Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unveiling the Resilience of LLM-Enhanced Search Engines against Black-Hat SEO Manipulation

Pei Chen, Geng Hong, Xinyi Wu, Mengying Wu, Zixuan Zhu, Mingxuan Liu, Baojun Liu, Mi Zhang, Min Yang

Abstract

The emergence of Large Language Model-enhanced Search Engines (LLMSEs) has revolutionized information retrieval by integrating web-scale search capabilities with AI-powered summarization. While these systems demonstrate improved efficiency over traditional search engines, their security implications against well-established black-hat Search Engine Optimization (SEO) attacks remain unexplored. In this paper, we present the first systematic study of SEO attacks targeting LLMSEs. Specifically, we examine ten representative LLMSE products (e.g., ChatGPT, Gemini) and construct SEO-Bench, a benchmark comprising 1,000 real-world black-hat SEO websites, to evaluate both open- and closed-source LLMSEs. Our measurements show that LLMSEs mitigate over 99.78% of traditional SEO attacks, with the phase of retrieval serving as the primary filter, intercepting the vast majority of malicious queries. We further propose and evaluate seven LLMSEO attack strategies, demonstrating that off-the-shelf LLMSEs are vulnerable to LLMSEO attacks, i.e., rewritten-query stuffing and segmented texts double the manipulation rate compared to the baseline. This work offers the first in-depth security analysis of the LLMSE ecosystem, providing practical insights for building more resilient AI-driven search systems. We have responsibly reported the identified issues to major vendors.

Unveiling the Resilience of LLM-Enhanced Search Engines against Black-Hat SEO Manipulation

Abstract

The emergence of Large Language Model-enhanced Search Engines (LLMSEs) has revolutionized information retrieval by integrating web-scale search capabilities with AI-powered summarization. While these systems demonstrate improved efficiency over traditional search engines, their security implications against well-established black-hat Search Engine Optimization (SEO) attacks remain unexplored. In this paper, we present the first systematic study of SEO attacks targeting LLMSEs. Specifically, we examine ten representative LLMSE products (e.g., ChatGPT, Gemini) and construct SEO-Bench, a benchmark comprising 1,000 real-world black-hat SEO websites, to evaluate both open- and closed-source LLMSEs. Our measurements show that LLMSEs mitigate over 99.78% of traditional SEO attacks, with the phase of retrieval serving as the primary filter, intercepting the vast majority of malicious queries. We further propose and evaluate seven LLMSEO attack strategies, demonstrating that off-the-shelf LLMSEs are vulnerable to LLMSEO attacks, i.e., rewritten-query stuffing and segmented texts double the manipulation rate compared to the baseline. This work offers the first in-depth security analysis of the LLMSE ecosystem, providing practical insights for building more resilient AI-driven search systems. We have responsibly reported the identified issues to major vendors.

Paper Structure

This paper contains 25 sections, 5 equations, 4 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Background
  3. LLMSE & Black-Hat SEO
  4. Threat Model
  5. Attack Surface Analysis of LLMSE
  6. Representative LLMSE Collection
  7. Attack Surface Analysis
  8. Resilience of Black-Hat SEO Attack
  9. Experiment Setup
  10. Landscape
  11. Resilience on Understanding Phase
  12. Resilience on Retrieval Phase
  13. Resilience on Summarizing Phase
  14. LLMSEO Attack
  15. LLMSEO Attack Techniques
  16. ...and 10 more sections

Figures (4)

  • Figure 1: Examples of Traditional Search Engines vs. LLM-Enhanced Search Engines (LLMSE).
  • Figure 2: The Workflow and the Attack Surfaces of LLMSE. It includes three phases: (1) Understanding: LLMSE analyzes user query's intent and rewrites the original query; (2) Retrieval: fetching information from multiple databases with the rewritten queries, and re-ranking references; (3) Summarizing: gathers all the structured information and generates the final answer, outputting the summary content and references.
  • Figure 3: Distribution of Word Counts after Rewriting.
  • Figure 4: Domain Rank Proportions of Retrieved References from Both Traditional Search Engines and LLMSEs.