Most of the internet is now bots and AI, not real people
By Colin PetersonJuly 8, 2026
Where the evidence lands →That organic human activity on the web (posts, replies, engagement, even entire personas) has been substantially displaced by bots, algorithmically generated content, and AI systems, and that in its strongest form this replacement is being deliberately orchestrated by governments or corporations to manufacture the appearance of consensus and manipulate the population.
Independently measured automated traffic really has overtaken human traffic on the open web. But that measurable trend doesn't support the theory's strong claim of a coordinated replacement of human activity.
Believed by: A minority, but a fast-growing share of heavy internet users
Fresh 2026 measurements pushed the 'bots have taken over' point further into the mainstream. Cloudflare reported that automated traffic crossed the halfway mark in June 2026, with bots generating about 57.5 percent of webpage requests, and cybersecurity firm HUMAN Security clocked traffic from AI 'agents' (software that clicks links and fills out forms) growing roughly 7,851 percent year over year, well ahead of a Cloudflare prediction that the crossover would not arrive until late 2027. As before, the measured fact, that automated and AI traffic now outweighs human traffic, is real; the strong conspiracy version, a single coordinated plan to replace people, is the part that remains unproven. source →
The theory picked up an unlikely mainstream endorsement. In October 2025, Reddit co-founder Alexis Ohanian said publicly that 'the dead internet theory is real,' telling interviewers that so much of the web is now 'botted' or 'quasi-AI, LinkedIn slop' that he had long subscribed to the idea once AI could pass the Turing test. Coming from someone who helped build a major social platform, the comments moved the phrase from fringe forums into business-press coverage. His point is narrower than the strong conspiracy version, though: he is describing an emergent flood of automated and AI content driven by incentives and cheap generation, not a single coordinated plan to replace people, which remains the unproven part of the claim. source →
The full story
An internet that started to feel empty
The Dead Internet Theory did not begin as a single claim but as a mood: a sense, shared across scattered imageboard posts in the mid-2010s, that the internet had quietly stopped feeling handmade. The people who first wrote about it were describing something specific: forums that once had regulars now filled with generic replies, threads that seemed to repeat themselves without anyone noticing, a sense of talking into a room that used to have people in it.
That mood crystallized into a named theory on January 5, 2021, when a user called “IlluminatiPirate” posted “Dead Internet Theory: Most Of The Internet Is Fake” to Agora Road's Macintosh Cafe, an esoteric-culture forum. The post, by the author's own account, synthesized earlier discussion from Wizardchan and 4chan's /x/ board into a single thesis: that “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 author described the modern internet as “entirely sterile” next to its 2007 self, cited online acquaintances who had “vanished without a trace,” and pointed to a 2018 New York magazine investigation, Max Read's “How Much of the Internet Is Fake?”, as evidence that the fakery was already well underway before AI text generation made it easy.
The post spread through YouTube explainers, Reddit threads, and eventually a wider audience after The Atlantic covered it that August under the headline “Maybe You Missed It, but the Internet ‘Died’ Five Years Ago.” From there it split, as most theories eventually do, into a modest empirical claim that turned out to be measurably true, and a much larger conspiratorial claim that has not been shown to be true at all.
The numbers the believers can actually point to
Take the strong version of this theory seriously, because unlike most entries in this encyclopedia, its believers can point to an annual, methodologically documented, independently produced measurement that keeps moving in their direction. Since 2013, the security firm Imperva (now part of Thales) has published a yearly Bad Bot Report, built from its global network's detection and mitigation of bot requests across thousands of domains. In its 2025 report, covering full-year 2024 traffic, automated programs accounted for 51% of all web traffic, the first time in the report's history that bots outnumbered humans. The 2026 report, covering full-year 2025 traffic drawn from 17.2 trillion detected bot requests, put automated traffic at 53%, with human activity down to 47% and roughly 40 percentage points of that automated share classified as “bad”, bots built to scrape, defraud, or manipulate rather than to perform a disclosed, legitimate function. That is not a rumor from an anonymous forum post; it is a year-over-year trend from a security vendor with a direct commercial interest in measuring it accurately, since its business depends on customers trusting the number.
Layer onto that the timing of generative AI. The theory was named in January 2021, roughly a year before consumer-grade large language models and image generators made machine-written text and machine-made images cheap, fast, and difficult to distinguish from human output at a glance. Researchers who have since studied the theory as an object in its own right, including a 2025 scoping survey in the Asian Journal of Research in Computer Science and a 2024 opinion piece in the journal AI & Society do not treat it as pure paranoia; they treat it as a folk hypothesis that anticipated a real shift in what the internet is made of, years before the tools that accelerated that shift were publicly available.
And the theory's darkest mechanism, that engagement itself is manufactured, is not speculative. Platforms including YouTube and Facebook have themselves disclosed removing billions of fake accounts and coordinated inauthentic-behavior networks over the past decade, and bot-farm operations selling fabricated views, likes, and followers have been documented and prosecuted. If you already knew, from a platform's own enforcement reports, that a meaningful share of what looked like organic popularity was fabricated, the leap to “maybe most of what I see here isn't real” is not an unreasonable one to make.
The theory's darkest specific mechanism, that engagement itself is being manufactured by coordinated, non-human actors, has a direct confirmation from an unexpected source: the AI companies whose own tools are allegedly doing the replacing. In May 2024, OpenAI disclosed that it had identified and shut down five separate covert influence operations that had abused its models: two linked to Russia, including the long-tracked “Doppelganger” campaign and a smaller operation the company named “Bad Grammar,” one linked to the persistent Chinese state-linked network known as “Spamouflage,” one linked to Iran, and one run by a private Israeli firm. Each had used generative AI to write social media comments and articles, invent biographies for fake accounts, translate propaganda, or debug the code running the bot infrastructure behind the operation, targeting audiences following the wars in Ukraine and Gaza and domestic politics in the United States, Europe, and India. This is not an inference drawn from aggregate traffic statistics; it is a named, dated, company-disclosed account of state-linked and commercial operations using AI specifically to manufacture the appearance of organic human opinion: the exact mechanism, and in some cases the exact actors, the theory's strongest version describes.
What the traffic numbers don't actually show
The trouble with the strong version of the Dead Internet Theory is not that its headline statistic is wrong; it is that the statistic does not mean what the theory needs it to mean. The Imperva figures measure traffic, not authorship of the content people actually read, and not human engagement. A large share of “bot traffic” is unglamorous and disclosed: search-engine crawlers indexing pages, uptime monitors pinging servers, price-comparison scrapers, and (a growing category the 2026 report tracks explicitly) AI crawlers fetching pages to answer chatbot queries. Imperva's own breakdown separates “good bots” performing these disclosed functions from “bad bots” built to scrape, defraud, or impersonate, and only the latter category resembles the sinister automation the theory describes. A server being pinged by a monitoring script a thousand times a day is not evidence that the humans who read that server's content have disappeared.
Researchers who have examined the theory directly make the same distinction. Writing on the theory in 2024, computer scientists Jake Renzella and Vlada Rozova of the University of New South Wales argued that “the dead internet theory is not really claiming that most of your personal interactions on the internet are fake” once the evidence is examined: the meaningful engagement most people actually experience, messages from known contacts, replies from real communities, still overwhelmingly involves real humans. What the data supports, they argue, is narrower and in some ways more mundane: bad actors deliberately deploying bots to farm advertising engagement, build credible-looking dormant accounts for later resale, and run disinformation campaigns: a real problem of profit-driven manipulation, not evidence that humanity itself has been replaced online.
The gap is starkest at the theory's original, most conspiratorial claim: that a government or corporation is centrally orchestrating this shift to manufacture the appearance of consensus. No bot-traffic report, leaked document, or academic study has identified any such coordinating actor. What the evidence does show is a lot of separate automation with a lot of separate, mostly banal motives (ad fraud, SEO spam, scraper bots training AI models, engagement-farming for hire), each explainable without a central conspirator. A 2025 scoping survey published in the Asian Journal of Research in Computer Science reviewing the academic literature on the theory reaches a similar conclusion: it finds real evidence of platform automation, algorithmic mediation, and AI-generated content, but explicitly frames the internet's problems as arising from engagement-driven business models rather than a deliberate plot to depopulate the web of people.
The clearest test of that distinction is the very report the strong version leans on hardest. OpenAI's May 2024 disclosure did find real, named, state-linked operations deliberately using AI to fabricate engagement, but the same disclosure is also the best evidence yet that such efforts are failing at the scale the theory requires. OpenAI's own investigators concluded that none of the five operations they caught, including the Russian “Doppelganger” and “Bad Grammar” campaigns and the Chinese “Spamouflage” network, “meaningfully increased their audience engagement or reach” by using its models. Real coordination exists, in other words, but the best-documented cases of it are ones that got caught and, by the disclosing company's own account, mostly failed to work: a pattern of contained, unsuccessful attempts rather than a demonstration that a coordinated replacement of human opinion is quietly succeeding anywhere at internet scale.
Why a real trend curdled into a conspiracy
The Dead Internet Theory is one of the rarer entries in this encyclopedia where the believers' starting instinct was empirically ahead of the skeptics'. Long before most people had a name for it, longtime internet users had noticed something real: the web they used in the mid-2000s ran on forums, personal blogs, and small communities, while the web after roughly 2015 increasingly ran on algorithmically ranked feeds optimized for engagement rather than connection. That shift was disorienting on its own, and it happened to be true, which gave the people who first named it an unusually solid foundation compared to most origin stories in this encyclopedia.
From that true foundation, the theory grew the way most conspiracy theories do: by filling the gap between “something changed and I can't fully see why” with the most emotionally coherent available explanation. Platform recommendation algorithms are genuinely opaque even to regulators and researchers who study them professionally, and that opacity is uncomfortable in a way that a single, nameable cause, “they replaced the humans on purpose,” resolves more satisfyingly than the true, messier answer, which is that thousands of uncoordinated actors, from state propaganda units to teenage ad-fraud operators to indexing bots, are each pursuing narrow goals that add up to a web that looks less human without anyone planning that outcome.
The generative-AI boom then supplied a second engine. Once cheap tools could produce plausible text and images at scale starting in 2022, the theory's believers had a fresh supply of exhibits arriving in real time (AI-written product reviews, synthetic social media personas, algorithmically generated “slop” content), each new example reads as confirmation of a theory that predates the technology that produced it, which is a uniquely persuasive pattern: the prediction came first, and the evidence appears to keep arriving afterward.
A kernel wrapped in overreach
This one splits cleanly rather than resolving in either direction. The theory's foundational claim, that automated, non-human traffic has grown to rival and now exceed human traffic on the web, is confirmed by repeated, independent, methodologically documented measurement: 51% automated in 2024, 53% in 2025, with roughly 40 points of that being traffic classified as malicious rather than benign crawling. That is a real and, on current trend lines, worsening kernel.
What remains unproven is everything the theory needs to become a genuine conspiracy rather than a description of a messy, commercially driven internet: that the humans behind meaningful content and engagement have been substantially replaced, and that any single actor is deliberately orchestrating that replacement to manufacture consent. No study, leak, or traffic report has shown either of those things. The most defensible summary is the one the researchers who have studied this directly converge on: the internet has become significantly more automated, more algorithmically mediated, and more cluttered with synthetic content than it was a decade ago, and that happened through the ordinary, uncoordinated incentives of advertising, fraud, and engagement-maximizing business models, not through a secret plan to depopulate it of people.
What's still unexplained
- How much of the content an actual person encounters in a feed, search result, or comment section (as opposed to raw server traffic) is bot- or AI-generated is still not cleanly measured. Imperva's figures track requests hitting servers, not the composition of what shows up in front of a real reader, and no independent study has published an equivalent 'share of content actually seen by humans' figure. The direction is almost certainly upward as generative tools get cheaper, but the size of the gap between measured traffic and lived experience is genuinely unresolved.
- Whether the shift toward a more automated internet is best explained as an emergent effect of thousands of uncoordinated profit motives, or contains a real element of deliberate coordination beyond what has already been caught, remains open. OpenAI's and Meta's own disclosures confirm that state-linked operations have used AI models and bot networks specifically to manufacture the appearance of grassroots opinion, which is genuine coordination; but the companies reporting on it also found that the specific operations they caught failed to gain substantial real audiences. Whether that finding reflects the true ceiling of coordinated manipulation, or just the limits of what has been detected so far, can't be settled from the public record.
- How the internet will look once a growing share of new AI-generated content is itself trained, directly or indirectly, on earlier AI-generated content is a live and unresolved research question. A 2024 study in Nature found that language models trained recursively on their own prior output degrade over successive generations, a documented effect researchers call 'model collapse.' What that feedback loop means for the practical texture of an internet increasingly built from AI output training on other AI output (whether it degrades quality in ways ordinary users will notice, or gets managed away by better data curation) is not yet known.
- Whether reliable, at-scale detection of AI-generated text and images is achievable at all, or whether generation and detection will stay locked in a permanent arms race with detection perpetually a step behind, is genuinely contested among researchers. Until that's resolved one way or the other, the basic question a growing number of internet users now ask of any given post, review, or reply (is this real?) may not have a stable answer, in either direction, for years.
Point by point
The claim: Automated bot traffic has overtaken human traffic on the web.
What the record shows: Confirmed by independent measurement. Imperva's annual Bad Bot Report, based on threat-detection data across its global network, found automated traffic at 51% of all web traffic in 2024 and 53% in 2025, with human activity at 47%, the first time in the report's history that bots have outpaced people. This is a real, repeatedly measured trend, not a rumor.
The claim: AI is now generating a large share of what looks like human-written content.
What the record shows: Directionally correct and rapidly worsening. Generative AI text and image tools became cheap and widely available after 2022, and researchers studying platforms since then document large volumes of AI-authored posts, reviews, and articles. Precise, agreed-upon percentages of 'AI-generated content' online are harder to pin down than bot-traffic share, but the trend itself is not in dispute.
The claim: Engagement metrics, recommendation feeds, and follower counts are frequently manipulated or fabricated.
What the record shows: Well documented. Reporting and platform-side enforcement actions going back to 2018 confirmed large-scale bot farms selling fake views, likes, and followers, and platforms including YouTube and Facebook have disclosed removing billions of fake accounts. This is the strongest, most concrete pillar under the theory.
The claim: This replacement of humans by bots is a deliberate, coordinated project run by a government or corporation to manipulate the public.
What the record shows: Not supported by the traffic or content data. The measured trends above describe an emergent, commercially driven mess (ad-fraud bots, scrapers, spam, engagement farms, each with its own mundane profit motive), not a single directing hand. Researchers who track the theory explicitly distinguish the well-evidenced 'a lot of the internet is automated' claim from the unevidenced 'a coordinating actor is deliberately hollowing out human presence' claim, and only the former holds up.
Timeline
- 2016–2018Discussion of the idea circulates in scattered form on imageboards including Wizardchan and 4chan's /x/ board, framed as a feeling that the internet had grown 'sterile' compared to its mid-2000s self.
- 2018New York magazine publishes Max Read's 'How Much of the Internet Is Fake? Turns Out, a Lot of It, Actually,' documenting bot-driven ad fraud and inflated engagement metrics, later cited as a key evidentiary source by the theory's originator.
- 2021-01-05A user called 'IlluminatiPirate' posts 'Dead Internet Theory: Most Of The Internet Is Fake' to Agora Road's Macintosh Cafe forum, synthesizing earlier imageboard posts into the named theory's foundational text.
- 2021-08The Atlantic publishes 'Maybe You Missed It, but the Internet ‘Died’ Five Years Ago,' carrying the theory to a mainstream audience.
- 2022–2024The public release of consumer generative-AI tools makes mass-produced AI text and images cheap and ubiquitous, and annual bot-traffic reports begin showing automated traffic approaching, then passing, half of all web traffic.
- 2024–2026The theory enters academic literature as a subject of study in its own right, with scholarly surveys and opinion pieces examining which parts of it hold up against traffic and content data.
Independently measured automated traffic really has overtaken human traffic on the open web. But that measurable trend doesn't support the theory's strong claim of a coordinated replacement of human activity.
Reviewed by Colin Peterson · Last reviewed July 27, 2026 · How we rate
Common questions
Is Dead Internet Theory true?
Independently measured automated traffic really has overtaken human traffic on the open web. But that measurable trend doesn't support the theory's strong claim of a coordinated replacement of human activity.
What is Dead Internet Theory?
The claim that the internet is now mostly fake: dominated by bots, AI-generated content, and algorithmic engagement farming rather than real people. Security researchers really have measured automated traffic overtaking human traffic; what remains unproven is the theory's stronger claim that this is a deliberate, coordinated replacement…
What does the evidence show?
Confirmed by independent measurement. Imperva's annual Bad Bot Report, based on threat-detection data across its global network, found automated traffic at 51% of all web traffic in 2024 and 53% in 2025, with human activity at 47%, the first time in the report's history that bots have outpaced people. This is a real, repeatedly measured…
Why do people believe it?
The core empirical claim is true and getting truer: independent, repeated measurements really do show automated traffic passing 50% of the web, so the theory's foundational instinct is validated by real data rather than invented from nothing.
What is still unresolved?
How much of the content an actual person encounters in a feed, search result, or comment section (as opposed to raw server traffic) is bot- or AI-generated is still not cleanly measured. Imperva's figures track requests hitting servers, not the composition of what shows up in front of a real reader, and no independent study has published…
Sources
- 1.2026 Bad Bot Report: Bots in the Agentic Age, Imperva (a Thales company) (2026)
- 2.2025 Imperva Bad Bot Report: How AI is Supercharging the Bot Threat, Imperva (a Thales company) (2025)
- 3.Disrupting deceptive uses of AI by covert influence operations, OpenAI (2024)
- 4.Dead Internet Theory: Most Of The Internet Is Fake (original post), Agora Road's Macintosh Cafe forum, user "IlluminatiPirate" (2021)
- 5.Maybe You Missed It, but the Internet 'Died' Five Years Ago, The Atlantic (Kaitlyn Tiffany) (2021)
- 6.How Much of the Internet Is Fake? Turns Out, a Lot of It, Actually., New York magazine (Max Read) (2018)
- 7.The Dead Internet Theory: A Survey on Artificial Interactions and the Future of Social Media, Asian Journal of Research in Computer Science (Muzumdar, Cheemalapati, RamiReddy, Singh, Kurian & Muley) (2025)
- 8.Artificial influencers and the dead internet theory, AI & Society, vol. 40 (Yoshija Walter) (2024)
- 9.AI models collapse when trained on recursively generated data, Nature, vol. 631 (Shumailov, Shumaylov, Zhao, Papernot, Anderson & Gal) (2024)
Embed this case file on your site
Paste this snippet to show this case file as a small card, with a link back to the full write-up. Free to use.
<iframe src="https://theconspiratory.com/embed/dead-internet-theory" title="The Conspiratory verdict" width="520" height="190" style="border:0;max-width:100%" loading="lazy"></iframe>Help us investigate
This is a living case file. If you spot an error or know evidence we missed, tell us, and weigh in on where you land.
Where do you land?
Cast your read on this one.
Comments
Add your take. Comments are read and approved by a human before they appear, so keep it on topic and civil. Please do not accuse named, living people of crimes.