What does "dead internet theory" mean? The slang, the origin and what this court tests

Dead internet theory is slang for the claim that most of the web is now bots and automated content. It bundles three claims: bot traffic, AI-written text and a deliberate plot.

By Jack Stovell · Published 29 October 2026

The short answer: what people mean when they say it

Wikipedia defines it as “a concept that asserts that the Internet consists primarily of bot activity and automated content manipulated by algorithmic curation”. It adds that the idea was “originally conceived as a conspiracy theory” and is now also used colloquially “to describe the impacts of generative AI”.

In everyday use the phrase is more a reaction than a thesis. Someone sees a reply thread full of near-identical comments, or a search result that says a great deal and tells them nothing, and reaches for the label. The feeling is real. Whether it counts as evidence depends on which claim sits underneath it.

Jake Renzella of UNSW and Vlada Rozova of the University of Melbourne, writing in The Conversation in May 2024, sum up the popular version: “AI and bot-generated content has surpassed the human-generated internet.” Their conclusion is more modest. The theory, they write, “is a reminder to be sceptical”.

Where the phrase came from

It began on internet forums. Wikipedia traces the best-known version to a 2021 post titled “Dead Internet Theory: Most Of The Internet Is Fake” on the forum Agora Road’s Macintosh Cafe. That post said it built on earlier imageboard posts. Wikipedia also says the precise origins are hard to pin down, and it marks its source for the very first post as needing a better one. Treat 2021 as the year the phrase spread. Nobody can say for certain when the idea first appeared.

From there it moved to YouTube, Reddit and Twitter, then into journalism and eventually academic papers. One of those papers is a survey by Prathamesh Muzumdar and five co-authors, posted to arXiv in January 2025 as “The Dead Internet Theory: A Survey on Artificial Interactions and the Future of Social Media”. Its abstract describes the theory as suggesting that “much of today’s internet, particularly social media, is dominated by non-human activity, AI-generated content, and corporate agendas”. The authors then redefine it around social media platforms that put engagement metrics ahead of human connection. The abstract presents that as an argument and reports no measurement.

Three claims hiding in one slogan

The slogan packs three separate claims together. Each needs a different kind of evidence.

Bot traffic. A large share of the requests hitting websites come from automated programs. The evidence is server logs and network measurements, usually published by security firms.
Machine-written text. A large share of what gets published (articles, posts, product pages) was produced by a language model. The evidence is samples of text, tested with detectors or with measurable features of the writing.
Deliberate control. Someone is running all this on purpose to steer what people think. Wikipedia calls this the part “where the conspiracy portion begins”, and the original post blamed the US government. The evidence would have to show intent and coordination, and no traffic log or word count can supply that.

The three are independent. You can accept the first and doubt the second. You can accept the second and reject the third outright. Most arguments about dead internet theory go nowhere because each side is defending a different claim from that list.

Bot traffic is not bot writing

The most quoted figure belongs to the first claim. Imperva, a security company, says in its 2025 Bad Bot Report: “For the first time in a decade, automated traffic has surpassed human activity, accounting for 51% of all web traffic.”

That is a vendor figure from a firm that sells bot protection, and it should be read as one. More to the point, it counts traffic. A request is a machine asking a server for something: a page, a login attempt, a price. A crawler that fetches ten thousand pages has written none of them. A bot hammering a login form produces no prose at all.

So “half the internet is bots” is a statement about the plumbing. It says nothing about who wrote the words on the pages being fetched. A headline that uses the traffic figure to prove the writing claim has swapped one question for another.

The one corner this court tests

This site tests a narrow slice of the second claim: has web writing converged on one machine voice since ChatGPT? The question and the reasoning behind it are on the theory page. Traffic is out of scope, and so is intent.

Even inside that slice, the evidence keeps mixing up two questions.

Prevalence asks how much text is machine-written. The websites nobody wrote is a prevalence case: a preprint that uses a detector to classify whole websites as LLM-dominant. Half the articles, and a plateau is another, a vendor study that runs Common Crawl articles through three commercial detectors. Both carry the caveat that comes with any detector. Some human writing gets flagged as machine writing, and across millions of pages a small error rate adds up to a lot of pages.

Convergence asks whether writing has become more alike. Did ChatGPT change how companies write? comes closer to that question. ScriptGrain’s own study compares company homepages with their own earlier versions, word by word and sentence by sentence, using code rather than a detector. The case’s headline puts its finding in seven words: “The words changed. The sentences did not.”

A high prevalence figure does not prove convergence. Machine text could still be varied, and human writers could drift together with no machine involved. Each question needs its own measurement. For a related idea, models trained on their own output, see what is model collapse.

How to use the phrase without overclaiming

Say which claim you mean.

If you mean traffic, say traffic, name the report and note who published it.
If you mean text, say how it was measured. A detector estimate needs its false-positive rate beside it.
If you mean sameness, ask for a convergence measure. A prevalence figure cannot answer that question.
If you mean the plot, call it a conspiracy claim and ask for evidence of intent. A feeling that the web is empty is no such evidence.

The phrase works as a prompt to check where something came from. It makes a poor conclusion to reach before you have checked.

Sources

Questions people ask

What does dead internet theory mean in slang?

It is shorthand for the claim that most online activity and content now comes from bots and automated output rather than people. Wikipedia notes it began as a conspiracy theory and is now also used loosely to describe the effects of generative AI. In casual use it often just means "this feels fake".

Why do people say dead internet theory?

Usually because a reply thread, a feed or a search result feels machine-made, and the phrase names that feeling quickly. Being short and memorable helped it spread. It also blurs bot traffic, AI-written text and the conspiracy into a single idea.

Is it true that AI has destroyed the internet?

That claim is too broad to answer yes or no, because it mixes traffic, writing and intent. This site tests one narrow part of it: whether web writing has converged on one machine voice since ChatGPT. So far the cases point in different directions.

Is the internet 51% bots?

Imperva's 2025 Bad Bot Report says "automated traffic has surpassed human activity, accounting for 51% of all web traffic." That figure comes from a security vendor, and it counts traffic: requests to servers. It does not tell you how much of what is written online was written by a machine.

Is dead internet theory a conspiracy theory?

The full version is. Wikipedia says it was originally conceived as a conspiracy theory, and the original post blamed the US government. The claims about bot traffic and AI-written text can be measured, so they stand or fall on their own evidence. The [about page](/about/) explains who runs the tests here.