The busier half of the network becomes half its population
The strong form of Dead Internet theory proposes that apparent online human activity has largely been replaced by automation, with ordinary users surrounded by generated posts and simulated participants. Bot traffic statistics seem to provide a numerical foundation. But the argument frequently moves between three different things: requests sent to websites, content published on them and the number of people using them. A statistic for one does not count the other two.
The automated workload is substantial
Imperva’s 2025 report gives a 51 percent automated share for 2024, divided into 37 percent bad bots and 14 percent good bots. Its methodology describes observations across the company’s network and the domains it protects. Those are measurements of activity in that dataset. The report’s traffic categories do not label 51 percent of human-looking comments as generated.
The distinction between good and bad automation also matters to the claim. An automated request can come from a crawler or another machine process without impersonating a person in a conversation. A request counter registers the work performed, regardless of whether anyone sees a corresponding post.
A heavily automated workload is therefore compatible with several different populations. A small number of very active automated clients could produce it. A larger number of less active clients could produce the same total. Distinguishing those cases requires requests per client, followed by a separate account of how clients relate to operators and people.
Hold traffic fixed and change who generates it
The instrument allocates 100,000 illustrative requests according to the reported split: 51,000 automated and 49,000 human-operated. The request total only sets a convenient scale. Two adjustable averages specify how many requests each kind of client sends during that same observation period.
At 1,000 requests per automated client and 10 per human-operated client, the totals correspond to 51 automated clients and 4,900 human-operated clients. Automated clients make up about 1.03 percent of those modelled clients while sending 51 percent of the requests. That is a coherent numerical possibility, not an estimate of the actual client population in Imperva’s data.
Set both averages to 10 requests per client and the automated client share becomes 51 percent. The traffic observation has stayed fixed. The inferred population has changed because the workload per client changed. This is the missing quantity whenever a traffic percentage is promoted into a population percentage.
Even the calculated clients are not people. One person can operate more than one browser or device, and one automated operation can use multiple clients. The instrument defines a client only as a request-producing unit with the selected average. It cannot count authors, determine whether a post was generated or measure how many independent operators control the resulting activity.
Translate traffic into a conditional client mix
Holds the reported 2024 automated request share at 51%. Both requests-per-client averages are adjustable assumptions over the same period.
Conditional on the chosen requests-per-client averages; a client is a request-producing unit, not a person.
The number needed to supply the automated part of this illustrative workload.
The number needed to supply the remaining workload.
Working tape
- Human-operated request share
1 − 0.51 = 0.49 - Automated clients per request unit
0.51 ÷ 1,000 = 0.00051 - Human-operated clients per request unit
0.49 ÷ 10 = 0.049 - All client units
0.00051 + 0.049 = 0.04951 - Automated client fraction
0.00051 ÷ 0.04951 = 0.010301 - Automated client percentage
0.010301 × 100 = 1.030095 - Automated clients for 100,000 requests
0.00051 × 100,000 = 51 - Human-operated clients for 100,000 requests
0.049 × 100,000 = 4,900
More automated requests do not mean fewer real people
At the selected request averages, automated clients make up 1.03 percent of the modelled clients while generating 51 percent of the requests. Equal requests per client would make both shares 51 percent. The reported workload remains fixed; the client mix depends on the missing activity-per-client measurement.
Machines perform a great deal of work on the observed web. Establishing that a community consists mostly of simulated people would require evidence about that community’s accounts, posts and operators. A network traffic counter measures how often something requests a resource; a population count must also establish who or what sent the requests.