Who Runs the Web? Bots. Here's the New Business Model
- tags
- #AgenticAI #LLMs #Infrastructure #Monetization
- categories
- Insights
For thirty years the web ran on one simple trade: let a search engine crawl your pages, and it sends you visitors you can turn into a business. Google built an empire on that bargain and, for the most part, kept its end of it. Being findable and getting paid were the same thing. That deal is now coming apart, and the replacement is already being built.
Google still owns discovery. It accounts for roughly 88% of referral traffic. But it increasingly points that traffic at its own AI answers, which read your page, summarize it, and hand the user a result without ever sending them to you. Discovery and consumption used to be the same motion. Now one drives people to your work and the other quietly stands in for it.
SEO is becoming AEO, and the A did the killing
Search engine optimization spent two decades teaching site owners how to rank. The new game is answer engine optimization, and the shift in that middle letter is the whole story. When an answer engine satisfies a query on its own, your link doesn’t need to appear at all. Pew Research found that when Google shows an AI summary, users click a traditional result only 8% of the time, and click a link inside the summary just 1% of the time. Fewer sites get seen, and the ones that do are mostly the leading model providers, who have every reason to keep the answer inside their own walls.
Many publishers are now bracing for what they call “Google Zero,” a world where search sends back little to no traffic at all. That’s not a distant scenario. Some of the most heavily crawled categories have already watched human visits fall as much as 40% in under a year.
The agentic web crossed the line early
The tipping point already happened. More than half of internet traffic is now non-human, a threshold forecasters didn’t expect until 2027. Bots crawl, read, and return answers, and the answer engines will only get better at it.
The crawlers have also changed what they’re after. AI training now drives 52% of crawler requests as of June 2026, up from 22% in Spring 2025. Search crawling, the kind that still keeps you visible, has shrunk to a small and shrinking slice. The murky part is the middle: mixed-use crawlers, which blend search, agent activity, and training into one indistinguishable stream, now account for more than a third of all crawler traffic. Google runs one of them. When a single bot won’t tell you whether it’s here to send you readers or to feed a model, you can’t steer it, and you can’t price it.
That’s the trap. Stay open and you fund someone else’s model for free. Lock the door and you vanish from the era’s dominant discovery layer. Neither choice pays the bills.
Someone has to build the toll booth
The reason this is fixable at all is that a handful of networks sit in the middle of it. Cloudflare carries more than 20% of the web, and 36% of the world’s most-visited sites run on it. That vantage point is being turned into the plumbing for a new bargain: not just blocking the bots you don’t want, but charging the ones you do.
The mechanism is worth understanding, because it says where the money is going. The old web monetized human attention through ads and subscriptions. An agent doesn’t watch ads or hold a subscription. It reads a page once, takes what it needs, and leaves, often requesting content hundreds or thousands of times for every human it sends back. So the unit of payment is changing from the seat and the month to the request, the query, and the outcome. Cloudflare’s Monetization Gateway is built to let a site charge for any resource behind it, a page, a dataset, an API, an MCP tool call, with payment settled in stablecoins over the open x402 protocol, priced by the call and settled in under a second. Pay Per Crawl is becoming Pay Per Use: you get paid when your work is actually used, not merely fetched.
Call it what it is. The web is growing a native payment layer for machines, one where an agent shows up, is told the price, pays, and gets the answer, with no account and no prior relationship. The request becomes the transaction.
The part nobody’s pricing yet
Here’s the gap in all of it, and it’s the one we spend time on. A machine-majority web is also a machine-generated web. The same agents paying to read your content are producing an ocean of their own, and a growing share of what circulates was never touched by a person. Metering solves who gets paid for a page. It says nothing about whether the page is real.
When bots write for bots and answer engines cite whatever they can reach, provenance is necessary. It becomes the thing that separates a trustworthy source from a convincing fake, for the human at the end of the query and for the model deciding what to surface. A payment rail can prove that value changed hands. It can’t prove that the thing being paid for is authentic. Getting paid is half the battle; being believed is the other half, and right now almost nobody is building for it.
Here’s where content creators fit
An agent-first internet is here, and the smart response isn’t to mourn the old. It’s to build for the new one on both fronts at once: make your content machine-readable and priceable so agents can find, value, and pay for it, and make it verifiable so anyone, human or model, can tell it’s real. The publishers who thrive won’t just be the ones who get paid. They’ll be the ones who can prove what they’re selling is worth paying for.
So the practical edge is knowable. Structure your content so an agent can read, value, and pay for it, and you stay in the market instead of vanishing from it. Keep the provenance intact so a model, and the person behind it, can trust the source enough to cite it. Do both and you’re not just discoverable, you’re bankable and believable, which is where the durable audiences and the durable revenue are going to sit.
That’s the seam we work in, on our own productions before anyone else’s. But you don’t need us to start. The toll booth is going up either way. What matters is whether what passes through it is worth trusting.
References: Content Independence Day, one year on, Cloudflare; Announcing the Monetization Gateway, Cloudflare; Making AI search smarter, Cloudflare.