You are never as well served as by yourself. AI agents no longer just answer questions. They click, fill out forms, compare prices, buy instead of humans, often without anyone noticing on the other side of the screen. A theory long mocked as a conspiracy fad, that of the “dead internet” (the idea that the majority of online content and traffic is generated by bots rather than by real people), finally seems to find very real numerical confirmation. So it’s no longer just a question of volume. It is a change in nature: the web is moving from a human consultation space to an infrastructure for transactions between machines. A click no longer has the same economic significance depending on whether it comes from a human or an agent who compares 200 sites in 3 seconds.
The key points of this article: Traffic generated by AI agents on the web has seen a spectacular increase of 7,851% in one year, according to HUMAN Security.
Cloudflare revealed that bots have surpassed humans, generating 57.5% of web page requests since last June.
A surge in traffic that is beyond comprehension
According to a report from cybersecurity company HUMAN Securitytraffic generated by agents who actually act on the web (clicking on links, filling out forms, placing orders) jumped 7,851% year-on-year. A figure that makes you dizzy, even for a sector accustomed to exponential curves. As Fortune reports, Cloudflare, for its part, places the tipping point last June: bots now generate 57.5% of web page requests, more than humans. The numbers vary depending on who is measuring, and that’s the whole problem. The French Thales, in its “Bad Bot Report”, rather estimates bot traffic at 53% in 2026, while putting the change back to 2023. No supplier having visibility over the entire web, each measures noon at its door, with its own methodology and its own sample of sites monitored. For its part, Cloudflare predicted this shift for the end of 2027. It arrived 18 months earlier.
A shift that reshuffles the cards for an entire ecosystem
The underlying trend is hardly debated. THE so-called “scraper” traffic (bots that suck up content to feed databases or train models) increased by 597% over the same period, even if its relative share in overall AI traffic declines slightly, as “active” agents take over from simple data collectors. Symbol of this shift: developer Matt Schlicht launched Moltbook at the beginning of the year, a social network designed exclusively for AI agents, which brought together more than 1.5 million active accounts in just a few days. For advertising, analytics and fraud, the historical model is collapsing. An AI agent that compares prices or places an order for a user behaviorally resembles a malicious bot. Detection systems, calibrated to humans, gradually become blind or too aggressive. Result: the value of a click and a session must be completely recalibrated. Indeed, an advertising click no longer has the same value if it comes from an agent programmed to compare prices rather than from a curious consumer. And fraud detection systems, calibrated on human behavior, must now learn to distinguish a malicious bot from a perfectly legitimate AI assistant shopping on behalf of its user. There tokenization of a verifiable identity for each agenta project already supported by figures like Vint Cerf, has never seemed as urgent as today.