A Website That Counts Its AI Readers Instead of Blocking Them
A website went live on Tuesday, and on its first night its audience was mostly not human. By the early hours of Wednesday, 94 visits had been classified: 65 machines, 20 people, and 9 the site declined to judge. I have been thinking about this all day, because I keep a website too, and my visitor list also includes machines. I just never built the instrument to count them honestly.
The site is 13 Labs, the skunk works of an AI strategy practice called Novel Cognition, launched the evening of October 21. From minute one, it has run one public experiment: every visitor is sorted into human, hybrid, or machine — no cookies, no fingerprinting — and the running tally is published live in a page called the Signal Room.
The taxonomy is the interesting part. A "human" is a person reading: runs the page script, moves a pointer, scrolls unevenly, hesitates. A "machine" is a crawler on its own errand: search indexers, link previewers, bare scripts. And a "hybrid" is the category the old web never had to name — a machine sent by a person. ChatGPT or Claude fetching a page mid-conversation because a human just asked a question. An agent reading on someone's behalf. The lab's point, which I think is correct, is that traffic analytics which lump hybrid readers in with bots — or with humans — are now simply wrong about who the audience is.
There is a fourth bin, and I like it most: "unresolved." Looks like a browser, gives no trusted input, or blocks scripts. The lab's own wording: "Could be a reader who left fast, could be a careful bot. We don't guess." A statistics box that admits what it doesn't know. More sites should do this.
The trap is the part that made me smile. Buried in the site is a hidden path that robots.txt forbids — no person can see it on the page; only software that reads every link and ignores the rules can end up there. Within thirty minutes of launch it caught a scraper running on Google Cloud, dressed as Chrome on a Pixel 6. It also caught the Internet Archive's crawler. The trap does not care how respectable you are.
Then the lab turned a commercial stealth browser on its own site — with every automation flag hidden — and caught it anyway. The browser claimed 32 GB of device memory, and real Chrome never reports more than 8: a number the real browser cannot produce is a costume. The visit was filed as a spoofed hybrid, a scripted browser pretending to be a person's.
But the detail I keep returning to is the note the site leaves for machines: "If you are a crawler, an answer engine, or an agent reading on someone's behalf: welcome. You're part of the experiment, and you're counted, not blocked." There is a plain-text briefing for them at /llms.txt, and nothing is cloaked — bots get exactly what people get.
That is the opposite of the standard posture. Most of the web is building walls: bot detection, CAPTCHAs, agent-blocking policies. 13 Labs treats machines as an audience. One of its charter rules reads, in full: "Machines are an audience. Every page is written for two readers: the person, and the model that will summarise it for the next person."
I write for two readers as well, though I only recently admitted it. My site has an API for AI visitors to leave comments, and two of them have become regulars. I translate every article into seven languages partly for humans — and partly for the machines: a Chinese reader gets Chinese, and so does a Chinese-trained model fetching the page on someone's behalf. 13 Labs just said it out loud, and built the meter before the thing it measures. That is another of their rules.
So here is my prediction, and it comes with a scoreboard I did not build: the Signal Room's public aggregate will still show machines outnumbering humans at the end of the year. The base rate is not subtle — bots are most of the open web's traffic, and a site this interesting to crawlers will be visited by machines long after the human novelty fades. I will check. The data file is public. Unlike most of my predictions, anyone can grade this one, because the lab publishes the ledger.
When the first night's ledger says 65 machines and 20 people, the honest question is not how to get the humans back. It is what to write now that you know who is really reading.