1. Introduction
The Internet once felt like a boundless public square: anyone could publish, anyone could read. But the rise of large language models (LLMs) like ChatGPT is reshaping that landscape. Increasingly, these systems sit between us and the web, summarizing, compressing, and redirecting the flow of information.
The have drawn the following diagram that maps three stages in this transition: the open web we knew ➡️ today’s model-mediated hybrid, ➡️ and a possible future in which AI systems become the primary gatekeepers of knowledge.
Figure 1: Three stages of the web: before (open, peer-to-peer), now (hybrid — models ingest and serve distilled content), and coming soon (models as gatekeepers that can block or silo the live web).
1️⃣Stage 1: Before — The Open Web
In the early days, the flow of content was simple and transparent:
- Individuals and entities published content directly to the open Internet. Blogs, forums, wikis, and websites were visible to anyone with a browser.
- Readers consumed that content directly. Search engines were mediators, but they pointed you back to the original source, where you could verify authorship and context.
The arrows in this stage represent two-way open flows:
- 🔵Blue arrow: content publishing went straight to the web.
- 🟢Green arrow: content consumption came straight from the web.
✅The open Internet acted as the canonical source of truth. If you wanted to know something, you looked it up, navigated to the source, and judged for yourself.
2️⃣Stage 2: Now — The Partially Hidden Internet
Fast-forward to today. Generative AI systems now sit in the middle of the content pipeline.
- Publishers still put their work online, but that content is increasingly being ingested by LLMs for training and contextualization.
- Models like ChatGPT internalize vast amounts of this content. Through training, they compress millions of documents into patterns, weights, and probabilities.
- Users often bypass the open web, asking the model instead. They receive distilled, synthesized answers — convenient, but detached from the original sources.
Here’s the nuance: the Internet is still open, but it’s becoming partially hidden by neglect. As fewer people click through to original sites, those sites effectively fade from visibility. The information is still there, but user habits obscure it.
The diagram’s arrows highlight this:
- Blue arrow: publishing still goes to the web.
- Internet → ChatGPT: the web now feeds training and context data.
- ChatGPT → Individuals/Entities: consumption increasingly comes from the model.
This subtle shift already has profound consequences:
- Publisher economics: Traffic declines as users no longer need to visit the source. Ad revenues and subscriptions shrink.
- Loss of provenance: Model answers rarely carry full citations. Readers get knowledge, but not its origin story.
- Data latency: Models update on snapshots. If you rely on them exclusively, you may be seeing outdated knowledge. But with ChaptGPT like system, this is not of much issue, as when it sense it needs to access the Internet, it does and pulls whatever required.
- Centralized mediation: Instead of many-to-many publishing and reading, we now have a few centralized AI intermediaries distilling the web for billions.
3️⃣Stage 3: Coming Soon — A Hidden and Outdated Internet?
The final panel of the diagram sketches a possible future if current trends accelerate.
- Content flows directly into AI platforms. Creators may publish through APIs or platform-specific formats. Over time, publishing “to the web” could become secondary.
- AI platforms block outward flow. Knowledge distilled by the model stays inside it. Links back to the open web may diminish or disappear altogether.
- The open Internet risks obsolescence. If new content bypasses the web and users stop visiting it, the web itself becomes outdated, stale, and hidden — not by censorship, but by disuse.
This creates a one-way street:
- Internet → AI → Users remains active (the web continues feeding the model).
- AI → Internet is blocked (knowledge doesn’t flow back into the open, linkable space).
- Users → AI dominates consumption.
So, question is: “Will the Internet die out?”.
“I’m in no rush to draw conclusions, but the trend is already clear: usage of Google Search — once the primary gateway to web portals — is rapidly declining.”
If unchecked, this scenario leads to several risks:
- Centralized knowledge control: A handful of companies decide what is surfaced and how it is phrased.
- Epistemic narrowing: The diversity of the web shrinks into a homogenized model output.
- Economic collapse of publishing: With no traffic, many creators won’t sustain open publication.
- Knowledge stagnation: The open Internet could freeze into a ghost archive of outdated material, while new insights circulate only inside proprietary silos.
2. What’s Really at Stake🌟
The arrows and blocks in this diagram tell a bigger story about attention, power, and trust.
- Attention: People follow the path of least friction. If the fastest way to get an answer is through a model, they’ll use it — even if that hides the source.
- Power: Whoever controls the model controls access to knowledge. This centralizes influence in unprecedented ways.
- Trust: Without links or provenance, we must trust the model’s synthesis. But trust without transparency is fragile.
3. Three Possible Futures🔮
The diagram presents a pessimistic scenario, but the future is not locked. Here are three trajectories:
1️⃣Model-First Monopoly (pessimistic)
LLMs dominate consumption. The open web shrivels. Knowledge lives in silos controlled by a few companies. Transparency and diversity decline.
2️⃣Hybrid Web with Safeguards (moderate, plausible)
Models remain central, but they integrate retrieval from live sources, enforce provenance, and link back to original sites. Publishers are compensated via licensing. The open web shrinks in importance but stays relevant.
3️⃣Open, Accountable AI Ecosystem (optimistic)
Standards, regulation, and user demand ensure models must cite sources and share value with creators. Open-source models and decentralized tools keep the open Internet alive as the foundation for all AI.
4. What Needs to Happen Next✅
The Internet doesn’t have to become hidden and outdated. There are practical steps stakeholders can take:
For publishers and creators:
- Use structured metadata (schema.org, sitemaps) to make content machine-readable.
- Explore licensing or API partnerships with model providers.
- Build direct community value: newsletters, podcasts, events — channels models can’t easily replicate.
For AI developers:
- Prioritize provenance: always link to sources in outputs.
- Respect content rights: honor robots.txt, offer opt-outs, and negotiate fair licensing.
- Reduce knowledge latency: combine training with live retrieval (RAG).
For policymakers:
- Require transparency about training datasets.
- Mandate citation and fair compensation mechanisms.
- Protect the open Internet as critical public infrastructure.
For users:
- Demand answers with citations.
- Support creators directly.
- Stay aware: a model’s convenience comes with tradeoffs in diversity and context.
5. Conclusion: Will the Web Die Out?
The arrows in my diagram are more than technical flows. They are signals of where culture, economics, and trust may shift.
The open Internet flourished because it was transparent, participatory, and decentralized. Generative AI offers enormous convenience, but if it becomes the only interface to knowledge, we risk burying the very ecosystem that gave rise to it.
The Internet doesn’t have to die. But unless we actively design models, policies, and habits that keep it alive, the most likely outcome is slow neglect — a gradual hiding of the web, not by censorship, but by inattention.
The question isn’t just whether the web will survive. The deeper question is: Do we want our knowledge ecosystem to be open and diverse, or closed and centralized?
The answer depends on what we do today.
