects incentives.. With setbacks in copyright cases and the current administration’s dismissal of copyright concerns referenced in the report coverage. the message many market actors appear to take away is that unauthorized scraping can come with few real consequences.. In that environment. legal claims and technical controls often fail to keep up. and the default operating pattern favors giving AI systems greater access.
For media companies, this reality creates a difficult strategic dilemma: block bots aggressively, or allow access and build around it.. Blocking, if effective, protects intellectual property but requires constant work to prevent new scraping methods and new actors.. Allowing scraping can mean conceding part of the fight—or outsourcing it indirectly to others—while also potentially preventing media firms from being stuck in a never-ending whack-a-mole.
At the same time. allowing bots and scraping may help publishers stop treating AI as purely a threat and start treating it as a distribution channel.. In this framing. AI engines serve simultaneously as intermediaries and as audience-facing systems that shape what people see and how information is summarized.. For publishers. the underlying question becomes whether the industry can capture value from that role rather than trying to eliminate it entirely.
A considered approach to the scraping ecosystem. as described. rests on five components—though not all may be feasible for every publisher.. The first is improving bot blocking through both technical and legal measures.. Major publishers are said to block bots at least in principle. but taking action beyond standard robots exclusion rules matters because these instructions are often ignored.
A key operational detail raised is that some companies may need highly sophisticated defenses.. People Inc.. CEO Neil Vogel. in comments referenced in the report coverage. has indicated the company had to become more advanced at blocking unauthorized bots.. For publishers without equivalent resources. technical partners and infrastructure providers can help. and the coverage notes that companies such as Cloudflare have moved toward copyright-protecting defaults.
Even with limited blocking capacity, publishers can still gather intelligence.. Rather than only watching bot traffic. the guidance is to regularly audit AI systems to identify where content has been appropriated and misused.. That matters because it can reveal not just access attempts but the form those attempts take once outputs are generated.
The second component is practicing “good GEO. ” a counterintuitive step because it treats access as something to manage rather than only prevent.. Regardless of whether scraping is happening. the argument is that publishers should make content as friendly to AI scrapers as possible.. If content is difficult for bots to interpret. the same difficulty affects both authorized and unauthorized systems. which can create visibility problems even for legitimate use cases.
The incentives here are partly commercial.. Scraping is already occurring. the coverage notes. so publishers may need to compete in summaries—even if they do not want to be present without compensation—so they can gain visibility and the qualified traffic that follows.. There is also a legal angle: a proactive approach can create a paper trail that supports auditing. and it may help prove publisher value if disputes escalate.
Good GEO is also described as important for future internal capabilities. including building an in-house agent or an MCP server for content.. In that sense. the “inputs and outputs” relationship becomes operational: how a publisher structures content can affect how internal and external AI systems work with it.
The third component shifts business strategy rather than relying only on enforcement.. The coverage argues that the media model built around monetizing anonymous traffic is diminishing. referring to changes associated with the Google era.. For publishers. that means diversifying revenue streams becomes necessary—through subscriptions. events. data-related offerings. and other non-ad-centered sources—rather than depending primarily on ad-driven discovery.
The fourth component is litigation, but it is framed with realism.. “Sue” is described as a path that is not available to everyone. since most media companies do not have the resources to take on companies such as OpenAI or Perplexity in court.. Still. the report’s emphasis on an industrial-scale shadow market suggests more targeted legal action could emerge. especially as the financial stakes become clearer.
The final component is regulation, particularly at the state level.. Federal action is described as unlikely in the current environment. but many states are attempting to regulate AI. including by requiring training-data transparency and disclosure rules.. The coverage suggests regulation may not need to overhaul copyright law wholesale; requiring bots to properly identify themselves could improve governance by preventing some bots from convincingly impersonating humans.
While these steps may sound practical, the broader tension underneath them is emotional as well as strategic.. As AI bots increasingly “eat the internet,” publishers may feel that scraping is inevitable, and helplessness can creep into decision-making.. The argument in the coverage is that inevitability should not become an excuse for paralysis. especially in a world dominated by agents.
Instead. publishers are urged to reassert agency: protect what they can. adapt where necessary. and avoid letting the future of content be decided entirely by the same companies that scrape it.. The core point is that the contest over AI scraping is not only about copyright—it is about who gets to define the value chain from publisher content to AI outputs to business outcomes.
AI scraping copyright lawsuits media industry bot blocking data brokers generative AI output harm