The Internet Feels Off: Here Is The Uncomfortable Version Of Why

Google just told a U.S. court the quiet part out loud: the open web is already in rapid decline. The company later narrowed what it meant, focusing on open-web display advertising, but the thrust stands. If you work anywhere near publishing or search, it tracks with lived reality: less referral traffic, more AI summaries, more sludge.

This is a straight look at the Dead Internet Theory and the world around it. I will separate what we can verify from what is speculation. Where I cannot verify, I will label it.

• Definition and history, without the creepypasta.
• The core claims from supporters, checked against current evidence.
• Manufactured consensus through modern influence operations, including Russian-linked campaigns in Germany, Poland, Romania, plus what this means with Moldova voting this month.
• The old internet, where this all started: research networks, collaboration first, revenue later.

If you want a tidy ending, you will not get one. You will get a map.

Table of Contents

• What People Mean By “Dead Internet Theory”
• What Google Just Admitted, And Why It Matters
• The Main Claims, Checked
  • Claim 1: Bots Are Overtaking Humans
  • Claim 2: Algorithms Bury Human Work And Reward Lowest-Effort Slop
  • Claim 3: Manufactured Consensus Is The Norm
  • Claim 4: The Open Web Is Collapsing Into Closed Gardens And AI Front-Ends
  • Claim 5: The Past Is Literally Disappearing
• Manufactured Consensus With Names And Dates
• The Old Internet Was A Different Beast
• So, Is The Internet Dead
• Practical Responses That Do Not Depend On Vibes
• A Note On Moldova
• The Hard Part

What People Mean By “Dead Internet Theory”

At its simplest, Dead Internet Theory claims the web you see is mostly not human anymore. Bots generate content, algorithms surface it, and coordinated actors steer conversation. The phrase gathered steam after a 2021 forum post by an anonymous user, then jumped into mainstream coverage. The summary most people carry around is this: it is a conspiracy theory that feels plausible because some of its inputs are true.

What is verifiably true in that bundle:

• Bot traffic is large and rising.
• AI-generated content is flooding indices.
• Platforms gate attention and optimize for ad economics.
• Foreign and domestic information operations exist and adapt.

What is not verified:

• [Unverified] That the majority of online content or engagement you personally encounter is fake.
• [Unverified] That a single, centrally coordinated campaign is controlling the internet at scale.

Hold both ideas at once. There is a lot of artificial activity. The totalizing version of the theory goes beyond what public evidence supports.

What Google Just Admitted, And Why It Matters

In a legal filing tied to the Justice Department’s ad-tech case, Google said the open web is already in rapid decline. Later, it clarified that the point was about open-web display advertising. The important part is not the PR. It is the admission under legal scrutiny that the economic engine for the open web is shrinking.

Pair that with independent data on how people use search when AI summaries appear. When an AI summary sits above the links, people click sources less. That compounds the traffic problem for the sites that still do the reporting or original work.

No, that does not prove the web is dead. It does show incentive structures tilting away from the messy, open, many-to-many web and toward closed, summarized, and platform-bound behavior.

The Main Claims, Checked

Claim 1: Bots Are Overtaking Humans

What supporters say: most traffic and a scary chunk of conversation are machines.

What the data supports: multiple annual reports now put automated traffic at a very large slice of the pipe, with malicious bots taking up a meaningful share. Numbers vary by sector and methodology, but the direction is not in dispute. If you run a site, your logs show it every day: scraping, credential-stuffing, inventory hoovers, price-scrapers, view-fraud. We live inside a machine ecology and it is growing.

What that means in practice: bandwidth share is not persuasion share. A crawler can inflate traffic, yet a few human posts can still set the tone. Bot tonnage does not automatically mean human culture has vanished.

Where speculation creeps in:

• [Unverified] “Most content you see day to day is bot-made.” We do not have platform-wide transparency to quantify that for a typical user.

Claim 2: Algorithms Bury Human Work And Reward Lowest-Effort Slop

What supporters say: ranking systems favor engagement farming over originality. Humans adapt their behavior to look like bots.

What we can verify: search is flooded with content that looks like it was created for search engines instead of people. The same period brought a clear strain on open-web economics. Separate studies show AI summaries reduce downstream clicks. Publishers are not imagining it.

Limits: platforms do not publish the weights and dials. Measuring suppression globally is not possible from outside. What we can say is simple: if AI answers satisfy the user on-platform and users click less, open sites will see less traffic. That is economics, not a plot twist.

Claim 3: Manufactured Consensus Is The Norm

What supporters say: a significant share of visible consensus is staged by state actors, PACs, agencies, or spam networks running fleets of accounts.

What is documented:

• Germany in 2025 saw Russia-linked networks seeding fake alerts and political narratives through pseudo-media outlets and social videos.
• Poland’s security services warned about Russian attempts to recruit locals through darknet channels to influence the election cycle.
• Romania reported waves of Russian-linked online disinformation across the 2024 to 2025 window, with authorities naming specific themes and tactics.
• The United States has a public record on 2016 that is beyond debate: sweeping operations by Russian entities to hack, dump, and flood feeds. Peer-reviewed research later found limited measurable persuasion effects in some panels, which is the nuance many miss. The operation is documented. Its precise effect size on vote choice is contested.

Then there is Moldova right now: the government has warned of a hybrid campaign ahead of the parliamentary vote, including disinformation, illicit funding, and vote-buying allegations. European institutions are paying attention. This is not theoretical. It is imminent.

Bottom line: manufactured consensus exists. The smart position is to demand proof for scope and specific outcomes. Do not confuse verified presence with quantified impact. The first is well documented. The second is contested and frequently overstated.

Claim 4: The Open Web Is Collapsing Into Closed Gardens And AI Front-Ends

What supporters say: the commons is being paved over by platform feedlots and chatbot answers.

What we can point to: legal statements about rapid decline in the open-web ad market, plus user behavior changes when AI summaries appear. If fewer people leave the results page, fewer people reach source sites. That is a structural shift, not a mood.

Caveat: dead is a big word. The open web can decline unevenly. Some niches thrive. Many wither.

Claim 5: The Past Is Literally Disappearing

Large portions of pages from the last decade are already gone. Link rot is not a metaphor. It is a statistical fact across news, government pages, and even Supreme Court opinions and academic journals. When the historical record dissolves at the URL layer, everything that depends on it becomes harder to verify. If human work decays and AI exhaust floods the index, the feedback loop starts to look like cultural amnesia.

Manufactured Consensus With Names And Dates

Specifics matter more than vibes.

• Germany, 2025: authorities flagged Russia-linked botnets and gray-propaganda outlets, including fake security warnings and targeted political content designed to inflame.
• Poland, 2025: warnings about attempted recruitment for influence activity around the presidential race. This tracks with earlier sabotage concerns and cross-border information operations.
• Romania, 2024 to 2025: a labeled wave of online disinformation and hybrid tactics, debated in parliament and covered by international outlets.
• United States, 2016 and after: an extensive public record showing operations that hacked, dumped, and astroturfed. Exposure was large. Studies on persuasion effects are mixed. Both truths can coexist.
• Moldova, this month: the government describes an active hybrid campaign. Deepfakes and fabricated content are part of the menu. The election is a live test of whether the region learned anything since 2016.

Conclusion on this pillar: manufactured consensus is not a theory. It is an operating model. Treat confident claims about exact electoral impact with skepticism unless they bring rigorous evidence.

The Old Internet Was A Different Beast

The network did not begin as an advertising machine. ARPANET linked research institutions. NSFNET extended an academic backbone. Tim Berners-Lee’s first proposal at CERN was about sharing documents among scientists. The web went public in 1991. Only later did the backbone commercialize and the attention economy take over. Incentives changed. So did behavior.

That early culture prioritized collaboration, openness, and persistence because the users were the maintainers. Strong norms around citations, mirrors, and community curation grew out of necessity. Today’s dominant culture prioritizes engagement metrics and growth because the users are the product. You are feeling the difference.

None of this is a call to nostalgia. The early web had gatekeeping, privilege, and glacial adoption outside universities. It did have one thing we are losing fast: a bias toward durable knowledge rather than ephemeral engagement. That matters when you are trying to build anything that lasts longer than a trend cycle.

So, Is The Internet Dead

Short answer: no. Blunt answer: large parts of it are on life support for the things that made it valuable.

Here is the honest split.

Verified, measurable problems:

• Automation at scale: bot share of traffic is high and rising. Scrapers, credential testers, view-bots, click-rings.
• Attention capture by platforms: AI summaries reduce outbound clicks. Open-web ad markets are shrinking.
• Information operations are real: Russian-linked campaigns are active in Europe and well documented in the United States.
• The historical record is decaying: link rot is pervasive across institutions, journalism, and academia.

Where the theory overreaches:

• [Unverified] The claim that most of what you see is fake.
• [Unverified] A single coordinated actor running the whole show.

What to watch next:

• How aggressively platforms keep users on-site with AI answers, and whether regulators force meaningful transparency on ranking and training data.
• Whether publishers pivot to direct relationships that do not depend on search referrals.
• Whether democracies build real counter-influence capacity without turning it into censorship theater.
• Whether we rebuild persistence: archiving, canonical links, stable citations, and explicit policy around AI training on human work.

Practical Responses That Do Not Depend On Vibes

You want concrete steps. Here are the ones that matter.

1. Re-bias toward sources. If you run content, push readers to the canonical home. If you read, click through from summaries to sources. Every click out of a summary box is a vote for the original.
2. Use persistence tools. If you publish, add stable citations and mirrors. If you cite, create durable links. If you maintain, snapshot what you own. The numbers on link rot should embarrass anyone who cares about knowledge.
3. Detect and limit automation that matters. Do not waste time on “is this commenter a bot.” Focus on high-impact automation: scraping, form abuse, credential testing, inventory fraud. Sector reports give a good starting baseline.
4. Demand disclosures. AI answers should display sources with first-class links. Ranking systems that materially hit public knowledge should have public documentation. If a platform insists clicks are stable, ask for the dataset and the method.
5. Treat influence claims like scientific claims. Require methods. Who measured what, over what time window, with what counterfactual. This cuts hype and forces better defensive work. The U.S. research record on 2016 shows both a real operation and mixed evidence on persuasion effects. Both truths are useful.

A Note On Moldova

Do not sleep on this election. Moldova is small, but the pattern is large: hybrid tactics, money, propaganda, and platform dynamics that reward emotional content. If you want to test whether Europe can handle an election in 2025 without being yanked around by outside actors and domestic opportunists, start here. The warnings are not hindsight. They are now.

The Hard Part

The early web was a commons born from research networks. The modern web is a product that optimizes for engagement and margin. That shift is not evil by definition, but it reliably produces sludge, starves originals, and empowers operations that thrive on speed and confusion. Pair that with automation that now consumes a big slice of the pipes, and you get the sensation so many people describe: a web that feels empty, repetitive, or oddly inhuman. The feeling is not proof. It is smoke from real fires.

What you can say, precisely:

• The infrastructure and incentives now favor closed answers over open exploration. Verified.
• Automated traffic is not an edge case. Verified.
• Foreign and domestic actors can and do push narratives at scale. Verified. Quantified effect varies by case.
• The archive is rotting. Verified.

What you should not say without proof:

• [Unverified] “Most of this is fake.”
• [Unverified] “The internet died in 2016.”
• [Unverified] “A single actor runs the show.”

The right posture is boring and effective: instrument the problem, publish the methods, preserve the record, demand disclosures, and fix incentives so original work is not charity. If that sounds less fun than declaring the internet dead, good. Fun is how we got here.