Bots Took Over the Internet. Guess What Still Works.
Machines generate 57.5% of web requests. The one channel where you can still count the humans is the one finance has been trying to kill since 2019.
Cloudflare reported that bots now generate 57.5% of web page requests. Sometime in June 2026, humans became a minority on the internet they built. Matthew Prince, Cloudflare’s CEO, had put that crossover at the end of 2027. It arrived more than a year early. Thales, in its annual Bad Bot Report, argues machines took the majority back in 2023. The two firms disagree about the date and agree about the fact.
Rita McGrath, writing about it in Fast Company on 28 July: “The ‘dead internet theory’ is real. It’s now just a measurement dispute.”
What changed isn’t the volume, it’s what the machines are doing. Traffic from agents that act on the web, clicking links, filling in forms, completing tasks, grew 7,851% year over year according to Human Security’s 2026 benchmark. Those aren’t crawlers indexing your pages for a search result, they are software running errands for somebody. Stripe says roughly 70% of the commands hitting its data APIs now come from agents rather than people.
Earlier this week I wrote that you never owned your traffic, you rented it. That was about who decides to send you a visitor. This is about what the visitor turns out to be.
Gary Vaynerchuk’s barbell: extreme tech on one end, extreme analog on the other
On 24 July, Gary Vaynerchuk published The Pendulum Is Swinging Back to Real Life. He isn’t responding to any of the above. He doesn’t mention bots, Cloudflare or agents once, and as far as I can tell he wrote it without the traffic story in front of him.
His argument is that the explosion of AI is the precursor to an explosion of analog. He sees a barbell coming: extreme technology on one end, extreme physical on the other, and the average middle squeezed out of existence. He isn’t writing it as a think piece either. Vayner hired a senior vice president of experiential in June and is pushing the agency into stores, sampling, restaurants and events. He points at Capital One’s cafés and airport lounges, Ralph Lauren’s cafés, Coach, Tiffany’s Blue Box, as evidence that brands with money have already started buying physical rooms.
His case is cultural. People are tired of screens, kids want the nineties back (I’m more for the eighties though ;-) ), run clubs and festivals are full. He is probably right, and yearning is a hard thing to defend in a budget meeting.
The traffic data says something narrower and more useful: A room is a place where you can count who is in it.
Where the barbell doesn’t transfer
Drive-in cinemas, run clubs, a café inside a handbag store. Those are consumer plays, and consumer brands are who Gary sells to mostly. If you run a B2B company with sixty accounts in Europe, extreme analog is not a music festival.
It’s the physical site visit. The user group of twelve people in a room who all have the same problem. The factory tour. The two days at a trade fair that finance has been trying to kill since 2019. The pilot installed at a customer’s plant with your own engineer standing next to it for a week.
Same barbell, different equipment. What transfers exactly is the cost structure. Analog doesn’t scale, a room holds what it holds, every contact costs real money and real hours. That constraint used to be the weakness of the channel. It’s now the thing that makes it countable.
Finance will cut the wrong thing, again
Most B2B budget arguments come down to cost per contact. Digital contacts are cheap and plentiful, in-person contacts are expensive and few, and the events line loses that argument nearly every year.
The comparison was already unfair, because a session and a conversation were never the same unit. Now one side of it contains machines. Agentic browsers push session counts up and bounce rates down, and standard analytics won’t flag a thing. So the review shows digital getting more efficient while events stay stubbornly expensive, and the recommendation writes itself. Cut the trade fair, move it into performance.
That would be cutting the one channel where you can verify a human, to fund the one where you can’t, on the advice of numbers that can no longer tell the difference. The spreadsheet looks great, the business is dying was about this reflex. Same mistake, new version, except this time the spreadsheet isn’t even reporting on people.
Four things to run this quarter
Ask a sharper question than “how much bot traffic do we have”. Your analytics never saw most of it anyway. Google Analytics is JavaScript running in a visitor’s browser, and the average crawler doesn’t run JavaScript, so it was invisible to you the whole time. Cloudflare’s 57.5% counts requests, your dashboard counts sessions, and comparing the two tells you nothing. What contaminates a marketing number is narrower and newer: agents that drive a real browser, render the page and fire your tag exactly like a person would. Ask whoever runs the site for rendered page loads broken out by known AI agent, GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot and the rest. Then look at your form submissions for completions in under a second and for datacentre IP ranges, because a polluted lead count will cost you more arguments than a polluted session count ever will. Depending on your CDN plan the detailed breakdown may sit behind a paid tier, so find out what you can actually see before you promise anyone a number.
Count how many deals you closed last year that had a face-to-face before signature. It’s in your CRM (well, it should be…) and most teams have never run it. That number is what holds the line when somebody proposes cutting travel and events, because it’s the one measure of the physical channel that no bot can inflate.
Give the agents an owner. McGrath’s point, and she’s right about it. Agents are a new customer segment. They don’t browse, they qualify on structured data, price, availability and reliability. Somebody should own how your company looks to a machine, the way somebody owns key accounts. Right now nobody does, which is either a gap or a head start depending on how fast you move.
Stop opening the monthly review with sessions. Open with meetings held, customers visited, and how many humans you can reach tomorrow morning without asking anyone’s permission. Whatever gets reported first is what gets defended in December or whenever you budget cycle is.
In the end
Thirty years of internet marketing ran on one promise: reach more people for less money. It delivered on both halves. Nobody wrote down that the people were the point, because at the time that was too obvious to write down.
Gary is betting his agency on the room, loudly and early, which is his pattern. You don’t need to open a drive-in cinema to act on the same read. You need to know how many of your customers you have actually met, and that answer is sitting in your CRM this afternoon.
Sources
Cloudflare’s 57.5%, the June 2026 crossover and Matthew Prince’s earlier end-of-2027 prediction: as reported by Fortune, 23 July 2026, “Turns out Dead Internet Theory was right”. https://fortune.com/2026/07/23/dead-internet-theory-bots-agents-majority-web-traffic/
“It’s now just a measurement dispute”, the personhood-as-premium argument and the agents-as-a-segment recommendation: Rita McGrath, “The ‘dead internet theory’ is real. And it’s killing the web as we know it”, Fast Company, 28 July 2026. https://www.fastcompany.com/91579817/dead-internet-theory-is-real-web-agents
Gary Vaynerchuk, “The Pendulum Is Swinging Back to Real Life”, LinkedIn and Substack, 24 July 2026.