MAGEO framework treats GEO as reusable strategy learning
Researchers Beining Wu, Fuyou Mao, and Jiong Lin propose MAGEO, a multi-agent framework that reframes Generative Engine Optimization as a strategy learning problem, enabling validated editing patterns to be distilled into reusable, engine-specific skills rather than optimizing each query in isolation.
Score breakdown
Practitioners building content strategies for AI-powered search engines can use MAGEO's reusable, engine-specific skill framework to systematically improve citation visibility across multiple generative engines rather than hand-tuning each piece of content independently.
- 01MAGEO is a multi-agent framework that reframes GEO as a strategy learning problem rather than per-instance optimization.
- 02Its agents cover coordinated planning, editing, and fidelity-aware evaluation, with validated patterns distilled into reusable engine-specific skills.
- 03A Twin Branch Evaluation Protocol enables causal attribution of content edits.
Generative engines (GEs) are changing how users access information by replacing ranked link lists with citation-grounded answers, but existing GEO approaches optimize each content instance in isolation — meaning no knowledge or strategy is carried over between tasks or across different engines. Beining Wu, Fuyou Mao, and Jiong Lin propose MAGEO, a multi-agent framework that reframes GEO as a strategy learning problem. Its execution layer consists of coordinated planning, editing, and fidelity-aware evaluation agents, while validated editing patterns are progressively distilled into reusable, engine-specific optimization skills — allowing the system to accumulate expertise over time rather than starting from scratch on each query.
They also release MSME-GEO-Bench, a multi-scenario, multi-engine benchmark built from real-world queries.
To support rigorous evaluation, the authors introduce two new contributions alongside MAGEO: a Twin Branch Evaluation Protocol designed for causal attribution of content edits, and DSV-CF, a dual-axis metric that jointly measures semantic visibility and attribution accuracy. They also release MSME-GEO-Bench, a multi-scenario, multi-engine benchmark built from real-world queries. Experiments on three mainstream generative engines demonstrate that MAGEO substantially outperforms heuristic baselines on both visibility and citation fidelity metrics. Ablation studies confirm that engine-specific preference modeling and strategy reuse are the primary drivers of these gains, pointing toward a scalable, learning-driven paradigm for trustworthy GEO. Code is publicly available at the project's GitHub repository.
Key facts
- 01MAGEO is a multi-agent framework that reframes GEO as a strategy learning problem rather than per-instance optimization.
- 02Its agents cover coordinated planning, editing, and fidelity-aware evaluation, with validated patterns distilled into reusable engine-specific skills.
- 03A Twin Branch Evaluation Protocol enables causal attribution of content edits.
- 04DSV-CF is a new dual-axis metric unifying semantic visibility with attribution accuracy.
- 05MSME-GEO-Bench is a new multi-scenario, multi-engine benchmark grounded in real-world queries.
- 06Experiments on three mainstream generative engines show MAGEO substantially outperforms heuristic baselines in visibility and citation fidelity.
- 07Ablations confirm engine-specific preference modeling and strategy reuse are central to performance gains.