Generative Engine Optimization, or GEO, is the practice of adapting content so that generative AI answer engines use and credit it, and the term comes from a specific 2023 research paper. The bottom line: the paper tested nine content changes and found that adding citations, quotations, and statistics improved visibility by roughly 30 to 40 percent on its measures, while keyword stuffing did not help, and later reviews caution that results depend on the engine, the topic, and how visibility is scored.
New to the term? What is answer engine optimization covers the basics.
The paper that named it
"GEO: Generative Engine Optimization" was written by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande, from IIT Delhi, Princeton University, the Allen Institute for AI, and Georgia Tech. It appeared on arXiv in November 2023 and was presented at the 30th ACM SIGKDD Conference in Barcelona, August 25 to 29, 2024. It proposed a black box optimization framework, which needs no access to a model's internals, and GEO-bench, a benchmark of 10,000 queries across 25 domains split into training, validation, and test sets.
What the paper measured
The authors did not measure whether a page was cited. They used two visibility scores: Position Adjusted Word Count, which credits the amount of an answer attributed to a source, weighted by where the source is cited, and Subjective Impression, a model scored blend of relevance and influence. That matters when you read the headline. "Up to 40 percent" is a maximum on these measures, in the authors' generative engine, not a promised lift in citations.
What worked and what did not
| Method | Result in the paper |
|---|---|
| Cite Sources | Among the strongest, roughly 30 to 40 percent relative gain on Position Adjusted Word Count |
| Quotation Addition | Among the strongest, with up to about 41 percent on that measure and 28 percent on Subjective Impression |
| Statistics Addition | Among the strongest, roughly 30 to 40 percent |
| Fluency Optimization and Easy to Understand | Also raised visibility, about 15 to 30 percent |
| Keyword Stuffing | Did not help, and performed worse than the baseline in the authors' engine |
The authors also found the effect varied by domain, that combining methods helped, and that the best pairing was fluency with statistics. They checked their best methods on Perplexity as a real world test. Note that our earlier version of this article listed different tactics from these, so if you read that version, this list is the one taken from the paper.
What later research adds
A critical survey of GEO work from 2023 to 2026 adds cautions worth taking seriously: heuristics transfer poorly across contexts, relevance often matters more than surface tricks, competition can erode a page's gains when many sources optimize for the same query, and over optimizing can hurt retrieval by the search index underneath. Different engines cite different sources, and answers vary across runs of the same question.
Using it in practice
- Add sourced statistics. Numbers with the source and date named.
- Add real quotations. From named people, with role and source.
- Cite your sources. Link to the primary source for each claim.
- Write clearly. Plain, fluent prose helped in the paper.
- Do not stuff keywords.
- Test on your own topic. Change a page and compare the share of runs that name you before and after.
GEO and AEO
GEO is the research framework, with a benchmark and measured results, and AEO is the broader industry label. In practice the work overlaps almost completely, and both sit on top of ordinary SEO fundamentals, which Google says still apply to its AI features.
Questions people ask
What is generative engine optimization?
Adapting content so generative AI answer engines use and credit it. The term comes from a 2023 paper by researchers at IIT Delhi, Princeton, the Allen Institute for AI, and Georgia Tech, presented at KDD 2024.
What did the GEO paper find worked?
Adding citations, quotations from relevant sources, and statistics gave the strongest gains, roughly 30 to 40 percent on its visibility measure. Keyword stuffing did not help.
Does GEO guarantee a 40 percent increase in citations?
No. The figure is a maximum on the paper's Position Adjusted Word Count measure in its test engine, and later reviews say results vary by engine, domain, and competition.
What is GEO-bench?
A benchmark of 10,000 queries across 25 domains that the authors built to test content changes against a generative engine.