News · 2026-07-28

Google loses its DMCA claim against a search scraper

Judge Yvonne Gonzalez Rogers of the Northern District of California dismissed both of Google's copyright anti-circumvention claims against the scraping company SerpApi on July 20, holding that Google cannot treat a general anti-bot wall around ordinary search results as a copyright access control. The core theory - covering search-result pages without copyrighted content - was dismissed permanently, without leave to amend.

Key facts

• The case is Google LLC v. SerpApi, LLC, No. 4:25-cv-10826-YGR, decided July 20, 2026 in the Northern District of California.
• Both DMCA Section 1201 claims were dismissed; the claim covering ordinary search results was dismissed without leave to amend.
• Google was given 21 days to amend the surviving theory, with discovery stayed meanwhile.
• Primary source: the court's order; Google's original complaint.

Google's suit centered on SearchGuard, a JavaScript challenge that runs in the browser to distinguish human-looking queries from mass automation. Google alleged SerpApi evaded it and sold structured Google results through a commercial API - an API whose customers include, prominently, AI companies that need search data.

The legal instrument Google reached for was Section 1201 of the Digital Millennium Copyright Act, the provision that makes it unlawful to circumvent a technological measure controlling access to a copyrighted work. It is the rule that stops you from breaking a DVD's encryption. Google's theory was that SearchGuard is such a measure and that its search-result pages are the protected work.

The court's reasoning turns on a distinction that is easy to state and consequential in practice: the anti-circumvention rule protects copyrighted works, not walls. If the gate does not guard something copyrightable, breaking the gate is not a copyright violation - whatever else it might be.

Applied to search results, that split the case in two. For ordinary result pages, Google's own pleading was fatal. Google described those pages as compilations of publicly available information, which are only sometimes accompanied by a Knowledge Panel containing licensed material. A gate that controls access to non-copyrightable compilations cannot be a copyright access control. Dismissed, without leave to amend - meaning that theory is finished, not merely deficiently pleaded.

For pages that do contain a licensed copyrighted component, the court found a different defect. Google had not pleaded that the owners of those licensed works authorized SearchGuard to act as their access control. This one was dismissed with leave to amend, so it was not decided on the merits, and Google had 21 days to try again with discovery stayed in the meantime.

SerpApi did not sweep the hearing either, and the losses are worth noting because they narrow how far the ruling travels. The judge rejected its argument that Google lacked standing to bring a DMCA claim at all, rejected its argument that circumvention was inadequately alleged, and rejected its argument that a measure cannot protect copyrighted material merely because it also blocks non-copyrighted material on the same page. SerpApi CEO Julien Khaleghy called the order a win for access to public pages and said the company would continue serving developers, researchers and AI companies - which is the defendant's advocacy, not the holding.

The one-line takeaway that survives scrutiny: Google cannot convert a general anti-bot wall around non-copyrightable search results into copyright protection. That is not a ruling that scraping is lawful. Google's Search spam policy still prohibits automated querying, and contract, trespass and computer-fraud theories are entirely untouched.

The knock-on effect lands on a bigger case. Reddit has a separate suit in the Southern District of New York against SerpApi, Oxylabs, AWMProxy and Perplexity, alleging that because they could not scrape Reddit directly, they evaded Google's controls and harvested Reddit content out of Google result pages. The California order does not decide that case, but it creates an obvious obstacle for any claim built on the premise that Google's gate protects the non-copyrightable material on a results page.

Underneath the legal question sits an economic one that surfaced the same day. An anonymous operator posted that AI bots made up roughly four-fifths of their site's traffic, with Meta the largest crawler and no visitors returned, and OpenAI producing a single referral - the widely shared "80,000 to 1" figure.

That number should not be used as an industry rate, and the best correction came from the replies: training crawls cannot generate referral clicks, so dividing all bot requests by search referrals compares unlike things. Cloudflare's own analysis separates AI crawling into training, search, user action and undeclared, and finds training accounts for nearly four-fifths of network-wide AI crawling, with user-action traffic a small minority. OpenAI's publisher documentation makes the same split operational: OAI-SearchBot handles discovery and referral, while GPTBot is the control publishers use to opt out of training. See Cloudflare splits AI crawlers into search, agent and training.

The honest caveat on the anecdote: no site name, no date window, no logs, no definition of whether "traffic" means requests or visits, and no stated bot-verification method. It is a real feeling accurately reported and a bad statistic.

Put together, the day exposes a structural mismatch rather than a rule. The largest search company's attempted legal moat around its public results has a hole in it, and the publisher-side exchange that was supposed to justify crawling often is not an exchange at all - machines consume pages for training while only a much smaller retrieval layer can plausibly send a human back. See also training data deduplication.

Primary source, verified: read the paper →

Ground Truth. (2026, July 28). Google loses its DMCA claim against a search scraper. Ground Truth. https://groundtruth.day/news/google-loses-its-dmca-claim-against-a-search-scraper.html

@misc{groundtruth:google-loses-its-dmca-claim-against-a-search-scraper,
  title  = {Google loses its DMCA claim against a search scraper},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {jul},
  url    = {https://groundtruth.day/news/google-loses-its-dmca-claim-against-a-search-scraper.html}
}

Topics: legal · policy · web-scraping · copyright · google · training-data