The Dead Internet Theory in the AI Era: Are We Building an Internet for Humans or Machines?

The internet was originally built around a simple idea:

People create things.

Other people discover them.

People talk, argue, learn, share, build communities, and create more.

But in the AI era, that cycle is beginning to change.

Today, content can be generated by AI.

Bots can publish it.

Algorithms can rank it.

AI crawlers can consume it.

AI agents can interact with websites.

Another AI can summarize everything for a human who may never visit the original source.

That raises an uncomfortable but increasingly relevant question:

What happens when machines become both the creators and consumers of a significant part of the internet?

This is where the idea known as the Dead Internet Theory becomes interesting.

Not because we should accept the theory as fact.

But because some of the questions behind it are becoming harder to ignore.

What Is the Dead Internet Theory?

The Dead Internet Theory is an internet theory that became widely discussed in the early 2020s.

Its broad claim is that a large portion of online activity is no longer created by real humans.

Instead, it suggests that the web is increasingly filled with:

  • bots
  • automated accounts
  • algorithmically generated content
  • manufactured engagement
  • spam
  • coordinated manipulation
  • artificial conversations

Some versions go much further and claim that the majority of the internet has effectively become artificial.

There is no strong evidence supporting the most extreme versions of that claim.

The internet is obviously still filled with real people creating, communicating, buying, learning, working, playing and building communities.

But something important has changed.

Automation now represents an enormous amount of internet traffic.

And generative AI is accelerating that shift.

The Internet Is Already Heavily Automated

The idea that most internet requests come directly from humans is increasingly inaccurate.

According to Imperva's 2026 Bad Bot Report, automated traffic represented more than 53% of global web traffic during 2025, compared with 47% attributed to human activity.

A year earlier, automated activity was approximately 51%.

This does not mean that 53% of everything you read online was created by bots.

Traffic and content are different measurements.

A crawler requesting millions of pages can generate enormous traffic without publishing anything.

But the statistic still tells us something important:

Machines are becoming first-class participants on the web.

Search engines have always used crawlers.

Monitoring systems have always used automation.

Companies have always operated APIs and scheduled scripts.

What's different now is the scale and capability of automated systems.

AI agents don't necessarily just crawl a page.

They may:

  • retrieve information
  • compare products
  • summarize documents
  • interact with applications
  • fill forms
  • call APIs
  • execute workflows
  • make decisions on behalf of users

Imperva describes AI agents as an emerging category of internet participant capable of retrieving data and performing actions rather than merely scanning websites.

That is a significant shift.

AI Crawlers Are Growing Quickly

Cloudflare provides another useful view into this transition.

Its 2025 Internet traffic analysis found that AI-related crawlers had become a meaningful portion of web requests.

Cloudflare reported that AI bots generated an average of approximately 4.2% of HTML requests during 2025, excluding Google's dual-purpose Googlebot.

Googlebot itself represented approximately another 4.5% of HTML request traffic.

Among verified bots, AI crawlers represented about 20% of traffic.

More interestingly, Cloudflare observed rapid growth in what it calls AI user-action crawling.

These are requests where an AI system visits websites because a user asked it to retrieve or work with information.

That category increased by more than 15× during 2025, according to Cloudflare's year-in-review analysis.

This may give us a preview of how people will increasingly interact with the web.

Instead of:

Human
↓
Search engine
↓
Website
↓
Read page

we may increasingly see:

Human
↓
AI assistant
↓
Websites / APIs
↓
AI processes information
↓
Human receives answer

The human still exists.

But the AI becomes the interface to the internet.

The Internet May Become Machine-to-Machine

This is where the Dead Internet discussion becomes more interesting.

Imagine the following cycle.

Human idea
    ↓
AI generates an article
    ↓
Automation publishes it
    ↓
Search / recommendation algorithms index it
    ↓
AI crawler reads it
    ↓
AI assistant summarizes it
    ↓
Another AI generates a post from that summary

Eventually we could reach a situation where:

AI-generated content is being consumed primarily by other AI systems.

The original human becomes increasingly distant from the information lifecycle.

This doesn't necessarily make the internet "dead."

But it makes it different.

The internet may evolve from being primarily:

humans communicating through computers

toward:

humans delegating communication and information processing to machines.

AI-Generated Content Changes the Economics of Publishing

Historically, creating content had a meaningful cost.

Someone needed to:

research,

write,

edit,

design,

record,

photograph,

or produce it.

Generative AI dramatically reduces that cost.

One person can now generate dozens or hundreds of pieces of content far faster than before.

At the same time, automated systems can distribute that content across many platforms.

That creates an obvious incentive:

More content
     ↓
More impressions
     ↓
More search visibility
     ↓
More advertising / leads / influence

When the cost of creating another page approaches zero, the internet can become flooded with material produced primarily to capture attention rather than provide value.

This doesn't mean AI-generated content is automatically low quality.

AI can help knowledgeable people create excellent work.

The important distinction is between:

AI-assisted creation

and

mass automated content generation with little human judgment.

The Web's Creator–Reader Relationship Is Also Changing

There is another important shift.

Traditionally:

Creator publishes content
        ↓
Search engine indexes it
        ↓
User clicks search result
        ↓
Creator receives visitor

AI search and assistants can create a different flow:

Creator publishes content
        ↓
AI crawler consumes it
        ↓
AI generates an answer
        ↓
User receives answer

The user may never visit the creator's website.

Cloudflare has described this as a growing crawl-to-referral gap.

Its 2025 analysis found that AI crawlers often retrieve far more pages than the number of visitors their platforms send back to publishers.

This creates a difficult economic question:

If AI systems consume the web's content but increasingly answer users without sending them back to the original websites...

what incentive remains for people to keep producing high-quality public information?

That may become one of the most important questions for the future of the open web.

From Search Engines to Answer Engines

For roughly two decades, our relationship with online information looked like:

Question
↓
Google
↓
10 links
↓
Visit websites
↓
Build your own answer

AI assistants increasingly change that to:

Question
↓
AI
↓
Answer

This is extremely convenient.

But convenience has consequences.

When people stop visiting sources directly:

  • publishers lose traffic
  • smaller websites lose visibility
  • independent creators lose discovery
  • users may not know where information originated
  • mistakes can become harder to identify
  • fewer people encounter viewpoints outside the generated answer

The web becomes less about finding information and more about receiving synthesized information.

AI Talking to AI

Agentic AI adds another layer.

Consider a future shopping interaction.

Instead of visiting five websites yourself:

You
↓
AI shopping agent
↓
Product APIs
↓
Merchant agents
↓
Recommendation engine
↓
Purchase

Or travel:

You
↓
Travel agent
↓
Airline systems
Hotel systems
Payment systems
Calendar
↓
Trip booked

Much of the internet interaction happens without a human directly seeing the websites involved.

Machines negotiate with machines.

APIs communicate with agents.

Software performs actions on behalf of people.

That internet isn't necessarily worse.

In many cases it could be dramatically more useful.

But it is structurally different from the web most of us grew up using.

The Identity Problem

Another major challenge is determining:

Who — or what — is on the other side of an interaction?

Today a website may receive traffic from:

  • a human visitor
  • search crawler
  • malicious bot
  • monitoring system
  • AI training crawler
  • AI search crawler
  • legitimate AI assistant
  • autonomous agent working for a user

The old model was approximately:

Human
or
Bot

The future may require something closer to:

Human
Human using AI
AI acting for human
Verified service bot
Search crawler
AI crawler
Commercial agent
Malicious automation
Unknown automation

This creates entirely new challenges for authentication, authorization and security.

Developers Will Need to Design for Agents

This matters directly to software engineers.

Historically, we designed applications primarily for humans.

We built:

  • graphical interfaces
  • buttons
  • forms
  • navigation
  • accessibility
  • responsive layouts

In an agent-driven internet, applications may increasingly have two interfaces.

Human interface

UI
Mobile App
Website

Machine interface

API
Structured data
Agent protocol
Tools
Permissions

The software may need to serve both simultaneously.

A human sees:

Book appointment

An AI agent sees something more like:

createAppointment(
    service,
    date,
    location,
    customer
)

That means API architecture, permissions and machine-readable capabilities may become even more important.

Trust Becomes More Valuable

When generating content becomes extremely cheap, human trust may become more valuable.

We may start caring much more about:

  • who created something
  • whether a person reviewed it
  • where information originated
  • whether sources are authentic
  • whether an account represents a human
  • whether a photo or video is synthetic
  • whether an agent is authorized

In other words:

As artificial content becomes abundant,

verified human judgment may become scarce.

Scarcity usually creates value.

This Doesn't Mean AI Is the Problem

It's easy to turn this conversation into:

AI is destroying the internet.

I don't think that's a useful conclusion.

AI also makes the internet enormously more accessible and useful.

It can:

  • translate information
  • summarize complex material
  • help people with disabilities
  • automate repetitive work
  • make programming more accessible
  • help people discover knowledge
  • assist researchers
  • improve productivity
  • create entirely new software experiences

The challenge isn't whether we should use AI.

We clearly will.

The real question is:

How do we preserve the qualities that made the open internet valuable while gaining the benefits of AI?

The Dead Internet Theory Is Probably the Wrong Name

Personally, I think "Dead Internet Theory" is too dramatic.

The internet isn't dead.

It may simply be entering another phase.

We had:

Web 1.0
Read

Web 2.0
Read + Write

Mobile Internet
Always connected

AI Internet
Read + Write + Generate + Act

The next internet may involve humans giving increasingly high-level instructions while software handles the underlying navigation and interaction.

Instead of:

Open this website.

We may say:

Find the best option and handle it for me.

Instead of:

Search for five articles.

We may say:

Research this topic and show me what matters.

Instead of:

Open my bank app and pay this bill.

We may say:

Pay all my approved bills before Friday.

That is a profound interface shift.

So Is the Internet Becoming "Dead"?

Not exactly.

There is evidence that automated traffic now represents a very large share of web activity.

There is evidence that AI crawling is growing.

There is evidence that AI assistants increasingly retrieve information on behalf of users.

And generative AI is undeniably making automated content creation easier.

But none of that proves that human participation has disappeared.

A better description may be:

The internet is becoming increasingly machine-mediated.

Humans are still behind the goals.

Humans still create many of the ideas.

Humans still make many of the important decisions.

But machines increasingly sit between humans and the internet.

The Question I Keep Coming Back To

Imagine this future:

AI writes content
↓
AI publishes content
↓
AI indexes content
↓
AI reads content
↓
AI summarizes content
↓
AI recommends content
↓
AI comments on content

And somewhere at the beginning and end...

there's a human.

At what point does the internet stop being primarily a place where humans communicate with each other?

And become infrastructure where machines communicate on our behalf?

Maybe the Dead Internet Theory isn't really about whether the internet is dead.

Maybe the more interesting question is:

Are we building the next internet for humans — or for agents?

Final Thoughts

I don't think we should fear an internet filled with AI agents.

But I do think developers, creators, publishers and platforms need to think carefully about what we're building.

We may need:

  • stronger identity systems
  • better attribution
  • machine-readable permissions
  • transparent AI labeling
  • better bot and agent detection
  • fairer relationships between AI platforms and content creators
  • systems that preserve original sources
  • tools that distinguish trustworthy automation from malicious automation

The internet has changed many times before.

AI may simply be the next major transformation.

The question isn't whether machines will become a bigger part of the internet.

They already are.

The question is:

What role will humans have when they do?

Sources & Further Reading

The figures and trends discussed above come from current industry research rather than from the Dead Internet Theory itself.

  • Imperva — Bad Bot Report 2026: Reports that automated traffic accounted for more than 53% of global web traffic in 2025, while human traffic represented about 47%.
  • Cloudflare — 2025 Internet Year in Review: Tracks verified bot traffic, AI crawler growth, and the rapid increase in AI user-action crawling.
  • Cloudflare — The Crawl-to-Click Gap: Examines how AI systems can crawl large amounts of publisher content while sending comparatively few users back to the original websites.
  • Cloudflare — Who's Crawling Your Site in 2025?: Provides additional data on the growth of AI and search crawler traffic.

The Dead Internet Theory itself should be treated as a theory, not as proof that most online content or interactions are fake.