The Dead Internet Theory in the age of AI
- Yoshi Soornack
- 5 days ago
- 3 min read
There’s a growing sense that something’s changed online.
In the last year more so, when I've scrolled through comment threads, read news summaries, skimmed posts on X, they feel familiar yet hollow. Engagement metrics climb while meaning thins. Something’s off.
What once sounded like a conspiracy in 2014 now reads like a system alert: the Dead Internet Theory. It suggests that most parts of the internet are no longer generated by humans at all, but by algorithms, bots, and AI models churning out content at industrial scale.
At first glance, that feels dramatic. But in light of recent developments, it’s getting harder to dismiss.
Where the Internet Ends and AI Begins
Recent reports suggest that up to 50% of all internet traffic is now bot-generated (Imperva, via Courier Mail). From fake engagement farms to AI-generated Medium articles to Reddit comment threads where no one is sure who’s real, we’re seeing a quiet transformation of the web’s social fabric.
Reddit CEO Steve Huffman even acknowledged this shift, calling Reddit’s “humanness” a competitive advantage under threat (Business Insider). Some researchers have gone further, with academic papers now analysing the AI-to-human ratio in social platforms (arXiv), noting the difficulty in discerning meaning when intent is absent.
The Financial Times took the critique deeper, applying philosopher Harry Frankfurt’s concept of bullshit to AI-generated language—outputs that sound convincing but are “indifferent to truth” (FT: Generative AI models are skilled in the art of bullshit).
This is what AI often produces: not lies, not truth, just patterns. Syntax without soul. Speech that lacks the grounding of human context, intention, or need.
Why This Matters for Project Delivery
This isn’t just about internet culture. In project delivery, the same pattern is playing out—only more subtly.
We now have LLMs reading our reports, summarising meetings, and generating documentation. Soon, they’ll be assessing compliance, drafting bids, or flagging early risks. But here’s the catch: they don’t understand what we mean.
AI doesn’t comprehend intention. It doesn’t reason from first principles. It simply predicts what words are likely to come next, based on training data.
If we want trustworthy, usable output from AI, we have to change how we write.
Writing for Both Humans and Machines
When your project brief becomes a source file for a language model, you’re no longer just reporting progress. You’re training your assistant. So the way you write, its clarity, structure, and semantic precision - directly impacts what AI gives back.
A few principles are starting to emerge:
Structure matters. Clear sectioning, consistent formats, and concise language help LLMs map and reason better.
Precision over ambiguity. Vague statements can mislead models, causing hallucinations or irrelevant recommendations.
Context is everything. Tie references explicitly to timelines, documents, or decisions. AI can’t infer them otherwise.
Metadata helps. Titles, roles, version dates—these help AI systems ground what’s being said and by whom.
This is less about over-engineering documents and more about writing with dual-readers in mind: humans and machines.
The Risk Isn’t Just Miscommunication. It’s Dehumanisation
The deeper risk is this: in tailoring communication for machines, we might lose something more fundamental. Our tone, our nuance, our ability to express complexity and doubt.
This is the paradox of working in an age of generative AI. To be understood by machines, we must become more structured. But to preserve meaning, we must stay deeply human.
The future of project delivery won’t just be about better software or smarter models. It will be about finding balance. Writing with intention, clarity, and structure, so that both human collaborators and AI systems can do their best work.
Because if we don’t shape how the internet speaks, it will keep speaking without us.
Rabbit hole:
Comments