Diagnosing and Repairing Citation Failures in Generative Engine Optimization

A team of researchers has published a new paper, "Diagnosing and Repairing Citation Failures in Generative Engine Optimization," introducing a critical fix for a major flaw in how AI models cite sources. The work addresses Generative Engine Optimization (GEO), the practice of optimizing content to appear in AI-generated responses from tools like ChatGPT or Perplexity. The researchers argue current GEO methods are flawed because they measure a document's general influence on an answer, not whether it actually gets a clickable citation—the real driver of creator traffic and revenue.

To solve this, the team built AgentGEO, an agentic AI system that acts like a diagnostic mechanic for broken citations. It uses a novel taxonomy to identify specific failure modes in the citation pipeline—such as a document being irrelevant, poorly formatted, or buried by competing information. Instead of applying generic rewriting rules to an entire article, AgentGEO selects targeted repairs from a tool library, like adjusting a headline or adding a key statistic, and iterates until a citation is secured. In tests, this precise approach yielded a 40% relative improvement in citation rates while modifying just 5% of the content, significantly outperforming baseline methods that achieved only 25% improvement.

The analysis revealed that blunt, generic optimization can actually harm visibility for niche, long-tail content, and some documents face structural challenges that optimization alone can't fix. This research has significant implications for creating a more equitable information ecosystem where diverse creators can be discovered and compensated through AI-mediated search, moving beyond simple SEO tactics to intelligent, diagnosis-driven content repair.