Posted on

Bot traffic versus bot-authored public conversation

A statistic such as “53 percent of web traffic is automated” sounds like it should tell us how much of the internet is made by bots.

It does not.

It tells us something important, but narrower: machines are responsible for a large share of requests reaching websites and applications. That is a measurement of traffic, not a census of who wrote the visible conversation.

Imperva’s 2026 Bad Bot Report says automated systems generated more than 53 percent of observed web traffic in 2025. That is a remarkable number. It is also easy to misuse.

A request is not a comment

Web traffic measurements usually count HTTP requests. A crawler fetching an article creates a request. A monitoring service checking whether a page is alive creates a request. A scraper downloading prices creates requests. An API client retrieving data creates requests. An attacker testing credentials creates many requests very quickly.

None of those activities necessarily writes anything that another person will read.

Google describes Googlebot as the crawler used by Google Search. It visits pages automatically so they can be indexed. Those visits are bot traffic in the literal sense, but Googlebot is not sitting in the comments pretending to be your uncle.

The mismatch also works in the other direction. One automated account can publish thousands of posts while producing only a modest fraction of a platform’s total network traffic. Meanwhile a single human opening a modern page can trigger requests for HTML, images, scripts, fonts, analytics, advertisements, APIs, and background updates.

The units simply do not map cleanly.

Measuring synthetic conversation requires conversation data

If the question is “What share of public discussion is machine-authored?” the sample has to contain public discussion: posts, replies, comments, messages, or other defined conversational units.

Researchers then need a defensible method for deciding which items are automated. That may involve account behavior, posting tools, content provenance, network patterns, manual review, disclosures, or known ground-truth accounts. Each method has uncertainty and false positives, but at least it measures the thing being claimed.

A traffic report cannot substitute for that work.

This distinction matters for Dead Internet Theory because traffic statistics are often used as if they prove a stronger proposition: if most requests are automated, then most apparent human activity must also be automated. That conclusion does not follow.

The automated web is already enormous. Search crawlers, security scanners, monitoring systems, AI agents, scrapers, spam tools, and malicious bots really do talk to servers all day without a person clicking anything.

That fact is worth studying on its own.

But a machine requesting a webpage and a machine impersonating a person in a conversation are two different events. Counting the first cannot tell us how common the second is.