Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

White Hat Search Engine Optimization using Large Language Models

Niv Bardas, Tommy Mordo, Oren Kurland, Moshe Tennenholtz, Gal Zur

TL;DR

This paper tackles white-hat SEO by using LLM-prompted document modification to influence future rankings for queries under undisclosed ranking functions. It introduces a two-part prompting framework and four context types—Pointwise, Pairwise, Listwise, and Temporal—together with metrics such as $TT$, $RF$, $OrigFaith$, $RCF$, and $CF$ to quantify ranking impact, faithfulness, and corpus faithfulness. Using LambdaMARTComp and E5Comp datasets (plus online bot-vs-student competitions), the study demonstrates that Pairwise and Listwise prompts achieve superior rank promotion and favorable corpus faithfulness compared to a feature-based baseline and human players, with some trade-offs in original faithfulness. The results highlight the practical viability of genAI-driven, white-hat document modification for both dense and sparse retrieval, while underscoring the need for faithfulness-aware evaluation to mitigate hallucinations and content drift.

Abstract

We present novel white-hat search engine optimization techniques based on genAI and demonstrate their empirical merits.

White Hat Search Engine Optimization using Large Language Models

TL;DR

This paper tackles white-hat SEO by using LLM-prompted document modification to influence future rankings for queries under undisclosed ranking functions. It introduces a two-part prompting framework and four context types—Pointwise, Pairwise, Listwise, and Temporal—together with metrics such as , , , , and to quantify ranking impact, faithfulness, and corpus faithfulness. Using LambdaMARTComp and E5Comp datasets (plus online bot-vs-student competitions), the study demonstrates that Pairwise and Listwise prompts achieve superior rank promotion and favorable corpus faithfulness compared to a feature-based baseline and human players, with some trade-offs in original faithfulness. The results highlight the practical viability of genAI-driven, white-hat document modification for both dense and sparse retrieval, while underscoring the need for faithfulness-aware evaluation to mitigate hallucinations and content drift.

Abstract

We present novel white-hat search engine optimization techniques based on genAI and demonstrate their empirical merits.

Paper Structure

This paper contains 6 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Document modification methods
  3. Experimental settings
  4. Experimental Results
  5. Conclusions
  6. Prompts

Figures (2)

  • Figure 1: Context-specific part of the Pairwise prompt which includes document pairs from the last three rankings. For each ranking, two documents, (a,b), (c,d), (e,f), were randomly selected and their rank is specified: r(a), r(b), r(c), r(d), r(e), r(f).
  • Figure 2: Context-specific part of the Listwise prompt which includes the latest ranking over documents: a, b, c, d (excluding the current document) and the previous ranking of the entire document list: e, f, g, h, i.