Generative engine optimization

GEO: Generative Engine Optimization reached arXiv on 16 November 2023, was revised to a third version in June 2024 and was accepted to KDD 2024^[2]. Its six authors were at Princeton University, Georgia Tech, IIT Delhi and the Allen Institute^[5]. They built GEO-bench, a benchmark of 10,000 queries drawn from nine sources including MS MARCO and Natural Questions, split into 8,000 training queries with 1,000 each for validation and test, every query annotated with five cleaned relevant sources taken from Google Search^[6].

An answer engine has no fixed ranking slots, so the paper scores a source with a Position-Adjusted Word Count: how much of its content reaches the generated answer, and how prominently^[7]. Measured that way, the three top methods - citing sources, adding quotations and adding statistics - gained 30 to 40% over unoptimised text. Rewriting for fluency and readability gained 15 to 30%. Keyword stuffing scored below the unoptimised baseline, and writing in a more authoritative tone produced no significant improvement^[8]. The single best method in the paper's own table, adding quotations, beat the baseline by 41%^[9]. The strategies that worked ask for the same evidence-bearing writing that E-E-A-T guidance and content structure work already push toward.

Which engine is being optimised for changes what the work targets. A 2025 comparative study ran controlled experiments across verticals, languages and query paraphrases and found AI search systematically biased toward earned media, meaning third-party authoritative sources, over brand-owned and social content, against a more balanced mix in Google's results^[10]. The same work found the AI services differing from each other in domain diversity, freshness, cross-language stability and sensitivity to phrasing^[10]. A programme tuned and measured on one engine therefore says little about the next; see AI citation sources and multilingual GEO.

Answer Engine Optimization (AEO), LLM Optimization (LLMO) and AI Optimization circulate as synonyms, and no consensus definition separates them from GEO^[1]. The Reuters Institute's 2026 trends report glosses AEO as ways content providers can get better visibility within AI chatbots and other AI driven interfaces, and cross-references GEO to it. The same survey records a net -25 score for effort on old-style Google SEO, against a +61 net score for distribution via AI chatbots^[11].

Google's Search Liaison Danny Sullivan summarised the company's view in September 2025 as "Good SEO is good GEO, or AEO, AIO, LLM SEO, or LMNOPO"^[12]. The developer documentation says the same thing without the joke: optimizing for generative AI search is optimizing for the search experience and thus still SEO, and a page must already be indexed and eligible for an ordinary snippet before an AI feature can use it^[4]. SEO and GEO collects the evidence on where the two overlap and where they part. AI visibility measurement covers how either is measured against surfaces such as AI Overviews, ChatGPT Search and Perplexity.