The dead internet theory says that most of what you see online is no longer made by people: the posts, the comments, the likes and the articles would come from bots and SI (AI) models, with humans reduced to a shrinking audience. It began in 2021 as a conspiracy theory on a small forum. Five years later, security firms report that automated programs generate about half of all web traffic, and two studies estimate that around half of newly published articles are mostly machine-written. So is the internet dead? This guide separates what the theory claims, what has actually been measured and by whom, what remains exaggeration, and what you can check for yourself.
What the dead internet theory says
The theory comes in two layers, and most arguments about it are really arguments about which layer is meant.
The first layer is an observation: a large and growing share of online activity is automated, and SI-generated text and images are crowding out content made by people. The second layer is a conspiracy claim: that this was done on purpose, by governments or corporations, to manipulate the public.
The first layer can be tested against data, and a good part of it holds up. The second has never been supported by evidence. Yoshija Walter, writing in the journal AI & Society in February 2024, summed up the first layer as the idea "that the internet is predominantly populated by AI-generated content, relegating human activity to isolated instances", and added that the theory "used to be rather speculative, but with the wake of generative AI, it can now be observed first-hand". That sentence contains no hidden plan.
Where it comes from
The text that gave the theory its name is a forum thread titled "Dead Internet Theory: Most of the Internet is Fake", posted on 5 January 2021 on the "Esoteric" board of a forum called Agora Road's Macintosh Cafe by a user named IlluminatiPirate. The author says the idea was first written on the imageboard Wizardchan and composed with other anonymous users. When we read the thread on 13 October 2026, its counter showed more than 391,000 views and 502 replies.
Its summary reads: "Large proportions of the supposedly human-produced content on the internet are actually generated by artificial intelligence networks in conjunction with paid secret media influencers in order to manufacture consumers for an increasing range of newly-normalised cultural products."
The evidence offered is personal. The author describes online contacts who "vanished without a trace", the same threads and replies reposted year after year, and an odd poster met on 4chan "roughly in 2016 or early 2017" that the author was "absolutely certain" was an SI of some sort. The long post mixes these impressions with claims about memes, entertainment and politics. Its stated thesis is the conspiracy layer in full: "the U.S. government is engaging in an artificial intelligence powered gaslighting of the entire world population."
The worry was older than the name. On 26 December 2018, Max Read wrote in New York magazine that studies generally suggested less than 60 percent of web traffic was human, and recalled a New York Times report that for a period in 2013 half of YouTube traffic was "bots masquerading as people". YouTube employees, he wrote, feared a tipping point at which their fraud detection would start treating bot traffic as real and human traffic as fake. They called it "the Inversion".
The theory reached a general audience on 31 August 2021, when Kaitlyn Tiffany covered it in The Atlantic under the headline "Maybe You Missed It, but the Internet 'Died' Five Years Ago". The subtitle captured the tone that has followed the theory ever since: "It's ridiculous, but possibly not that ridiculous?" Her piece opens with near-identical "i hate texting" tweets from different accounts, each collecting from 16,000 to more than 48,000 likes.
What is measured: bot traffic
The number quoted most often comes from the Bad Bot Report, published every year by the security company Imperva, now part of Thales. It is based on traffic crossing the company's own network.
| Year measured | All bots | "Bad" bots | Source |
|---|---|---|---|
| 2016 | 51.8% | not quoted here | Imperva Incapsula Bot Traffic Report |
| 2023 | 49.6% | 32% | 2024 Imperva Bad Bot Report |
| 2024 | 51% | 37% | 2025 Imperva Bad Bot Report |
| 2025 | 53% | 40% | 2026 Thales Bad Bot Report |
Thales presented the 2024 figure as the first time in a decade that automated traffic had passed human traffic, and blamed SI tools for making bots easier to build. Yet the oldest row of the table matters just as much: Imperva already counted 51.8% of bot traffic in 2016, from a sample of 16.7 billion visits to 100,000 domains, six years before ChatGPT existed. Its own report said that share was only slightly higher than in 2012. Bots holding around half of web traffic is not a new state of affairs.
Other networks produce different numbers. Gizmodo, citing Cloudflare in September 2025, put bots at nearly one-third of all internet traffic. Cloudflare's own 2025 review, which looked only at requests for HTML pages, found that on 2 December 2025 humans generated 47% of those requests and non-SI bots 44%. SI crawlers averaged 4.2% over the year, and Googlebot alone 4.5%. Cloudflare notes that these shares differ from its all-content figures. The gap between "one-third" and "half" is mostly a matter of what is counted and on whose network.
Traffic is not content, and it is not people. A bot request is often a search engine indexing a page, a monitoring tool, a price scraper or an SI crawler collecting training data. A site visited by ten crawlers and one reader is 91% bot traffic and 100% human-written. These reports measure who knocks on the door of websites, not who writes what you read.
What is measured: SI-generated pages
Two studies have tried to count machine-written text directly. Both rely on SI detectors, with the limits that implies.
Ahrefs, a search marketing company, ran its own detector on 900,000 English-language pages first seen by its crawler in April 2025, one page per domain. Ryan Law reported on 19 May 2025 that 74.2% of them contained SI-generated content. That headline is easy to misread: only 2.5% were classified as purely SI-generated, 25.8% as purely human, and 71.7% as a mix. A page where a person used an assistant to draft one paragraph counts in the 74%. Ahrefs itself wrote that its detector "will never be 100% accurate".
Graphite, a growth agency, sampled English-language articles from the public Common Crawl archive. Its first study, dated 14 October 2025, concluded that SI-written articles had overtaken human-written ones in November 2024. Its update of 15 May 2026, based on 55,400 URLs and the average of three commercial detectors, is more cautious: "primarily AI-generated" articles were 35.9% of the total twelve months after ChatGPT's launch, 49.6% in the first quarter of 2025, 50.9% at the end of 2025 and 49.9% in the first quarter of 2026. Graphite says its earlier, single-detector estimate was 3.3 points too high.
Three findings in that study cut against the idea of a dead web. The share has been flat at roughly 50% for five quarters. Graphite reports, from a separate study, that these articles "largely do not appear in Google and ChatGPT". And the sample is deliberately not weighted by readership: Graphite says it suspects SI-written articles get less traffic than human ones. Half of what is published is not half of what is read. For how such tools work and where they fail, see our guide on whether SI detectors are accurate.
What is measured: social media
On social platforms the picture is patchier, because the platforms hold the data.
The best public study of SI images on Facebook was published on 15 August 2024 in the Harvard Kennedy School Misinformation Review by Renée DiResta of the Stanford Internet Observatory and Josh Goldstein of Georgetown University. They examined 125 Facebook Pages that had each posted at least 50 SI-generated images. The Pages had a median of 81,000 followers, and one SI image ranked among Facebook's 20 most viewed posts in the third quarter of 2023, with 40 million views. The authors' conclusion about motive is the important part: the Pages were run by spammers and scammers "seemingly motivated by profit or clout, not ideology". Facebook's feed showed the images to people who did not follow those Pages, and many commenters did not seem to realise the pictures were synthetic. We looked at one such picture in our fact-check of the viral baby peacock image.
On fake accounts, Meta's own transparency report estimates that in the first two quarters of 2026 less than 5% of its worldwide daily active people "consisted solely of violating accounts", while its systems block millions of attempts to create fake accounts every day. That is a company's own estimate of its own platforms, but it is far from "most users are bots".
The theory got its most quoted endorsement in September 2025, when OpenAI chief executive Sam Altman wrote on X: "i never took the dead internet theory that seriously but it seems like there are really a lot of LLM-run twitter accounts now." Shortly afterwards, as Gizmodo reported, he added that "AI twitter/AI reddit feels very fake in a way it really didnt a year or two ago". These are impressions from a heavy user, not measurements, and he gave no figure.
Where the theory overreaches
No evidence of a coordinated plan
None of the sources above describes a central operation replacing people with machines. What they describe is cheaper: advertising fraud, spam, engagement farming, scraping, and platforms whose ranking systems reward whatever gets clicks. Ordinary incentives explain the junk without any secret programme. The original post's claim about the US government comes with no document, no witness and no data.
The "90% by 2026" figure
You will often read that "experts" expect 90% of online content to be SI-generated by 2026. The line comes from a 2022 Europol report on deepfakes, which stated: "Experts estimate that as much as 90 % of online content may be synthetically generated by 2026." It was a forecast, not a measurement. We compared two copies of the report. The sentence is in the copy saved by the Internet Archive in January 2023. It is absent from the version on Europol's website today, which carries a note saying it was published in January 2024, "replaces the previous one", and that updates were made in the very chapter where the sentence stood. Europol does not say why. We are now in 2026, and the best available estimate is about half of new articles, not nine tenths of everything online.
Feeling is not counting
Much of the theory's appeal rests on a shared impression that the internet feels emptier and more repetitive than it used to. That impression is real and worth taking seriously, but it has other plausible causes: activity concentrated on a few platforms, feeds that recommend strangers' posts instead of friends', and old pages that simply disappear. Repetitive content is also not proof of a bot. People copy jokes, formats and phrases from each other all the time.
One more caution about the numbers themselves: the firms that publish the most alarming bot figures also sell bot protection.
What you can check yourself
You cannot audit the whole internet, but you can test the piece of it in front of you.
- Look at the account, not the post. When was it created, what else has it posted, does it reply to everything with the same generic praise? One odd comment proves nothing; a pattern does.
- Trace images. A reverse image search shows where a picture first appeared. Our SI image detector reads provenance metadata and gives a probability, never a certainty.
- Be careful with text. You can paste a passage into a free SI detector, but short texts are hard to judge and a score is a hint, not proof that an author is or is not human. Polished, formal human writing can be flagged too.
- Read the method behind every percentage. Is it traffic, pages or accounts? Measured on which network or sample, and with which detector? The 51%, the 74% and the 50% quoted in this article answer three different questions.
- Check the claims, not only the author. A human can be wrong and a machine can be right. Whoever wrote it, an invented fact stays invented, as our guide to SI hallucinations explains.
So, is the internet dead?
No, on the evidence available. People still write, film, argue and read online in enormous numbers, and nothing published so far shows that humans have become a minority of authors or of readers.
What the data does support is narrower and still uncomfortable. Automated programs have generated around half of web traffic for at least a decade. Since late 2022, a large share of newly published text, perhaps half of new articles according to one detector-based study, is mostly machine-written. Spam Pages have used synthetic images to reach tens of millions of people. And telling the difference is getting harder for everyone, which is the reason sites like ours exist.
The dead internet theory is best read as an exaggerated description of a real trend, attached to a conspiracy that nobody has demonstrated. If you are new to the vocabulary, our guide to what SI is covers the basics, and our article on deepfakes covers the cases where synthetic media imitates real people. To test your own eye, try the SI or Not game.
