AI Is Killing Web Traffic (And It Might Destroy Itself Because of It)

Something ugly is happening to the web and the numbers are now impossible to ignore.

Zero-click searches, where users get an answer from Google or an AI chatbot without ever visiting a website, now account for roughly 65% of all Google queries. For news-related searches that number climbs even higher. Google search traffic to publishers fell 33% globally in just the past year. Some individual sites have lost 70, 80, even 89% of their organic traffic. NPR called it an "extinction-level event" for online publishing.

And yet the AI tools responsible for this are also the ones that depend on those same publishers to survive. If the web hollows out, the training data dries up. The models get dumber. The whole thing collapses inward.

This is not a hypothetical. It's already in motion.

Jump to Section

• How Web Publishing Actually Makes Money
• What Zero-Click Search Is Doing to Publishers
• The Real Numbers Behind the Collapse
• How AI Companies Are Responding
• What Happens to AI When the Web Dies
• What Survives: Tools Over Content
• FAQ

---

How Web Publishing Actually Makes Money {#how-web-publishing-makes-money}

The business model of most content sites on the internet is dead simple: write useful stuff, get people to visit, show them ads, collect revenue. That's it. A million page views at a $10 CPM is $10,000 a month. Scale that up and you have a real business. Scale it down and you're writing for free.

Subscriptions work for a handful of elite outlets. The New York Times pulled it off. A few others. But the vast majority of the open web, the how-to guides, the tech tutorials, the niche blogs, the independent news sites, the developer documentation, all of it, lives or dies on advertising revenue tied directly to page views.

No traffic. No impressions. No money.

The deal the web made with Google for 25 years was simple: you write stuff, we send you readers, everyone wins. That deal is being quietly shredded.

---

What Zero-Click Search Is Doing to Publishers {#zero-click-search-impact}

Google's I/O 2026 announcements essentially confirmed what publishers have been screaming about for two years. Google described the overhaul as "the biggest change to search in 25 years." The new experience is built around AI Mode, conversational follow-ups, and background agents that monitor the web on your behalf. Every one of those features is designed to keep users on Google's results page rather than sending them to a third-party site.

The result? You ask a question, get a full AI-generated answer, and close the tab. The publisher whose content informed that answer gets nothing.

This is the zero-click problem, and it has been accelerating hard. According to data from Similarweb, zero-click searches reached 60% of all queries after Google's I/O 2026 rollout. For AI-augmented results specifically, that rate runs between 60% and 93% depending on the platform. ChatGPT, Perplexity, and Google AI Mode are explicitly designed to answer your question without sending you anywhere.

Google's position on all of this is predictably corporate: AI Overviews actually generate more high-quality clicks because engaged users go deeper. Maybe. But HubSpot lost 70–80% of its organic traffic. Chegg reported a 49% decline. Business Insider's traffic dropped 55% between 2022 and 2025, which led to cutting 21% of its staff. DMG Media, which owns MailOnline, documented drops as steep as 89% for certain query types.

Those aren't "higher quality clicks." Those are bodies.

---

The Real Numbers Behind the Collapse {#real-numbers-collapse}

Let's put some actual data on the table because the scale of this matters.

Google Network advertising revenue, which is what publishers earn through Google's ad systems, fell 4% in Q1 2026, the sharpest quarterly decline in recent reporting history. That's Google's own ecosystem eating itself. Meanwhile, Google search revenue grew 19% in the same period because Google is keeping the ad dollars on its own pages while the traffic that used to flow outward dries up.

Publisher-specific traffic data from Press Gazette covering 400+ news sites shows that Google search traffic share to news publishers collapsed from 51% in 2023 to 27% in Q4 2025. In the US specifically, Google search referrals dropped 38%. Facebook is down 43%. Twitter/X is down 46%.

A study cited by Search Engine Land projects that organic search traffic could decline by 43% by 2029 as AI answers replace informational queries at scale.

Some smaller publishers have already shut down. More will follow in 2026. The ones still standing are consolidating, cutting staff, or throwing up paywalls to survive.

---

How AI Companies Are Responding {#how-ai-companies-responding}

There's a growing acknowledgment from AI companies that they can't just hoover up the web indefinitely without consequences. The response has been a wave of licensing deals, though the terms often favor the AI companies far more than the publishers.

OpenAI has signed deals with the Associated Press, the Guardian, News Corp (reportedly $250 million over five years), Axel Springer, the Financial Times, and others. Google has struck deals with AP and Reddit. Meta, Microsoft, and Amazon have their own licensing agreements in place. These deals typically give AI companies access to publisher archives and real-time content in exchange for attribution, some traffic referral, and cash payments.

The problem is the deal structure is deeply unequal. Major licensing agreements exclusively involve large publishers. News Corp gets $250 million. Reddit gets $60–70 million per year. The thousands of independent publishers and niche sites that collectively make up most of the useful information on the web? Nothing.

And there's a harder structural problem. As Nieman Lab put it bluntly, as long as participating in Google's search index means participating in AI training, the market price for publisher content is effectively zero. If you want to be findable, you have to be scrapeable. Most publishers can't afford to opt out of search entirely just to protect their training data rights.

The lawsuits are piling up too. Dozens of publishers are suing OpenAI, Perplexity, and Google over unauthorized scraping, copyright infringement, and the reproduction of content in AI outputs. The Danish media body DPCMO took OpenAI to court in February 2026. Encyclopedia Britannica and Merriam-Webster joined a lawsuit claiming ChatGPT reproduces their content verbatim. None of these cases are resolved. The legal framework for all of this is still being built in real time.

---

What Happens to AI When the Web Dies {#what-happens-to-ai}

Here's the part that should genuinely worry the AI industry.

These models were trained on the open web. The breadth, diversity, and quality of human-written content that accumulated over 30 years is what made them useful. But training is not a one-time event. Models need fresh data. They need current information. The ones deployed right now depend on web retrieval to stay relevant since their training cutoffs freeze their base knowledge at a fixed point in time.

If the web degrades, both things get worse simultaneously. The retrieval layer returns lower quality results because the sources it's pulling from are lower quality or simply gone. And the next generation of models gets trained on a web that's increasingly dominated by AI-generated slop, recycled summaries, and content farms that optimized for AI citation rather than human usefulness.

This phenomenon, called model collapse, is not theoretical. Researchers have documented that models trained heavily on AI-generated text begin to degrade in measurable ways. The diversity of ideas narrows. Factual reliability drops. Edge cases get handled worse. The rich variation of human perspective and experience that made the original training data valuable simply cannot be replicated by AI generating more AI text.

As one industry observer put it, publishers function as the plankton of the digital ecosystem. Kill the plankton and everything up the food chain eventually starves.

Gartner predicts 80% of AI training data will be synthetic by 2028. The synthetic data market is growing fast precisely because the natural supply is being threatened. But synthetic data supplementing human content is very different from synthetic data replacing it. Every AI researcher who is honest about this knows the difference.

---

What Survives: Tools Over Content {#what-survives}

Not everything is equally screwed by this shift, and that distinction matters.

Content that can be fully summarized by AI suffers the most. How-to guides, explainers, news summaries, product descriptions, any content where the value lives entirely in the information itself is getting absorbed directly into AI answers. If the user can get the gist without clicking, they will.

What survives is harder to absorb. Original reporting with firsthand sources. Community. Interactive tools. Real-time data. Things you actually have to use rather than just read.

That's part of why building developer tools rather than only writing developer content is a more durable strategy. An AI can explain how to format JSON in natural language, but it can't be the JSON formatter you run right now in your browser. It can describe how HTTP headers work, but it can't be the HTTP header inspector that shows you exactly what your server is returning. It can sketch out a system architecture in words, but it can't be the diagram editor where you actually build and export the thing.

The same logic applies to any developer resource with genuine utility. The tokenizer on this site exists to do a job, not just describe how tokenization works. The API directory is a live resource you navigate, not an article about APIs. Those things are harder for AI to replace because they're not information, they're function.

This is also why developer tools as a category tend to hold up better than general publishing. Developers need to run code. They need to test things. They need environments and utilities that execute, not just explain. A site that gives developers tools they actually use in their workflow has something AI can reference but can't fully replicate.

The content that's going to survive this transition is the content that does something, not just says something. Original voices with real authority, interactive utilities, community-driven knowledge, things that require actual human participation to maintain their value.

Everything else is running on borrowed time.

---

FAQ {#faq}

What is a zero-click search? A zero-click search is any Google query where the user gets their answer directly on the results page without clicking through to any website. This includes AI Overviews, featured snippets, knowledge panels, and AI Mode answers. Currently about 65% of all Google searches end this way.

Why are AI chat tools bad for publisher ad revenue? Ad revenue for most publishers depends on page views. When users get answers from AI chat tools or AI search without visiting the source website, those publishers get no traffic, no impressions, and no revenue even though their content may have informed the AI's answer.

How much traffic have publishers lost because of AI search? Numbers vary widely but are consistently severe. HubSpot lost 70–80% of organic traffic. Business Insider lost 55% over three years. Some DMG Media properties reported drops as steep as 89% for certain query types. Across the industry, Google search traffic to news publishers dropped from 51% share in 2023 to 27% by late 2025.

Are AI companies paying publishers for their content? Some are. OpenAI, Google, Microsoft, Meta, and Amazon have signed licensing deals with major publishers like News Corp, AP, Reuters, the Guardian, and the Financial Times. But these deals almost exclusively involve large publishers. Independent and niche sites generally receive nothing despite contributing significantly to training data.

What is model collapse and why does it matter for AI? Model collapse is what happens when AI models are trained heavily on AI-generated content rather than human-written text. The models gradually degrade in quality, losing diversity, factual reliability, and the kind of nuanced understanding that comes from learning on authentic human expression. If the web's supply of quality human content shrinks because publishers can't afford to keep operating, future AI models could end up significantly worse than the ones we have today.

Can publishers block AI crawlers from scraping their content? They can attempt to via robots.txt, but it creates a brutal tradeoff. Blocking AI crawlers often means blocking the search crawlers that drive discoverability. Since Google's AI training and its search indexing are increasingly intertwined, many publishers face a binary choice: be visible and be scraped, or block scrapers and lose your search presence entirely. Most can't afford the latter.

What kinds of websites are most at risk from AI search? Sites whose primary value is informational content that can be summarized are most at risk. How-to guides, news aggregators, product review roundups, explainer content, definition pages, anything where AI can give users the gist without sending them to the source. Sites with interactive tools, original reporting, user-generated community, or real-time data are somewhat more insulated.

What is generative engine optimization (GEO)? GEO is the emerging discipline of optimizing content to be cited and referenced by AI-generated search responses rather than just ranking in traditional blue-link results. The goal is to be included in AI Overviews, AI Mode answers, and chat tool responses. It's becoming a required layer of SEO strategy in 2026 alongside traditional search optimization.

Will AI content licensing deals save publishers? Probably not at the scale needed. The deals announced so far are concentrated among a small number of major publishers and the terms heavily favor the AI companies. Structural issues, like Google's control over both search indexing and AI training, limit publishers' negotiating leverage. Licensing revenue may help large players survive but is unlikely to meaningfully compensate the long tail of the web.

What does this mean for developer tool sites? Developer tool sites that provide actual utilities, not just written content, are in a stronger position than pure publishing sites. Tools that execute real functions in the browser, like formatters, validators, inspectors, tokenizers, and diagram editors, have utility that AI can reference but not replace. That said, traffic from informational content around those tools is still vulnerable to zero-click erosion, so diversifying beyond Google search traffic remains important.