An NPR podcast episode published on July 14, 2026 attached specific figures to the zero-click web: one independent site in Berlin has lost roughly three quarters of its traffic since Google introduced AI summaries, and its operator says income remains down about 30% even after renegotiating affiliate rates.

The episode, produced for The Indicator from Planet Money, opens with a trivia question about the actor Paul Rudd and a search that never leaves the results page. That small demonstration frames a larger argument the programme then develops across twelve minutes: that the economic exchange underpinning much of the open web, in which search engines took content and returned visitors, has been quietly dismantled.

Presenters Darian Woods and Adrian Ma describe the phenomenon as the zeroclick internet. According to the episode, 68% of all United States Google searches now end without a click, compared with 45% a decade ago. The programme includes a disclosure that Google is one of NPR's financial supporters.

A figure that shifts with the measurement

That 68% number sits awkwardly alongside other tracking published this year, and the gap is worth examining rather than glossing over.

Similarweb research covered previously put zero-click Google searches at close to 69%, up from 56% before AI Overviews arrived in May 2024. Yet the Datos and SparkToro Q1 2026 State of Search report recorded United States zero-click searches at 22.4% in March 2026, down from 24.5% in December 2025, and 19.6% across the European Union and the United Kingdom. Same phenomenon, radically different scale.

The divergence is definitional and methodological. Datos analyses desktop clickstream behaviour across a panel of tens of millions of users and counts searches that produce no click of any kind. Other approaches classify a click that lands on another Google-owned surface as functionally zero-click, since no third-party site receives a visit. Panels differ, devices differ, and mobile behaviour is not captured identically across studies. For anyone budgeting against these numbers, the underlying definition matters more than the headline percentage.

What the various datasets do agree on is direction of travel for third-party publishers. A randomised field experiment involving 1,065 desktop Chrome users produced the first causal evidence that AI Overviews cut outbound organic clicks by 39.8% while lifting zero-click searches by 34.5%, with no measurable improvement in how users rated the search experience. Ahrefs research from February 2026 correlated AI Overviews with a 58% reduction in click-through rates for top-ranking pages, nearly double the 34.5% figure the same firm published in April 2025.

Amanda Natividad, who works in marketing at SparkToro and co-authors the forthcoming book Zero Click Marketing, told the programme the trend predates generative AI. "Platforms discovered they could capture more value by owning the answers, right? By owning the audience and the niche and giving less and less for the rest of us," she said. Her framing places AI summaries as an accelerant rather than an origin point, consistent with a decade of social platforms suppressing outbound links.

Asked directly whether AI is killing web traffic, Natividad's answer in the episode was affirmative. She also conceded the obvious counterpoint: for many users, getting an answer without a click is simply a better experience.

One site, three quarters of its audience

The episode's most concrete material concerns All About Berlin, a guide for newcomers to the German capital covering everything from immigration bureaucracy to where supermarkets shelve eggs. Its operator, identified in the episode as Nicolas, moved from Canada to Germany about a decade ago and built the site into full-time work funded by affiliate commissions.

Affiliate marketing pays a site a percentage when a visitor follows a link and buys something. It requires traffic. Since Google introduced AI summaries in 2024, traffic to All About Berlin has fallen roughly 75%, according to the episode.

"Every month is the worst month of the website since two years," he said, comparing the experience of checking his analytics to looking at a bank balance while unemployed.

He responded by renegotiating his affiliate commission rates to take a larger cut per conversion. It helped. Income is still down about 30%.

That pattern, partial mitigation that fails to close the gap, recurs across the sector. Teads reported pageview declines of 10 to 15% across its premium publisher network during Q3 2025 and noted that publishers seeking higher revenue per thousand impressions could not price their way out of volume losses. Chartbeat data showed small publishers lost 60% of search referral traffic over two years, medium publishers 47%, and large publishers 22%, with ChatGPT referrals accounting for roughly 0.02% of total publisher traffic.

Some sites do not mitigate at all. Overfishing.org shut down after 21 years when AI Overviews absorbed the search traffic that sustained it. Bauer Xcel Media Germany closed in April 2026 with 160 jobs eliminated, its parent company citing AI Overviews reducing clickthroughs, advertising pressure, and fragmented affiliate and commerce models.

Nicolas framed the aggregate risk in terms of invisibility. "We are slowly going to lose a lot of websites, but we won't even notice until they're gone," he said.

Time builds a second website for machines

Mark Howard, chief operating officer at Time, described the magazine's response as a set of parallel bets. Time blocks many AI crawlers. It has also signed a partnership with OpenAI under which the magazine shares content in exchange for access to the company's technology.

The third measure is structural. "Humans continue to come to the site the way that they always have and the way that you experience Time.com if you go there today. But the AI bots go to what we call a markdown page," Howard said.

A markdown page strips a webpage down to text and metadata. "So we've taken all the visual design cues, all the HTML off the page and all it is is the actual content itself with the metadata that supports the content on the page. And we route the AI bots directly to those pages. It requires less computational um power by them. We're making it easier for them to actually get to that content," Howard said.

The commercial logic behind that generosity is that bots which retrieve content cheaply and cleanly are more likely to cite it. Time's markdown conversion has been documented alongside its use of the TollBit marketplace, which claims a markdown fetch completes in about a quarter of a second against more than a minute for a full HTML page.

Asked whether the approach is working, Howard said it is early days.

Whether stripped-down machine-readable files reliably attract AI attention remains unsettled. Research on llms.txt, a related proposal for publishing machine-readable summaries, found that adoption rose 8.8 times while 97% of the files received zero AI requests. Publishing for machines is not the same as being read by them.

Howard also characterised the pace of change as unprecedented in his experience. "Not only is this new, but it is rapidly changing at a pace that I don't think we've seen before," he said.

Courts, crawlers and the construction of a market

The episode places litigation and infrastructure alongside each other as the two organised responses to the zero-click shift.

On the legal side, publishers including The New York Times are suing AI companies over model training on their articles without permission. That case has generated its own procedural history, including an order requiring OpenAI to hand over 20 million anonymised ChatGPT conversation logs in November 2025. Other actions have gone differently: Ziff Davis filed against OpenAI in April 2025 over content from more than 45 properties, while Anthropic agreed a 1.5 billion dollar settlement in September 2025 in a case brought by authors.

On the infrastructure side, Howard pointed to companies including Cloudflare building what amounts to a paywall for bots. That work is further along than the episode's framing suggests. Cloudflare launched Pay Per Crawl in private beta in July 2025, using HTTP 402 Payment Required responses to let publishers allow, charge for, or block individual crawlers. A year later the company replaced per-crawl charging with a model that pays publishers when their content contributes to a generated answer, citing internal data showing more than half of legitimate crawl traffic re-fetches pages that have not changed.

The measured imbalance is stark. Cloudflare figures published on July 1, 2026 showed crawl-to-referral ratios spanning 118 crawls per referral at the low end to nearly 50,000 at the high end. By early June 2026, bots accounted for 57.4% of all traffic to HTML content across Cloudflare's network.

Blocking, meanwhile, carries costs of its own. Research from Rutgers Business School and The Wharton School found that news publishers who blocked AI crawlers via robots.txt lost 23.1% of monthly visits without achieving proportional protection, because robots.txt remains a voluntary protocol. Legislators have started responding: New York passed a bill in June 2026 requiring AI crawlers to identify themselves to news sites.

Howard expects the picture to look different within a year or two. "There's no doubt that when we have this conversation in 12 or 24 months, it'll be a totally different world in terms of how publishers have navigated this transition," he said. The presenters immediately raised the counterpoint: whether the sites currently watching their traffic disappear will still exist to see it.

Why it matters for marketers

Three implications follow for advertising and marketing professionals.

First, inventory supply is contracting unevenly. The open web that carries programmatic display is funded by the same referral traffic now being absorbed at the results page. Index Exchange analysis of 1,200 publishers found 69% experienced year-over-year ad opportunity declines in 2025, with news down 7% while health and careers verticals fell 40 to 50%. Media buyers planning against category-level reach assumptions built before 2024 are planning against a supply curve that no longer exists in the same shape.

Second, affiliate and performance content faces a specific structural problem. Review sites, comparison guides and how-to content are precisely the formats an AI summary can compress into three sentences. All About Berlin is a small example of a large category. The publisher mix beneath those programmes is consolidating toward larger operators holding licensing deals, a pattern already visible in Google's arrangements with major titles while independent publishers absorb the same declines without compensation.

Third, measurement discipline has become a competitive variable. Zero-click figures ranging from 22.4% to 69% describe the same web. Similarweb research has attempted to trace AI recommendations through to actual site visits, finding that conventional analytics undercount the influence of AI-mediated discovery. Attribution models that treat direct traffic as unattributable will misread an increasing share of demand.

The episode closes with the presenters asking an AI summary for the solution to the zero-click web. It returns advice about abandoning raw traffic goals in favour of brand authority. The hosts note the obvious conflict of interest in that answer. The exchange is played for laughs, but it describes the actual pivot much of the search marketing industry has spent eighteen months attempting, and the Datos data suggests AI tools remain below 2% of desktop visits even as clicks to the open web keep falling.

Fact checking for the episode was credited to Sierra Juarez. It was produced by Corey Bridges, engineered by Jimmy Keeley, and edited by Kagen Canon.

Timeline

Summary

Who: NPR's The Indicator from Planet Money, presented by Darian Woods and Adrian Ma, featuring Amanda Natividad of SparkToro, Nicolas of the independent site All About Berlin, and Mark Howard, chief operating officer of Time. The episode discloses that Google is one of NPR's financial supporters.

What: An examination of the zero-click web, reporting that 68% of United States Google searches now end without a click compared with 45% a decade ago, that All About Berlin has lost roughly 75% of its traffic and about 30% of its income since AI summaries arrived, and that Time now serves approved AI bots a stripped-down markdown version of its site while blocking others and licensing content to OpenAI.

When: The episode was published on July 14, 2026, running twelve minutes and three seconds.

Where: The programme is produced in the United States and distributed by NPR. Its case studies span a Berlin-based independent publisher operating in Germany and a United States magazine, with infrastructure and legal developments drawn from Cloudflare and the federal courts.

Why: Search referrals historically funded advertising, subscriptions and affiliate commissions for independent websites. AI summaries answer queries at the results page, breaking that exchange. The episode documents what the loss looks like at the level of a single site, what a large publisher is building in response, and why the outcome is unresolved: litigation, bot paywalls and machine-readable content formats are all in progress, but the sites losing traffic now may not last long enough to benefit from them.