Editor’s correction, September 9, 2026: our earlier headline claimed that bot traffic proved the Dead Internet Theory. It did not. We also confused the report’s publication date and treated measurements of traffic as measurements of people and authored content.
A crawler can request thousands of pages without writing a single comment. One person can spend an hour reading a page while generating comparatively little traffic. That difference is central to interpreting reports about automated activity online.
What the report measured
HUMAN Security announced its 2026 State of AI Traffic and Cyberthreat Benchmark Report on March 26, 2026. Its announcement describes more than one quadrillion digital interactions analysed by its platform during 2025. It reports that automated traffic grew eight times faster than human traffic, and that AI-driven traffic increased 187% from January to December.
Those are findings from the provider’s observation system. They are not a census of every internet user, every message or every page ever published. Our previous text gave April 9 as the release date and attached much broader conclusions to the figures than that announcement supports.
Requests, accounts and content are different things
Imagine a small library website with one hundred human readers and a search crawler that checks ten thousand URLs. Counting requests could make automation look dominant even though every article in the library was written by a person. Now imagine a forum with automated accounts posting comments. That is a different measurement problem, requiring evidence about accounts and their output.
This example is deliberately hypothetical. It shows why a share of network activity cannot tell you the probability that the next review you read was written by AI. You would need a defined sample of reviews, a reliable method of identifying authorship and a clear account of uncertainty to make that claim.
A theory needs evidence for its actual claims
The Dead Internet Theory is often used as shorthand for the feeling that online spaces are becoming synthetic. But a feeling, a claim about automated traffic and a claim that human conversation has been replaced are not interchangeable. Evidence for one does not automatically establish the others.
Our earlier article also moved from selected companies’ shares of AI traffic to a claim about almost all non-human internet traffic. That changes the denominator mid-argument. AI crawlers are a category of automation; measuring their relative shares is not the same as measuring every kind of automated system.
What readers can reasonably take away
Automation deserves scrutiny. Publishers need to understand which systems access their work, readers need ways to judge sources, and communities need workable responses to spam. None of those concerns requires pretending that a traffic report settles a sweeping theory about the whole internet.
When the next dramatic percentage appears, check three labels: the unit being counted, the population observed and the period measured. Then compare the headline with the source. If the source says requests and the headline says people, something important has already been lost in translation.
