Do AI Overviews Benefit Search Engines? An Ecosystem Perspective

Yihang Wu, Jiajun Tang, Jinfei Liu, Haifeng Xu, Yao Fan

TL;DR

AI Overviews provide instant answers but cannibalize the traffic that content creators rely on, potentially collapsing the very ecosystem that feeds the AI. This paper uses game theory to prove that search engines must implement citation or compensation mechanisms to remain profitable in the long run, offering a "win-win" blueprint for the GenAI search era.

What Problem Does This Paper Address?

When you search for "how to fix a leaky faucet," an AI Overview might give you the steps directly. You get your answer, but you never click the link to the DIY blog that wrote the original content. This is the "ecosystem" problem: if creators lose traffic, they lose the incentive to produce high-quality content. Without new content, the AI has nothing to learn from, and search engine quality—and eventually profit—plummets.

The authors tackle two main questions:

  1. Do AI Overviews actually benefit search engines in the long run once creators adjust their behavior?
  2. How can search engines design incentives (like citations or payments) to keep creators engaged while maximizing their own profit?

Methods

The researchers built a game-theoretic model based on the classic Position-Based Model (PBM). In this world, $n$ creators compete for a query by choosing an "effort level" (content quality). A creator’s traffic depends on their rank (position bias) and their effort.

The paper introduces two novel intervention mechanisms:

  • Citation Mechanism: The AI Overview cites specific organic results, giving them a visibility boost within the AI response itself.
  • Compensation Mechanism: The search engine directly pays creators based on their effort/quality.

The authors solve for the Mixed Nash Equilibrium (MNE)—the state where no creator wants to change their strategy—under both symmetric (all creators are equal) and asymmetric (high-ability vs. low-ability creators) settings. They then optimize these mechanisms to maximize search engine profit, which they define as revenue from user welfare minus any compensation costs.

Datasets

To ground their math in reality, the authors conducted a real-world user click experiment:

  • Participants: 50 users.
  • Queries: 24 search tasks (12 fact-based, 12 open-ended).
  • Interfaces: Three versions of a simulated search engine ("Baigle"):
  • Type A: Traditional organic results only.
  • Type B: AI Overview without citations.
  • Type C: AI Overview with four citations.
  • Output: They estimated the "position bias" (inherent visibility) of AI Overviews versus traditional links, finding that an AI Overview captures nearly as much attention (~0.99) as the top-ranked organic result in a traditional search.

Key Insights and Ideas

  • The "Golden Goose" Risk: The model confirms that while AI Overviews might boost engagement today, they act as a "tax" on creators. Without intervention, creators reduce effort, leading to a lower-quality web and lower long-term search engine profits.
  • Multiplicative Payoffs: Unlike standard "all-pay auctions" where the winner takes a fixed prize, search traffic is multiplicative. Your reward is (Visibility) × (Your Quality). This nuance significantly changes how creators compete compared to traditional economic models.
  • The "Uniform-But-Last" (UBL) Strategy: For symmetric creators, the most effective citation strategy is to equalize the visibility boost across almost all top creators, excluding only the very last ones. This maintains a "competitive tension" that keeps everyone working hard.
  • Piecewise-Constant Incentives: A surprising finding in the mechanism design is that search engines don't need complex, granular payment tiers. Near-optimal profit can be achieved by paying all creators of the same "type" (e.g., all "experts") the same amount.

Key Results

  • AI Overviews are Cannibalistic: Introducing AI Overviews without incentives reduces long-term search engine profit across all tested scenarios.
  • Incentives Work: Implementing the proposed citation (UBL/TPBL) or compensation mechanisms can restore and even increase long-term profit compared to a world without AI.
  • Profitability vs. Cost: The effectiveness of these mechanisms is highly sensitive to the search engine's "profitability constant" ($\alpha$) and the "cost of effort" ($\beta$). Incentives are most valuable for queries with high creation costs (e.g., specialized academic topics) where creators are most likely to quit if traffic drops.

Limitations and Future Work

The authors acknowledge several areas for further exploration:

  • Binary Types: The asymmetric analysis focuses on only two types of creators (High vs. Low ability). Real-world ecosystems have a continuous spectrum of creator capabilities.
  • Inference Challenges: The model assumes the search engine can accurately infer a creator's effort and cost structure, which is difficult in practice.
  • Static User Behavior: The study uses the PBM model, but user behavior might evolve as people become "power users" of AI search, potentially changing how they value citations versus the raw AI summary.
  • Dynamic Games: The current model is a one-shot game; a multi-round dynamic model could better capture how creators build reputation over time.

Summary updated

Abstract

The integration of AI Overviews into search engines enhances user experience but diverts traffic from content creators, potentially discouraging high-quality content creation and causing user attrition that undermines long-term search engine profit. To address this issue, we propose a game-theoretic model of creator competition with costly effort, characterize equilibrium behavior, and design two incentive mechanisms: a citation mechanism that references sources within an AI Overview, and a compensation mechanism that offers monetary rewards to creators. For both cases, we provide structural insights and near-optimal profit-maximizing mechanisms. Evaluations on real click data show that although AI Overviews harm long-term search engine profit, interventions based on our proposed mechanisms can increase long-term profit across a range of realistic scenarios, pointing toward a more sustainable trajectory for AI-enhanced search ecosystems.
BibTeX
@misc{wu2026do,
  author = {Yihang Wu and Jiajun Tang and Jinfei Liu and Haifeng Xu and Yao Fan},
  title = {{Do AI Overviews Benefit Search Engines? An Ecosystem Perspective}},
  year = {2026},
  eprint = {2601.22493},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.22493},
}
Figure