Generative Engine Optimization (GEO) is the academic framework for improving content visibility in AI generative engines, the systems that synthesise answers from multiple web sources. It was formalised in a 2024 paper by researchers at Princeton University, IIT Delhi, and the Allen Institute for AI, presented at the ACM SIGKDD 2024 conference. The paper is the closest thing the field has to a primary, peer-reviewed source.
The study built GEO-bench, a benchmark of 10,000 real queries across nine domains, and measured how content edits changed a source\'s visibility inside a generative answer. The headline finding: the best techniques lifted visibility by up to 40%. The breakdown matters more. Every edit that worked was a change to the writing itself, adding citations, quotations, and statistics, not keyword manipulation.
GEO is the research foundation for what practitioners call AEO. The terms overlap almost entirely in practice. GEO is used in research and by technical specialists; AEO is the term marketers and consultants use. This guide explains the study\'s findings and how to apply them.
Where Does the Term GEO Come From?
A search system that uses a large language model to gather and summarise information from multiple web sources to answer a query, rather than returning a list of links. The term "generative engine" was formalised in the Princeton GEO paper. Examples include ChatGPT with search, Perplexity, and Google AI Overviews.
The term comes from the 2024 paper "GEO: Generative Engine Optimization" by Pranjal Aggarwal, Vishvak Murahari, and others, published at KDD 2024 in Barcelona. The authors framed generative engines as a step change replacing traditional search, and identified a gap: content creators had little control over how their content appeared in AI-generated answers. GEO was proposed to address this.
The paper introduced GEO-bench, a large-scale benchmark of 10,000 queries across diverse domains, and a set of optimisation methods tested against a baseline. It also validated the methods on Perplexity.ai, a real deployed generative engine, demonstrating real-world impact. The work is open and the code and data are publicly available.
What Did the Princeton GEO Study Actually Find?
The study tested content edits against a baseline with no optimisation. The results ranked by visibility lift on the Position-Adjusted Word Count metric:
- Quotation addition: Up to 41% lift. Adding relevant quotations from credible sources had the largest effect.
- Statistics addition: Up to 33% lift. Inserting specific data points wherever relevant increased visibility.
- Citing sources: Up to 30% lift. Adding the source of a statement significantly boosted citation.
- Fluency optimisation: Up to 28% lift. Improving readability and flow helped.
- Authoritative wording: Up to 12% lift. Using confident, authoritative language had a smaller but positive effect.
Traditional SEO tactics failed. Keyword stuffing performed 10% worse than the baseline. The lesson is consistent: generative engines reward evidence and credibility, not repetition. Combining methods amplified the effect, with the best combination (fluency plus statistics) outperforming any single strategy by over 5%.
How Was GEO Validated on a Real Engine?
To demonstrate real-world impact, the authors tested the methods on Perplexity.ai, a deployed generative engine with a large user base. The results confirmed the findings:
- Quotation addition performed best, with a 22% improvement on Position-Adjusted Word Count.
- Cite Sources showed improvements up to 9%.
- Statistics Addition showed improvements up to 37% on the Subjective Impression metric.
- Keyword stuffing again underperformed, by 10%, reinforcing that traditional SEO tactics do not transfer.
The validation matters because it shows the methods generalise across different generative engines, not just the authors\' custom system. It also demonstrates content creators can apply the methods directly with minimal effort. See our dedicated guide on Perplexity optimisation for practical application.
How Does GEO Relate to AEO and SEO?
The three terms describe related disciplines with different origins:
- SEO: Optimising for ranking positions in traditional search engines. The foundation. Without it, content is not indexed or found.
- GEO: The academic framework for optimising visibility in generative engines. Research-driven, focused on what content edits increase citation.
- AEO: The practitioner term for the same goal. Used by marketers and consultants. Functionally interchangeable with GEO in day-to-day practice.
They are complementary, not competing. SEO gets you indexed and ranked. GEO/AEO tactics increase the chance your ranked content is cited inside AI answers. For Google specifically, where AI features reuse core ranking, the overlap is strongest. For ChatGPT and Perplexity, GEO tactics matter more because there is less reliance on traditional ranking signals.
What Are the Limitations of the GEO Study?
The study is valuable but has important limitations the authors acknowledge and practitioners must respect:
- Best case, not average: The 40% figure is the headline from the best-performing techniques in the most responsive domains. A typical page should expect much less.
- Domain dependence: Effectiveness varies by topic. What works in one field may do little in another. Debate-heavy topics behave differently from factual ones.
- Historical engine: The study used a 2023-era engine. Today\'s models differ. The broad finding holds, but treat exact percentages as historical, not current.
- Test before relying: The authors recommend testing GEO tactics on your own content before counting on uniform results.
Treat GEO as evidence that well-sourced, clearly structured content shows up more often in AI answers, not as a guarantee of a specific lift. Apply the tactics, measure your own results, and adjust. Our measurement guide covers how to track outcomes.