Sundar Pichai vacuums up British journalism to feed his AI summarizer and dares the publishers to opt out.
It is true, and the Competition and Markets Authority’s announcement is at pains to note it, that for the first time a major regulator has forced Google to offer publishers a separate control over whether their content appears in AI-generated search summaries, rather than burying the AI opt-out inside the all-or-nothing choice of withdrawing from Google Search entirely. The News Media Association called it a “significant step”; Tom Lewis, the former CMA director now representing news publishers, called it a “big win.” The test, the CMA says, begins today on a subset of UK media sites, with Google promising a global rollout to follow. The trouble is that the control being offered is a valve on a pipe the company still owns, and the pipe is feeding a tank that Google has filled with nearly every drop of search traffic in the country. What the CMA has secured is not a structural remedy but a procedural one, and the distinction between the two has been the difference, in every platform-monopoly case of the last two decades, between a constraint that bites and a concession that decorates.
The mechanism the opt-out addresses is the one Cory Doctorow named enshittification: a platform uses its lock on users to extract value from the businesses that depend on it, then turns that value against them. Google’s search monopoly — the CMA puts its share of general search in the United Kingdom at more than 90 per cent — means that a publisher who wants to be found online has no choice but to be indexed, crawled, and, as of the last two years, scraped for the text that populates the AI summaries appearing at the top of the results page. Before yesterday, the only way to block that scraping was to block all of Google, which for a news organisation dependent on search traffic is the equivalent of going off the grid. The opt-out lets a publisher pull its content from the AI summaries while staying in the traditional search results, which is, in the narrow sense that regulators usually mean, an improvement. The trouble is that the underlying power relationship has not been touched. Google still controls the index, the ranking algorithm, and the user’s first glance. The publisher who checks the opt-out box is not trading from a position of parity; it is asking a monopolist to please not extract one particular category of value, while continuing to deliver every other category the platform demands.
It is worth being precise about what the AI summary actually does, because the public discourse has the habit of treating it as a new technology, a fresh feature, a thing separate from the search monopoly on which it sits. It is not. The summary is a parasite on the same crawl infrastructure Google built to index the open web. It reads the publisher’s article, compresses it into a paragraph that appears above the organic links, and answers the user’s question without the user ever needing to visit the page where the answer was found. The drop in click-through traffic that news organisations have been documenting since the summaries appeared is not a side effect. It is the revenue-side expression of a technical architecture in which the platform that controls discovery also controls consumption, and the consumption happens inside the platform’s own frame, on the platform’s own terms, with the publisher’s work serving as the raw material while the publisher’s ad inventory goes unfilled. The same monopoly that let Google charge advertisers nearly five hundred pounds for every UK household, by the CMA’s own estimate, is the monopoly that now lets it convert newsroom output into summary output at zero marginal cost, and the opt-out is a permission slip the company has graciously agreed to issue.
The comparison that has to be sat with carefully, because it clarifies the limits of what yesterday’s announcement will actually change, is the remedies ruling Judge Amit Mehta issued in United States v. Google last September. The Department of Justice had asked for a divestiture of Chrome and a ban on the default-search payments that lock Google into the search box on every iPhone; Mehta denied those structural remedies and instead imposed behavioral ones — annual renegotiation of the default contracts, limited data-sharing with “qualified competitors,” a six-year supervisory framework. The behavioral remedy left the monopoly intact and gave the monopolist a schedule for complying with its appearance. The CMA’s AI opt-out is the same shape at a smaller scale: an adjustment to the terms of extraction that does not alter the fact of extraction. A publisher who opts out of AI summaries is still indexed, ranked, and dependent on Google’s search traffic; the platform still harvests the clickstream, the behavioral data, and the structural position that lets it determine what the British public reads. The opt-out addresses the symptom — the summary box — while leaving the disease untreated.
Lina Khan’s tenure at the Federal Trade Commission established, through a string of merger challenges and complaints, that platform power is architectural power, and that architectural power cannot be cured by consent decrees. The CMA’s move follows the same logic in reverse: it attempts to install fairness through a dashboard toggle rather than through the structural separation of the index from the summarizer, the host from the parasite. The regulator has designated Google with strategic market status and imposed a conduct requirement; the regulator has not required Google to license the crawled content on fair, reasonable, and non-discriminatory terms, nor to divest the AI-summary product from the search index, nor to open the ranking signal so that publishers can audit the mechanics that determine whether their work reaches an audience or disappears into the summary’s one-paragraph maw. The opt-out is what happens when an agency with real teeth decides to use them to rearrange the furniture.
There is a consultation underway, and the CMA says further announcements on Google’s search business are coming in the weeks ahead. The timeline matters because the only way a behavioural remedy becomes a structural one is if the next round moves past the toggle and into the architecture. The opt-out will be tested on a subset of UK media sites, then rolled out globally; Google’s blog post this morning called it a tool for “website owners” to “manage how their links and content appear,” language that casts the company as a benevolent platform operator providing a dashboard. The publisher who clicks the box will still be paying the same search-optimisation tax, still gambling on the same ranking algorithm, still one summarizer-update away from finding its content repurposed under a different label. The extraction curve has been smoothed, not flattened.
The bar mill at Selkirk had a safety shutoff on the bar-mill drive; the operator could hit a button to stop the rollers if a piece of steel fouled. The button did not change the speed at which the mill ran, the hours the men worked, or the profit the owners extracted. It made the existing operation survivable while the operation remained what it was. The CMA’s AI opt-out is the button. The publishers who reach for it will find themselves still standing in the same mill, still feeding the same machine, still hoping that the schedule leaves enough in the pay packet for the next edition. The machine does not know the button is there, and the men who own the machine are already routing the output to the next process, which has its own summarizer, its own index, and its own opt-out, coming soon.