Are Websites Dying?

There has never been more information online, yet many publishers have never felt less certain that readers will arrive to see it. That is the strange shape of this moment. The pages are still there. The servers still answer. What is being dismantled is quieter: the arrangement that rewarded websites for making information available.

To see the change clearly, stop asking whether websites are dying and ask a more useful question: what were they living on?

For a generation, answering a question and visiting a website were usually the same act. Search engines organised the web; publishers supplied the material; readers clicked outward. AI search separates those acts, and changes who captures the value created between them.

How the old web actually worked

The open web ran on a bargain so familiar that it became almost invisible. A publisher allowed search engines to crawl and index a page. In return, the search engine sent people to it. The publisher converted those visits into advertising, subscriptions, sales, donations, reputation or a direct audience, and used that value to produce the next page.

Crawl, index, rank, click. That loop became a load-bearing structure beneath independent publications, blogs, forums, review sites, recipe pages and countless specialist resources.

An entire discipline, search engine optimisation, grew around the loop. Ranking mattered because ranking produced visits. Underneath it all sat one quiet assumption: the search engine would point the user towards the answer rather than absorb the answer into itself.

Zero-click did not begin with AI

It is tempting to describe AI search as a sudden rupture. In reality, search engines had been learning to answer questions directly for years. Weather panels, sports scores, currency converters, knowledge cards and featured snippets gradually moved information from the destination website onto the results page.

Each step looked like a convenience. Together, they changed the destination.

How an AI answer is actually built

A modern AI search answer generally involves three stages. First comes retrieval: the system identifies relevant pages, often drawing on the same web index that supports conventional search. Next comes synthesis: a language model combines information from several sources into one response. Finally comes presentation: the response appears above or instead of the familiar list of links.

The publisher’s reporting, data or explanation may still power the result. But the reader’s need can be satisfied before the source is opened. The value is extracted from the page; the visit is no longer automatic.

What AI search actually changes

The most visible change is a generated paragraph. The structural change is that an intermediary which once distributed attention can now retain it.

Pew Research Center analysed 68,879 Google searches associated with the browsing activity of 900 US adults. When an AI summary appeared, users clicked a traditional result in 8% of visits, compared with 15% when no summary appeared. A cited link inside the AI summary received a click in only 1% of visits.¹

Those figures do not prove that every publisher will lose the same amount of traffic. They measure a particular sample, period and search environment. But the direction is difficult to dismiss: a complete answer at the top of the page reduces the need to continue the journey.

A 2026 study using the staggered rollout of Google AI Overviews estimated that exposure reduced traffic to affected English-language Wikipedia articles by approximately 15%, with larger effects where a short synthesis could satisfy the user’s intent.²

Another 2026 measurement study found that AI Overviews appeared for 64.7% of question-form queries in its sample. It also reported that 11% of the individual claims it analysed were not supported by the cited pages, a reminder that compressing several sources into one confident voice creates an accuracy problem as well as a traffic problem.³

Who gets hurt, and who does not

There is no single thing called “a website”. A page defining a simple term, a newsroom conducting original reporting, a banking portal and an interactive calculator all use the same medium, but perform different jobs. AI search therefore does not threaten them equally.

The most exposed pages are those whose useful core can be compressed into a few sentences. The most defensible sites either create something that did not previously exist or allow the user to do something an answer cannot complete.

The real divide is not big versus small. It is replaceable information versus distinctive value. AI has not erased the value of good work; it has made generic, interchangeable work much harder to monetise through search.

Test the defensibility of a website

No score can predict the future, but the exercise below reveals the variables that matter. Select a page type and notice how the need to visit changes.

Google’s dilemma

Google is caught in a genuine dilemma. Its search engine became valuable because the open web was rich, and the open web was partly financed by the attention Google returned to it. The two systems grew together.

Now the instant AI answer is a competing interface. Google must make Search more conversational and self-contained or risk losing users to products that do. Yet every answer that prevents a visit can weaken the publishing ecosystem on which future answers depend.

This is not merely a dispute about fairness. It is an incentive problem. Publishers want their work discovered and rewarded. AI companies want broad access to current information. Users want the quickest reliable answer. The party controlling the interface decides how those interests are balanced.

Who feeds the machine?

AI answers can make information look as though it simply exists. It does not. Someone reported the event, inspected the documents, ran the experiment, maintained the database, tested the product, photographed the place, wrote the code or explained the concept.

Generative systems hide that production chain unusually well. Sources with different levels of confidence, context and cost are transformed into one smooth voice. The spring that fills the reservoir disappears from view.

The paradox is not inevitable. Licensing, subscriptions, stronger brands, direct audiences and new technical controls may break the loop. But the tension is real: an information system cannot indefinitely reduce the rewards for producing the information it consumes.

AI systems consume the web differently

Traditional search crawling was closely connected to referral. AI crawling can serve model training, search, retrieval for current answers or user-directed actions, and those uses return very different amounts of human traffic.

Cloudflare’s data covers traffic on its own network and is not a complete census of the internet. Even so, it exposes a striking imbalance. In July 2025, Cloudflare estimated that Anthropic crawled approximately 38,000 pages for each visitor it referred. It also reported that training accounted for nearly 80% of AI bot crawling during the period it studied.⁴

The new bargain being built

If the click was the old currency, publishers and infrastructure companies are now searching for a replacement. Some publishers are licensing archives to AI companies. Others are restricting crawlers or requiring negotiated access.

In July 2025, Cloudflare introduced “pay per crawl”, allowing a domain owner to permit, block or charge an AI crawler for access. It also changed its default approach for new customers so that AI crawlers could be blocked unless permission was granted.⁵

A licence could become a new mechanism for financing the information supply, a payment for access replacing some of the value once delivered by the click. It could also concentrate power if only the largest publishers can negotiate meaningful terms. The web is attempting to price something it previously gave away in exchange for attention.

Both stories can be true. Search can grow overall while particular classes of site lose traffic. Billions of clicks can remain enormous while a smaller share of an informational journey reaches the open web.

The decisive questions are therefore not merely how many clicks exist, but who receives them, for which queries, and whether the sites doing expensive original work can still finance it.

What people often misunderstand

“Websites are dying” collapses three separate developments into one panic. Discovery is changing: people are finding answers through AI interfaces. A business model is weakening: information no longer guarantees traffic. But the web itself remains intact: pages, domains, protocols and the ability to publish globally are not disappearing.

The web also helped create its own vulnerability. Years of competing for rankings produced vast quantities of thin, interchangeable content. That is precisely the material an AI system can replace most easily. The content most threatened by the answer box is often the content written primarily for an algorithm rather than for a reader.

What it means if you run a website

Do not compete to become the tenth adequate answer to a common question. Ask instead: if an AI can already state the answer, why would someone still come here?

The answer might be original evidence, a recognisable voice, specialist judgment, a useful tool, a transaction, a community, a dataset, an archive or an experience that cannot be completed inside a summary box.

The website increasingly becomes headquarters: the canonical home of the work, the full evidence, the archive and the identity. Discovery may happen through search, AI answers, newsletters, video, podcasts or social platforms, but the owned website remains where a publication can establish depth, permanence and a direct relationship.

Why it matters

For readers, the convenience is real and the cost is easy to miss. An answer with no visit is also an answer whose source may never be examined, whose author may never be weighed and whose publisher may not be rewarded for producing the next piece of useful information.

For the open web, the deeper risk is concentration. If only the largest brands, licensed archives and most defensible niches can finance original work, the web may remain technically open while becoming culturally narrower. That outcome is not inevitable, but it is the structure now being negotiated.

So, are websites dying?

No. But one version of the web is weakening.

The old assumption was that publishing useful information would naturally generate visits. AI search breaks that automatic connection. A website can now be crawled, extracted, summarised, cited and relied upon without being opened.

That is not the death of the website. It is the separation of being useful from receiving the user.

The sites most likely to endure will not survive by producing more words than AI. They will survive by producing what AI cannot originate on its own: new facts, accountable judgment, useful tools, distinctive experience and communities that choose to return.

The web is not dying. It is being asked to justify itself without a search engine quietly paying its way. The website is not disappearing; it is being forced to prove why the visit still matters.