If It's All Bots, What's the Point?

There is a moment in every technology story when the impressive number stops being the interesting part. Adriel recently highlighted one of those numbers: bots now account for a majority of requests to web pages. But the figure does not really become strange until you keep following the chain. Who made the page? Who is reading it? Who is deciding whether it matters? Who is summarizing it? Who is acting on the summary? Increasingly, the answer to several of those questions can be some form of software.

The underlying milestone comes from Cloudflare, not Adriel. Cloudflare Radar tracks the share of HTTP requests to HTML content that its systems classify as automated or human, and in early June 2026 the balance crossed decisively toward machines. Reporting based on Cloudflare's data put the split at roughly 57 percent bot requests to 43 percent human requests. Cloudflare CEO Matthew Prince, who had expected the crossover later, said agentic traffic had pushed the web past the line sooner than he predicted. The measurement is specifically about requests for web-page content, which is exactly why it matters here: this is the part of the internet built to be read.

The shift did not arrive from nowhere. Cloudflare's 2025 year-in-review data already showed non-AI bots responsible for about half of HTML-page requests at the beginning of that year, while AI bots averaged another 4.2 percent and Googlebot alone averaged 4.5 percent. By December 2025, human requests had climbed back to about 47 percent and non-AI bot requests sat around 44 percent, but the larger trend was obvious. Automated systems were already a huge part of the audience. The arrival of agents capable of browsing, comparing, purchasing, booking, researching, and acting on behalf of users simply gave the machine side another accelerator.

HUMAN Security's 2026 State of AI Traffic report illustrates the acceleration. Across its customer data, automated traffic grew eight times faster than human traffic in 2025. AI-driven traffic nearly tripled over the course of the year, while traffic from AI agents and agentic browsers increased 7,851 percent year over year. Most of that activity concentrated in retail and e-commerce, streaming and media, and travel and hospitality - exactly the places where a machine can search large amounts of structured information and then make or influence a decision for a person.

That sounds efficient until you ask what the web becomes when the efficiency works too well. A human shopping for a camera might open a handful of pages, compare a few options, get distracted by a review, decide one feature matters more than expected, and eventually buy something. An agent can inspect hundreds or thousands of pages because it does not experience browsing as time, annoyance, visual clutter, or decision fatigue. Request counts therefore explode even when the number of humans with an actual need has not changed. One person can now arrive on the web wearing a very large mechanical footprint.

Cloudflare has started describing the next phase openly as the "Agentic Internet." In an August 2026 post, the company said well-behaved bots were already making billions of requests to re-fetch pages that had not changed, producing what Cloudflare called an enormous amount of machine effort attached to no outcome at all. Its proposed answer is not to send the bots away. It is to rebuild parts of the web for them: machine-readable formats, cryptographic bot identity, agent-oriented search, payment rails, WebMCP tools that expose actions directly, and even browsers stripped of the human-facing overhead that agents do not need.

That is a perfectly logical engineering response. It is also where the question gets wonderfully ridiculous. We built the web for humans. Then we built machines to read the web for humans. Now the machines are generating so much traffic that we are redesigning the web so the machines can read it more efficiently. The hero image, the button, the clever layout, the carefully tuned call to action - all of the things designed to persuade a person - can become irrelevant to an agent that wants the underlying fact or action and would rather receive it as structured data.

Meanwhile, the supply side is becoming more automated too. Ahrefs analyzed 900,000 English-language pages newly detected in April 2025 and estimated that 74.2 percent contained at least some AI-generated content. Only 2.5 percent were categorized as purely AI-generated, while most of the AI-involved pages mixed human and machine work, but the direction is still important. The machines are not merely reading the shelves. They are increasingly helping stock them.

Then another machine comes along to decide which shelf item is useful. Cloudflare's crawl-to-refer data showed how lopsided that relationship can become. In July 2025, Anthropic's crawlers made roughly 38,000 page requests for every visitor referred back to a source site. OpenAI's ratio was about 1,091 to one. Cloudflare also found that nearly 80 percent of AI crawling activity was related to training rather than search or direct user actions. The old bargain of the web was imperfect but understandable: let a search engine crawl your material and, in return, people may be sent to you. AI systems can consume vastly more while returning vastly fewer humans.

This is where the web begins to resemble the same conceptual puzzle appearing in autonomous racing and robot sports. The Abu Dhabi Autonomous Racing League is trying to build an international championship for fully autonomous racecars, and RoboCup has maintained an even more audacious research goal since 1997: develop a fully autonomous robot soccer team capable of beating the human World Cup champions by 2050. Both are legitimate research programs. Racing gives engineers a punishing real-world test of perception, prediction, control, and safety. Robot soccer forces machines to solve cooperation, movement, strategy, and decision-making under pressure.

But the spectacle still has a human frame. People designed the challenge. People watch it. People decide what counts as progress. The autonomous car does not care that it won at Imola, and the robot striker is not going home devastated after missing a penalty. The competition means something because humans have chosen to make the performance meaningful. Remove that human purpose and you are left with an extraordinarily sophisticated benchmark running laps for itself.

The web faces a similar risk. An agent searching on behalf of a person still has an obvious point. If it finds the cheapest flight, books the hotel, compares insurance policies, locates a difficult piece of research, or spares someone 40 minutes of clicking, the chain ends in human value. The machine is an intermediary. The problem begins when more of the system becomes self-referential: AI-assisted pages created partly to rank in AI answers, crawled by AI systems, summarized by AI systems, measured by AI-visibility tools, and optimized so another AI system will prefer them next time.

At that point, traffic itself becomes a very strange success metric. A million machine requests can sound like reach while representing almost no human attention. A page can become highly legible to agents without being memorable to a person. A company can win the competition for machine discoverability and still have to answer the embarrassingly basic question that marketing was supposed to solve in the first place: did anybody actually care?

There is a business-model problem hiding underneath that question. Prince put it simply in June: bots do not click ads. The web's advertising economy was built around human attention, and much of publishing assumed that a crawler created value by eventually routing people back to the source. If an agent can absorb the useful portion of a page, synthesize it elsewhere, and complete the user's task without a visit, the publisher has supplied the raw material while losing the moment where attention used to become revenue. Cloudflare's push toward a payable agentic web is one possible response: if bots are the visitors, perhaps bots need to become customers too.

That may be where this is heading. Websites may increasingly operate in two layers: an expressive human layer built for curiosity, trust, identity, storytelling, and experience, and a machine layer designed for agents that want facts, permissions, tools, and transactions. There is nothing inherently wrong with that. In fact, it may be a much cleaner architecture than forcing software to pretend it has eyes, move a fake cursor around a screen, and click a button labeled "Buy Now."

But the human layer cannot become ceremonial. The reason to publish, report, photograph, test, review, investigate, and experience things is not merely to manufacture another unit of machine-readable material. Original observation matters precisely because it starts outside the loop. Someone went somewhere. Someone tried the product. Someone interviewed the person. Someone noticed the contradiction. Someone laughed at the absurd thing. Someone cared enough to make a judgment that did not exist until a human being made it.

That is also why the bot-majority milestone is more interesting than a content-strategy warning. It forces a definition. The internet can survive being mostly traversed by machines if those machines are doing work for people. What becomes harder to defend is an internet where machines increasingly produce material for machines to consume so that machines can report to other machines about how effectively the machines are being reached.

If that is the destination, the problem is not that the bots won. It is that we forgot what the race was for.

SOURCE NOTES

• Adriel - Bots Now Outnumber Humans Online. Here's What It Actually Means for Your Content Strategy (June 17, 2026)
• Cloudflare Radar - Bot vs. Human traffic measurement
• Cloudflare - 2025 Radar Year in Review
• Cloudflare - Building an open Agentic Internet (Aug. 6, 2026)
• Cloudflare - The crawl-to-click gap: AI bots, training, and referrals
• Cloudflare - Control content use for AI training
• HUMAN Security - 2026 State of AI Traffic & Cyberthreat Benchmark Report
• Ahrefs - 74% of New Webpages Include AI Content
• A2RL - Autonomous racing international debut at Imola
• RoboCup Federation - Objective