An AI-Based Multi-Agent Framework for Search Engine Optimization

Abstract

In this paper, we propose a research-grounded intent-aware multi-agent framework for the automation of Search Engine Optimisation (SEO) content generation and evaluation. The framework decomposes the entire workflow into specialised components for research, content generation, technical validation, and final evaluation. The Content SEO Agent builds structured SEO assets based on the evidence it finds, while the On-Page Technical SEO Agent looks at metadata, keyword placement, heading structure, readability, formatting, and Schema.org structured data. The produced output is then evaluated by a Fusion and Evaluation Agent that combines rule-based, semantic, technical and source-faithfulness criteria. The framework also includes an evaluation and refinement mechanism that identifies the unsatisfactory outputs and performs a targeted correction to the concerned part of the workflow instead of re-executing the whole process. The proposed framework was evaluated via retrieval experiments and an extended SEO benchmark. The extended evaluation achieved 92.46% accuracy for intent classification and a final internal approval rate of 98.99% after refinement. The results indicate that the separation of responsibilities of generation, validation and evaluation can contribute to a more structured and traceable approach to AI-assisted SEO automation. But the study had its limitations too. In particular, the classification of navigational intent needs further evaluation and the experiments did not include a directly matched baseline where failed outputs were regenerated without targeted feedback.