The reader isn't human anymore

Google users click a search result about 8% of the time when an AI-generated summary sits at the top of the page. Without the summary, they click 15% of the time. The link inside the summary, the source the AI is quoting, gets clicked 1% of the time. Those three figures come from Pew Research Center, which watched 68,879 real searches run by 900 US adults in March 2025.

Part 1 argued that the web's audience stopped being human-majority. This is the first consequence, and it is a money problem: the open web was funded by a human who saw an ad or created an account, someone whose attention the page could charge for. Answer engines serve the answer without the click, and agents were never going to see the ad at all. The reader the business model was written for is leaving, and the revenue is leaving with them.

The Pew gap is a correlation, not a proven cause: Pew's own breakdown shows summaries surface most on longer, question-shaped queries, the kind people may never have clicked through on anyway.

Traffic was always a stand-in for attention

For twenty years "traffic" was the number everyone optimized, because it stood in for the only thing you could actually sell: human attention. A visitor might see a banner, remember a brand, fill a cart, or come back tomorrow. Page views were a proxy for that, and the proxy held for as long as the visitor was a person.

Machine traffic keeps the number and drains what it measured. A crawler registers in your analytics as a request, the same as a reader does, but there is nobody behind it whose attention can be sold. You can serve ten million requests and monetize none of them.

Answer engines do the same thing from the other end. The old search bargain was that a crawler indexed your page and, in exchange, sent you a human. AI summaries keep the first half and quietly drop the second: the system reads your work to compose the answer, delivers it in place, and the reader has no reason to come to you.

Ads, subscriptions, and SEO each priced a human

Advertising is a bet on eyes. The advertiser pays for an impression on the theory that a person will glance at it, and pays much more for a click on the theory that the glance can be turned into a customer. An agent fetching a page to extract one fact renders nothing and buys nothing from the sidebar. There is no impression to sell because there is no attention to interrupt.

Subscriptions and accounts assume someone who will come back. A login is only worth acquiring if the person behind it returns often enough to be upsold and retained over months. An agent creates no account and keeps no session. Between calls it forgets you exist. Lifetime value assumes a lifetime, software has none.

Search optimization is the strangest case, because it was the one pillar built explicitly for a machine, on the understanding that the crawler was a courier carrying a human back to you. Rank well, earn the visit, monetize the visitor. The newer crawlers broke the contract: they index the page and answer the question themselves, so the ranking you fought for now feeds a summary that replaces the click it used to earn.

The extraction ratio

The clearest way to see the imbalance is to count how much a crawler takes against how much it gives back. Cloudflare did exactly that, measuring pages crawled per human visitor referred across its network in July 2025.

Those figures are from Cloudflare's "crawl-to-click gap" analysis, and two things about them need saying out loud. First, they are one company's view of its own network on one date, and they move fast: Anthropic's ratio was 286,930:1 in January 2025 and had fallen roughly 87% by July. Second, the spread is enormous. Google sits near parity because it still runs a search product that sends clicks back. Anthropic, which mostly does not, sits about 7,600 times further out.

The reason the ratios skew so hard is what the crawling is for. By Cloudflare's count, 79% of AI crawling in July 2025 was for training, against 17% for search. Most machine reads of your work were never going to send anyone back, because sending a human back was never the goal. The content is raw material for a model, and the model is the product.

The first casualty

If you want to know what happens to a business whose readers get intercepted before they arrive, look at Chegg. It sold homework help and study tools to students who found it through search, and that funnel was the company. When the funnel narrowed, so did the company: Chegg cut about 45% of its workforce, 388 people, in October 2025, after a 22% cut earlier the same year. In February 2025 it sued Google, arguing that AI Overviews were keeping students on the results page and starving the sites those answers were drawn from. By its own account, its non-subscriber traffic went from down 8% in Q2 2024 to down 49% by January 2025.

Chegg is hollowed out and pivoting. Its troubles predate AI Overviews: some of that traffic was already going to free tools and shifting study habits. AI search is the accelerant here, not the sole cause. The vulnerability was already sitting in every business whose only real asset was a search referral. AI answers found it, and they take the referral away faster than it can be replaced.

Chegg is legible because it was almost purely a search-funnel business. Most publishers are the same story with more padding: a bit more brand loyalty, a bit more direct traffic, the same dependence on a click that answer engines are learning to keep for themselves.

If not ads, then what?

Strip out the drama and the problem is plain. The open web was paid for by a human who saw the page, and the two fastest-growing kinds of reader (the answer engine that summarizes it and the agent that consumes it as data) remove that human from the loop while still consuming everything the page cost to produce. The old model fails business by business, starting with the ones whose only asset was a search referral.

The costs remain, and the question of who covers them is wide open. The obvious first move is to make the machine that takes the content pay for it at the moment it takes it. That is possible today, through a door the web built in the 1990s and never opened: a payment mechanism it reserved and then left blank for twenty-five years. The next part walks through it.

This is Part 2 of The Machine Web, a series on what changes when the web's main character stops being human.

Part 1: The Machine Web

AKENA is a blockchain engineering studio. We build the infrastructure AI agents run on: agent-facing RPC and MCP endpoints, on-chain data pipelines, and the products on top of them. If your readers are turning into machines and your revenue was built for humans, we should talk.