The Zero Click Internet Has a Supply Problem

That is not a projection. SparkToro, working from Similarweb's clickstream panel, put the figure at 68.01% across the first four months of 2026. A decade ago it was roughly 45%. In 2024 it was 60.45%. So most of that movement happened in about twenty four months, which tells you something about how fast the ground shifted.

The phrase people use for this is the zero click internet. You ask, you get an answer, you close the tab. Nobody visits anybody.

For most people this is just better. You wanted to know when the store closes and now you know when the store closes, without loading a page carrying four ad units and a newsletter popup. Hard to argue with that. The trouble is that the ad unit and the newsletter popup were paying for the person who found out when the store closes.

What actually changed

AI Overviews get most of the blame and they deserve a decent share of it. They appear on more than a fifth of searches, and when one shows up, click through rates fall by roughly 60%. Seer Interactive measured organic CTR dropping from 1.62% to 0.61% on queries where an Overview appeared.

But the Overview is a symptom. The bigger shift is that answering the question inside the results page turned out to be good business, so every surface that can answer a question directly now does. Featured snippets. Knowledge panels. Related questions that expand into more questions. Search results that lead to more search results rather than to a website. SparkToro found the share of searches leading to another Google search climbed 7.2 points over the same two years.

Then there are the chatbots, which skip the pretense entirely. No results page, no links, just the answer.

The part that doesn't add up

Here is the technical picture, briefly.

• Bots now generate more web traffic than people do. Cloudflare Radar clocked automated requests at 57.5% of HTML traffic in June 2026, against 42.5% human. Cloudflare's CEO had predicted the crossover would happen in 2027.
• Most of that crawling is not search. In May 2026 Cloudflare attributed 51.8% of AI crawler requests to model training and under 10% to search. Training crawlers send back nothing, because there is no link to send.
• The exchange rate is lopsided and getting worse. Cloudflare tracks a crawl to referral ratio: pages taken per visitor sent back. Google sits around 5 to 1. Reported figures for AI crawlers range from the low hundreds to the tens of thousands to one, depending on the operator and the month.

Traditional search had a bargain underneath it. You let the crawler in, the crawler sent you readers, the readers paid the bills through ads or affiliate commissions. It was never a generous bargain but it was a bargain, and it is the reason a one person website about a narrow topic could support a person.

Strip out the referral and there is no bargain. There is just extraction.

The loop

Now follow it forward.

AI systems need content. Not the content they trained on last year, either. Fresh content, about things that happened recently, written by people who went and found out. The store hours change. The law changes. The recommendation that was right in 2024 is wrong now.

That content comes overwhelmingly from the same open web that the systems are quietly defunding. Not the big publishers, who can sue or sign a licensing deal or both. The small ones. The obsessive niche sites that document one city or one hobby or one piece of software better than anyone else, and that exist because affiliate commissions made them viable as a job rather than a hobby.

When those sites stop being viable, they do not announce it. They just get updated less, then not at all, then the domain lapses. Nobody notices for years, and by then the answers have quietly gone stale and nobody can tell you why.

Which puts the AI companies in a strange position. The thing they are disrupting is their own supply chain.

What people are trying

Nobody has solved this. Several things are being attempted:

• Licensing deals. Content in exchange for money or technology access. Works if you are big enough to have a phone number the AI labs will call.
• Litigation. Several publishers are suing over training use. Slow, expensive, and the outcome will take years.
• Blocking. Intuitive, and less effective than it sounds. Research from Rutgers and Wharton, revised in April 2026, found news publishers who blocked LLM crawlers did not recover the traffic they expected. Robots.txt is a request, not a wall.
• Charging. The most interesting one. Cloudflare has built pay per crawl into the plumbing, returning an HTTP 402 to unpaid crawlers, and from September 2026 will block mixed use crawlers by default on monetized pages for new sites. Roughly a fifth of the web runs through Cloudflare, so this is not a gesture.
• Optimizing to be cited. The consensus advice, which is to stop chasing raw traffic and get your brand named inside the answer instead. Brands cited in AI Overviews do see meaningfully better click through rates than uncited ones. Whether being mentioned pays as well as being visited is a question nobody has answered honestly yet.

Where this leaves us

The honest answer is that the economics of the open web are being renegotiated in public, quickly, and by parties with wildly unequal leverage.

Ask an AI what to do about it and you will get told to build brand authority. It would say that. It is the one answer that requires nothing from anyone except you.

But the underlying question is not a marketing question. It is whether the next decade produces the sort of web worth summarizing at all, or whether we spend it recycling a snapshot of 2024 and slowly forgetting that somebody used to go and check.

Sources

• SparkToro / Similarweb, zero click study, June 2026
• Seer Interactive, AI Overview CTR analysis
• Cloudflare Radar, bot traffic share, June 2026
• Cloudflare, crawl to refer gap data and Pay Per Crawl announcements
• Rutgers Business School and The Wharton School, LLM crawler blocking study, revised April 2026