Some publishers fear that Google’s AI-powered search will be a catastrophe

Whenever Google makes a change to its search product, it inevitably generates a lot of anxiety among news outlets, many of which rely on the revenue they get from traffic driven by the platform. But the response to the company’s most recent announcement was arguably a class apart. “It’s the End of Google Search As We Know It,” read a headline in Wired. Danielle Coffey, the CEO of the News Media Alliance, a lobby group that represents newspapers and other publishers, told CNN that the changes would be “catastrophic” for its members; others called the Google news a “death blow.” Nilay Patel, the editor in chief of The Verge, argued that the latest changes will accelerate what he calls “Google Zero,” whereby Google search traffic vanishes completely for some publishers.

The changes in question involve a fundamental reworking of Google’s search product, which will incorporate a variety of features powered by the company’s artificial intelligence engine, which it calls Gemini. (The new features were announced last week at Google’s annual developer conference in Mountain View, California, an event known as I/O.) As Wired explained, the latest announcement is the culmination of a series of moves that Google has made over the past year or so. Last year, the company created a separate section of its Search Labs—which allows users to experiment with new features—for what the company called the Search Generative Experience. The announcement last week essentially incorporates those experimental features into the main search product.

The most obvious change is that, for many searches, AI-generated summaries of the information a user has requested will appear at the top of the search page, above Google’s traditional blue links; when a journalist from Wired asked, for example, where the best place to see the aurora borealis would be, Google’s AI-generated response advised the Arctic Circle as part of a longer passage of text. Liz Reid, the head of Google’s search team, told Wired that AI answers won’t appear for simple searches that could be satisfied by linking to a specific website, but will be used for more complex queries. (If you want the old-fashioned results without the AI summary, you can still get them by clicking on the “Web” heading under the “More” tab at the top of the search page. According to a number of reports, you could also try adding a few characters to your Google search URL.)

While links to external websites do appear below the AI-generated answers, some publishers and groups such as the News Media Alliance are afraid that they will attract precious few clicks, since the AI will have already provided the most obvious answer to the user’s question. Google’s response to these criticisms has been twofold. On the one hand, the company argues that its job is to help users get the information they need as quickly as possible; in many cases, giving them an answer without making them click on a link is the best way to do that. As Reid told Wired, “If you have an opportunity with technology to help people get answers to their questions…why wouldn’t we want to go after that?”

At the same time, Google says that it remains committed to referring traffic to publishers and that it believes users will still click on links even if they have already gotten an answer to their query. In a video interview with Patel, Satya Nadella, the chief executive of Google, said the company recognizes the “symbiotic” value of the ecosystem that it has created with publishers—if there aren’t sites making unique and useful content, Nadella said, “then what are you putting together and organizing?” Echoing a blog post by Reid, Nadella insisted that, according to Google’s own research, many users click on links even when AI-generated summaries are available. “Yes, there are times people come and all they want is a quick answer,” he said. But AI summaries will lead to “growth for high-quality content.”

To say that some publishers are skeptical of this hypothetical future would be an understatement. As Kevin Roose noted in the New York Times, Google executives argue that AI summaries take the legwork out of searching—but this legwork, Roose writes, “pays for a lot of journalism,” which in turn makes the Web more useful by providing high-quality information that can be searched in the first place. Despite Google’s reassurances about clicks on AI-generated results, Gartner, a technology research firm, estimates that Google search traffic overall will fall 25 percent by 2026, while Raptive, which provides services to thousands of websites, estimates that some sites could lose up to two-thirds of their previous Google traffic, at the cost of as much as two billion dollars to site creators.

Google’s push to integrate AI into search and other products isn’t being driven solely by an interest in serving readers—or even a desire to experiment with new technology—but also by a kind of competitive pressure that the company hasn’t experienced in some time. Ever since OpenAI came on the scene with products such as ChatGPT, the existing major platforms have scrambled to either cut licensing deals with OpenAI or develop their own in-house AI engines and products. Microsoft, for example, has invested in OpenAI and has integrated the latter’s technology into its Bing search engine, as well as a new chat-oriented AI search engine that it calls Copilot.

As I wrote recently in a piece for CJR, news publishers have themselves taken divergent paths in response to the AI tsunami. A number of news companies, including Axel Springer and the Associated Press, have signed licensing deals with OpenAI that involve payments for their content—and, in some cases, links to their content in ChatGPT’s results—in exchange for letting ChatGPT train itself on their archives. But other news companies—including the New York Times, The Intercept and eight newspapers owned by Alden Global Capital—have taken a different tack, and are suing OpenAI for using their content without permission to train its AI engine. Some publishers are also working on developing their own “large language models,” or AI engines that are trained exclusively on their own content.

As they have trialed new strategies for dealing with a disruptive new technology, many publishers have continued to rely on the traffic they get from Google search links. But even this older technology has hardly been static over time. As Wired noted, the traditional “ten blue links” that used to be Google’s calling card have been declining in prominence and value for a number of years. First came advertisements, which began breaking up pages of search results; then Google started adding “info boxes” and highlighting specific results in what the company calls the “carousel” at the top of its search results page, as well as linking directly to other Google products. In that sense, AI-generated answers or summaries of links are a continuation of a pattern of evolution, rather than a totally new threat. In his interview with Patel, Nadella noted that some of the same fears about declining traffic greeted previous technological changes, like the arrival of smartphones.

One challenge for Google—and other AI-powered search companies such as Microsoft and OpenAI, which is expected to launch its own search engine at some point—is the propensity for the large language models that power such technology to simply invent results, a problem that some observers refer to as “hallucinating.” (Last year, The Atlantic asked Google to name a country in Africa whose name starts with the letter K; the results, powered by ChatGPT, said that there wasn’t one—which was likely news to the people of Kenya.) AI engines have also invented criminal charges against innocent people, among other significant errors. At least one legal expert has argued that Google could be legally liable for AI-generated search answers in ways that don’t apply to its listing of external links, which are protected from liability under Section 230 of the Communications Decency Act.

Some media companies have argued that the unreliability of AI-generated text reinforces the need for the verifiable facts that news publishers can provide; in some cases, publishers have used this argument as a rationale for licensing their content to OpenAI. In his interview with Patel, Nadella said that rewarding originality is “a complex question”—but that Google is trying to satisfy the expectations of users, who are “voting with their feet” when it comes to which search results they see as most useful. Even for Google, the AI future is accelerating rapidly. But will it be an extinction event for news publishers, or an opportunity?

Other notable stories:

• Yesterday, the Murdoch-owned publisher News Corp became the latest media company to strike a licensing agreement with OpenAI, allowing the latter to use content from News Corp titles in the US, the UK, and Australia to train and supply its AI tools in a deal that could be worth more than two hundred and fifty million dollars to News Corp over five years. In other news about AI and the news, the Washington Post’s Heather Kelly reports that a chatbot run by Meta is regurgitating news outlets’ reporting for users of the company’s platforms—even though Meta has otherwise walked away from promoting news of late. And a German magazine paid compensation to the family of Michael Schumacher—a Formula 1 racing driver who has not been seen since entering a coma following a 2013 skiing accident—after it published an AI-generated “interview” with him.
• Also yesterday, Will Lewis, the new publisher of the Post, convened a meeting with staff to discuss a pivot in strategy for the paper, which he said has lost 50 percent of its audience since 2020 and seventy-seven million dollars in the past year alone. Going forward, the Post intends to add new subscription tiers, including in the hope of adding more corporate subscribers, as well as more thoroughly integrating AI into its newsroom. Meanwhile, Matea Gold, a managing editor at the Post, responded to a report claiming that another editor instructed colleagues not to promote a story involving allegations that Lewis helped cover up phone-hacking when he worked for Murdoch in the UK; Gold described the discussion as “routine” and said the Post is covering the story rigorously.
• Vivek Ramaswamy, the anti-“woke” former Republican presidential candidate, has taken a 7.7 percent stake in BuzzFeed as an activist investor, saying in a regulatory filing that he believes the struggling company is undervalued and that he plans to speak with its management. Business Insider’s Peter Kafka weighed four theories for Ramaswamy’s interest in a company that, on its face, does not remotely share his politics; they run the gamut from sincere financial play to trolling. (A spokesperson for BuzzFeed, which shuttered its news division last year, said that the company “is purposely structured to protect its editorial integrity,” but is “open to hearing ideas from our shareholders.”)
• Jem Bartholomew, a frequent contributor to CJR, is out with a new newsletter, Conversations on Poverty, which will “feature conversations about how the media covers poverty—and discuss the urgent task of doing it better.” For his first interview, Bartholomew talked with Sarah Stillman, of The New Yorker, whose work, he writes, finds “the emotional core at the heart of broken systems.” Often, “we talk about significant socio-political issues as ‘structural,’” Stillman told Bartholomew, “but I’m interested in exploring what that actually means where the rubber meets the road.”
• And Rishi Sunak, Britain’s prime minister, called surprise early elections for July 4 in a speech that was drowned out by torrential rain and a nearby protester playing “Things Can Only Get Better,” a song associated with the opposition Labour Party’s landslide election victory in 1997. (“Things can only get wetter,” one newspaper quipped this morning.) Later, a security guard physically ejected Darren McCaffrey, a journalist with Sky News, from Sunak’s campaign launch as McCaffrey continued to broadcast.

ICYMI: Stanis Bujakera Tshiamala on reporting, imprisonment, and an attempted coup in the DRC

Mathew Ingram is CJR’s chief digital writer. Previously, he was a senior writer with Fortune magazine. He has written about the intersection between media and technology since the earliest days of the commercial internet. His writing has been published in the Washington Post and the Financial Times as well as by Reuters and Bloomberg.