Google on May 19 unveiled an AI-driven search box the company says will let users “easily ask questions” of the web and let a Gemini-powered agent do the looking. The new search box, with AI Mode at its center, is the biggest upgrade to Google’s search bar in over 25 years, the company said at its annual I/O developer conference. It lands on a news industry where several major outlets have lost 30 to 40 percent of their Google search traffic in the past year, and on a public that has told pollsters, twice, that it does not want AI for its news.
What Google is selling as a smarter front door, the news industry is reading as the next turn of a screw.
The Search Box That Replaces Links
Google framed the new interface as a generational redesign. “Today at I/O, we shared the next step in our journey to bring together the best of a search engine with the best of AI,” Liz Reid, Google’s vice president of search, wrote in the official I/O post. The company said the new search box “dynamically expands to give you space to describe exactly what you need” and accepts text, images, files, videos, or Chrome tabs as inputs.
The numbers Google pointed to give a sense of the bet’s scale. AI Mode, the conversational search tab the company launched a year ago, has “surpassed one billion monthly users” and “queries more than doubling every quarter since launch.” Google is also “expanding Personal Intelligence in AI Mode to more people in nearly 200 countries and territories across 98 languages, no subscription required.” Search agents are the second piece of the announcement, and the company said they will operate around the clock, scanning the web, news sites, and social posts for material relevant to a user query. Information agents will roll out first to paying subscribers this summer, with agentic booking and shopping reaching everyone in the U.S. later this summer.
For users, the pitch is convenience: ask a longer question, get a synthesized answer, and continue the conversation. For the sites that built the open web, the pitch is the same one Google has made since AI Overviews first appeared at the top of results in May 2024. The new search box buries the link further down the page. And it gives Google’s AI the front row, with its link index sitting behind.
Publishers Are Already Bleeding
That further demotion is the part publishers have been sounding the alarm about. Helen Havlak, publisher of The Verge, was blunt about it in an interview with NPR.
The extinction-level event is already here. And a bunch of small publishers have already gone out of business.
Havlak told NPR in July 2025.
The damage is showing up in the data. Similarweb figures published by NPR in July 2025 showed traffic to CNN’s website had dropped about 30% year over year, and to Business Insider’s and HuffPost’s sites by around 40% in the same period. A Pew Research analysis of the browsing behavior of 900 U.S. adults in March 2025 traced the mechanism: users who encountered an AI summary clicked on a traditional search result link in 8% of all visits, compared with 15% of visits when no summary was present, and the summary’s own cited links got a click in 1% of all visits. After seeing an AI summary, 26% of users ended their browsing session altogether, versus 16% of users on pages without a summary. The Pew study also found that 18% of all Google searches in its sample produced an AI summary, and 60% of question-form queries did.
| Search page type | Clicked a result link | Ended the session |
|---|---|---|
| With an AI summary | 8% of visits | 26% of pages |
| Without an AI summary | 15% of visits | 16% of pages |
Danielle Coffey, who heads News/Media Alliance, called Google’s behavior “parasitic” and “unsustainable” in a statement to NPR, arguing that the opt-out from AI Overviews is effectively an opt-out from Google search.
The Audience Has Made Its Preference Clear
The audience, surveys show, was never asking for any of this. A Pew Research Center survey of 5,153 U.S. adults conducted in August 2025 found that 75% of Americans say they never get news from AI chatbots like ChatGPT or Gemini. About one in ten say they get news ‘often’ (2%) or ‘sometimes’ (7%) from chatbots, with another 16% doing so ‘rarely.’ Fewer than 1% of Americans prefer to get news from chatbots rather than from other sources.
For those who do use chatbots for news, the experience is mixed. A third of chatbot news users told Pew they find it difficult to determine what is true and what is not. About half said they at least sometimes come across news there that they think is inaccurate, and 16% said they saw this ‘extremely often or often.’
The age split is sharper on the accuracy question. Among adults 18 to 29 who use chatbots for news, 59% said they at least sometimes saw news they thought was inaccurate. For adults 30 to 49, the figure was 51%, and among 50 to 64 it dropped to 43%. For those 65 and older, it was 36%.
- 5,153 U.S. adults surveyed in Pew’s August 2025 study of chatbot news habits.
- 88% of AI summaries in Pew’s March 2025 study cited three or more sources.
- 67 words: the median length of an AI summary in that same study.
- 36% of full-sentence searches in the study generated an AI summary.
What Happens When the Bot Gets It Wrong
Trusted news outlets have corrections policies, named editors, and a public masthead. AI summaries do not. A study by Forum AI, a research firm that audits AI outputs, found that election-related answers from leading chatbots failed the firm’s accuracy, neutrality, or source-quality checks 90% of the time, with nearly 36% of election answers containing fact errors.
The risk of bot-generated errors in news is not hypothetical. In May 2025, the Chicago Sun-Times published a ‘Summer Reading List for 2025’ that recommended 15 book titles. The paper itself later confirmed that 10 of the 15 were fake. The list was generated by a third-party AI vendor and ran as a paid special section. Other major newspapers, including The Philadelphia Inquirer, ran the same section.
Chatbots cannot report. That’s something journalists can do and robots cannot.
Klaudia Jaźwińska, a researcher at Columbia University who studies how AI is upending the news industry, told NPR in July 2025.
A Resistance Movement Is Already Built
The audience is not the only group that has pushed back. Across the country, communities are organizing against the physical infrastructure of the AI buildout: the data centers that supply the model and search compute.
The numbers carry weight. In 2025, projects with publicly disclosed values totaling at least $156 billion were blocked or delayed by coordinated local opposition, moratoria, and litigation, according to a Q3-Q4 2025 update from Data Center Watch. At least 48 data center projects were affected, in 42 U.S. states. Opposition has hardened into law, elections, and policy, influencing races in Virginia, New Jersey, and Georgia. The trend is setting the stage for the 2026 midterms.
The opposition spans party lines. State legislatures, utility regulators, and ballot initiatives have all become battlegrounds, and moratoria in multiple states are complicating timelines for new builds. The pattern matters for the AI search story, because the same data centers feeding Google’s search agents are the ones the public is fighting. Beyond the direct fights, the AI buildout is reshaping the global chip market, and a 2026 memory shortage tied to data-center demand is making laptops and phones cost more for ordinary buyers.
What Publishers and Readers Can Do
The publisher side has started to organize. Cloudflare, the internet infrastructure firm, launched a ‘Pay Per Crawl’ product in July 2025 that lets website owners charge AI crawlers a per-request fee to scan their content.
News/Media Alliance, the trade group that represents more than 2,000 outlets, is pushing for compensation deals. Some publishers, including The Verge, are rebuilding as platforms in miniature, with newsletters, podcasts, and short-form feeds designed to keep readers on the site. Major outlets are also litigating: a dozen copyright suits against AI firms are now in court, including The New York Times’ federal suit against OpenAI. Cloudflare chief executive Matthew Prince has framed the new model in similar terms, telling NPR that if content creators cannot get compensated for their content, they will stop creating it.
Readers have their own levers. They can visit trusted outlets directly, build a small list of bookmarks, and subscribe to newsletters filtered into the primary inbox. They can route DNS through a privacy resolver that does not phone home to Google. A $20 Raspberry Pi keeps DNS queries off Google’s servers and runs Pi-hole and Unbound, a small hardware project that is well-documented for beginners. They can pay for the journalism they want to keep, and read past the AI summary to the source, every time.
The audience has been polling consistent on this question. About half of chatbot news users say they at least sometimes see news they think is inaccurate. Forum AI found chatbot election answers fail on accuracy checks 90% of the time. The public, on its own surveys, keeps saying the same thing.
A search box that answers with AI is not the same thing as a public that wants AI to answer. The result is a new kind of friction between the largest search company on the web and the journalists whose reporting fills it.
- Visit news sites directly, instead of starting every question in a search engine.
- Bookmark and subscribe to a small list of trusted outlets; filter their newsletters into the primary inbox.
- Route DNS through a privacy resolver that does not log queries to Google, using Pi-hole or a comparable setup.
- Pay for the journalism you want to keep, even if the amount is small.
- When an AI summary is the only thing shown, click through to the source, and read the source.
Frequently Asked Questions
What did Google announce at I/O 2026?
Google unveiled what it called the biggest upgrade to its search box in over 25 years: a redesigned, AI-driven “intelligent” search bar that the company said can accept text, images, files, videos, or Chrome tabs as inputs. The company also announced Search agents that operate 24/7 in the background, scanning the web, news sites, and social posts for material relevant to a user query. AI Mode, the conversational search tab, has passed one billion monthly users, according to Google.
How much Google traffic have news publishers lost?
Significant amounts, with the steepest declines in the past year. According to Similarweb data published by NPR, traffic to CNN had dropped about 30% year over year as of mid-2025, and to Business Insider and HuffPost by around 40% in the same period. Pew Research found that users who encountered an AI summary clicked a traditional result link in 8% of visits, compared with 15% when no summary was present.
Do Americans want to get news from AI?
No, by a wide margin. A Pew Research Center survey of 5,153 U.S. adults in August 2025 found that 75% of Americans never get news from AI chatbots, and fewer than 1% prefer chatbots over other news sources. About half of chatbot news users said they at least sometimes see news there that they think is inaccurate.
What is “Google Zero”?
It is the scenario publishers describe when Google stops sending meaningful traffic to news sites altogether. The phrase was coined by tech observers to describe an AI-first search experience in which the chatbot answers the question and the reader never clicks a link. Havlak told NPR the trajectory is already an “extinction-level event” for the industry.
