CMA forces Google: publishers can dodge AI Overviews

CMA forces – A UK regulator has ordered Google to let publishers opt out of AI Overviews, forcing a split between what search bots index and what answer engines summarize. The ruling lands as media companies face a traffic-to-licensing shift, where bots increasingly decide

For years, media companies built their strategy around one idea: get noticed by search, earn the clicks, and keep the audience coming back. But now the “audience” is starting to look less like readers and more like bots.

When generative AI was new. the question of what it should mean under copyright law was often treated like a misunderstanding still waiting to be resolved. In the background. another assumption went unchallenged for about two decades: that the mass indexing used by search engines—Google being the obvious example—was simply part of how the internet works. Google. after all. wasn’t replicating full articles; it was pointing people with a headline and a link. functioning more like a distributor than a republisher.

Answer engines flip the mechanics. They ingest content, summarize it, and combine it with other information to build a response. That approach can make it unnecessary for users to interact with the original publisher—convenient for the user, and far less so for the publication.

The evidence of value tends to show up inside the answers themselves: a citation in the form of a named source with a link. That’s part of what fuels the current turmoil. where the issue has already produced “several lawsuits and existential panic in the media industry.” Yet a growing view is taking hold: what AI does to content has more in common with syndication than distribution.

That brings the change with real leverage. The Competition and Markets Authority (CMA). a UK regulatory board. has ordered that Google must provide publishers a way to opt out of AI Overviews—the summaries that appear at the top of search results. Up to now. Google used the same bot for search indexing and AI crawling. meaning opting out of one meant opting out of both.

The CMA’s order creates a new choice: publishers will be able to decide whether to show up in AI Overviews. It also bars Google from punishing the search rankings of sites that choose AI invisibility.

Google, in turn, appears to be complying. The company turned the ruling into its own message to publishers through a blog post promising “new opportunities with generative AI in search.”

This is one decision by one regulator in one country. and it doesn’t instantly rewrite the global AI licensing landscape. But it does offer a practical off-ramp—and the timing matters—because the same pressure making publishers fight for control is also reshaping who gets to be “seen.” AI systems are rapidly becoming publishers’ primary audience.

Traffic was the old war. Clicks were the trophy. The traditional playbook was clear: publish, optimize for SEO, and measure success in how many people clicked.

But the numbers paint a stark picture. Data shows that people who click through from AI search are more engaged—yet the share of total traffic they represent is small compared with what came before. TollBit has logged scrape-to-referral ratios of 179:1 for OpenAI, 369:1 for Perplexity, and 8,692:1 for Anthropic. Digital Trends counted 4.1 million bot scrapes against 4,200 human referrals in a single week.

And the human portion is shrinking. Cloudflare CEO Matthew Prince said bot traffic has passed human traffic for the first time: 57.4% of requests versus 42.6%. He said the crossover came 18 months ahead of his own forecast, with agentic traffic growing eight times faster than human activity.

All those bots need information—both to answer human queries and as context for agents. That changes the game from “How do I get people to click?” to “How do I get bots to pay for my content.”

A high-quality archive stops being bait for traffic and becomes a supply of data. The user shifts from reader to a machine working on that user’s behalf.

Who the “user” is and what AI they’re using are not small details; they shape the value exchange. The source describes it through a simple idea: someone crossing a desert will value a glass of water differently than someone hiking near a mountain stream. The same logic applies to content. A general chatbot may value a specialized information archive lightly—though never at zero. But a specialized service that also serves the publisher’s exact audience would likely value it more.

In an ideal world, publishers would build strict levers to control access. The prescription laid out here is practical: optimize where visibility helps by making content easy for bots to read. parse. and cite via GEO—and learn the AI funnel the way the industry once learned the search funnel. Where the value is high, the argument goes, publishers should offer bots a paywall and block unauthorized crawlers aggressively.

The scraping data itself becomes evidence. It can demonstrate how valuable the content is, and it is not just theoretical: in negotiations with OpenAI, Time pointed to TollBit data to secure a licensing deal.

The market, as described, is splitting into two lanes. One is a paid lane, where OpenAI’s licensing deals sit. The other is a litigated lane, involving CNN, NYT, and News Corp against Perplexity. The UK opt-out gives the litigators more leverage. with the possibility that it could push prices higher in the paid lane.

None of this happens automatically. The ruling matters because it changes the question publishers ask. It stops being only how much AI sends traffic and becomes what each AI path to a reader is worth—and what publishers should charge to be included.

The off switch, in that framing, is really a price list.

There is still a catch: timing. The opt-out exists because the politics around AI are hot right now. with regulators. courts. and even the Vatican pushing in the same direction. Pressure like that doesn’t stay forever. If publishers wait for someone else to set the rates, the leverage built over years gets spent on nothing.

For media companies, the shift feels less like a distant policy debate and more like an immediate choice: whether to stay in the old traffic war—or to price the bot paths that now decide which content gets used.

Competition and Markets Authority CMA Google AI Overviews generative AI media licensing copyright law bot traffic syndication search rankings OpenAI Perplexity Time TollBit Cloudflare Matthew Prince GEO paywall scraping-to-referral ratios