KDD 2024 GEO paper

"GEO: Generative Engine Optimization" (arXiv 2311.09735) is the foundational research paper on how to structure content so generative AI engines like ChatGPT and Perplexity cite it. Authored by Pranjal Aggarwal and colleagues at Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi, it was presented at KDD '24 and remains the most-cited reference in the AEO/GEO space as of 2026.

The paper tested seven content rewrites against Perplexity.ai and a controlled GPT-3.5 + top-5-Google retrieval rig, measuring citation lift via the Position-Adjusted Word Count (PAWC) metric. Three rewrites delivered statistically significant lifts: direct quotations from named sources (+41% — the highest single tactic), statistics with cited numbers (+31% on average, +37% specifically on Perplexity), and inline citations to authoritative sources (+27% on average, with a striking +115% lift on currently low-ranked pages).

The finding that low-ranked pages gain disproportionately from citations is particularly important for small businesses: it means GEO content patterns can democratize who gets cited, not just amplify existing authority. A page outside the top 3 organic Google results can still be heavily cited inside ChatGPT's answer if it has dense, well-cited, statistic-rich content.

The paper also identified two patterns that did NOT help and one that actively hurt: keyword density and authoritative tone alone produced no significant lift, and keyword stuffing measured roughly 10% worse than baseline — meaning classic SEO black-hat patterns actively harm AI-citation ranking.