Overview

This project studies how Generative AI reshapes the internet as an information ecosystem, focusing on both the supply of online content and the quality of information users consume.

I develop a simple equilibrium model in which:

  • users choose between traditional web sources and AI,
  • content creators decide whether to enter and produce information,
  • and AI both aggregates existing knowledge and changes production costs.

The key insight is that GenAI introduces two opposing forces:

  • it improves information access by aggregating dispersed content,
  • but it also diverts attention away from the web, weakening incentives to produce that content.

As a result, the effect of GenAI is non-monotonic:

When AI is weak, it can reduce both content supply and consumed quality; when sufficiently strong, the same forces can reverse and improve outcomes.

Introduction

The central question is:

How does Generative AI affect the supply and quality of online information?

The internet is now a core infrastructure for learning, decision-making, and knowledge production. Changes in how information is produced and accessed have direct implications for productivity, innovation, and welfare.

The project frames this as a general equilibrium interaction between three actors:

  • Users (demand side) choose how to access information: via the web or via AI.
  • Content creators (supply side) decide whether to produce based on expected traffic and costs.
  • Generative AI acts both as:
    • a competing information source, and
    • a technology that lowers production costs.

This dual role is central: AI simultaneously substitutes for the web and reshapes the economics of content creation.

Methodology

The Internet as an Equilibrium System

The model treats the internet as a market for attention and information:

  • Users allocate attention across sources.
  • Creators supply content if it is profitable.
  • AI both competes for attention and depends on the content it aggregates.

This creates a feedback loop:

  • AI relies on web content,
  • but also weakens incentives to produce it.

Demand: Choosing Between AI and the Web

Users choose between:

  • AI, which provides a single aggregated answer,
  • and the web, which consists of many heterogeneous sources.

As AI improves, it captures a larger share of user attention. This is the business-stealing channel: AI diverts traffic away from traditional sources.

Supply: Content Creation

Content creators differ in quality and decide whether to enter based on:

  • expected user traffic,
  • and fixed production costs.

GenAI affects entry through two channels:

  • Negative: less web traffic reduces revenues
  • Positive: lower production costs make entry easier

The net effect on content supply depends on which force dominates.

AI as an Aggregator

AI aggregates information from the web rather than producing it independently.

Its quality depends on:

  • its own capability,
  • and the amount and quality of available web content.

This implies that AI performance is endogenous to the health of the information ecosystem it relies on.

Core Mechanism

The model highlights three forces:

1. Aggregation Effect (positive)

  • AI improves how information is synthesized.
  • Users may receive higher-quality answers.

2. Business-Stealing Effect (negative)

  • AI diverts users away from the web.
  • Lower traffic reduces incentives to create content.

3. Cost-Reduction Effect (positive)

  • AI lowers production costs.
  • More creators can profitably enter.

Main Result: A Threshold Effect

The effect of GenAI is not monotonic:

  • When AI is weak:

    • Users shift away from the web,
    • Cost reductions are limited,
    • → Content supply and consumed quality decline

  • When AI is sufficiently effective:

    • Aggregation becomes strong,
    • Cost reductions support entry,
    • → Content supply and quality increase

GenAI can therefore initially degrade the information environment, but improve it once it becomes sufficiently capable.

What the Model Predicts

Key predictions:

  • Higher AI capability

    • increases AI usage,
    • reduces direct web traffic,
    • weakens incentives to produce content,
    • but improves aggregation quality.

  • Stronger cost reductions

    • increase entry,
    • expand the stock of information,
    • offset part of the demand-diversion effect.

These forces generate regions where GenAI:

  • reduces information supply and quality,
  • or improves both.

Current Scope and Next Steps

Current stage

  • A tractable equilibrium model of:
    • user choice,
    • content supply,
    • and AI aggregation.

Next steps

  • Bring the model to data
  • Estimate key parameters
  • Quantify the importance of:
    • demand diversion vs. cost reduction

Conclusion

This project provides a framework for thinking about GenAI as a general equilibrium shock to the internet.

The core tradeoff is:

  • AI improves access to information,
  • but may undermine incentives to produce it.

Whether GenAI improves or degrades the information environment depends on:

  • how strongly it substitutes for the web,
  • how much it lowers production costs,
  • and how effective it is at aggregating knowledge.

The future quality of information online depends not only on AI itself, but on whether the ecosystem it relies on continues to exist.