On 3 November 2026, the most cited generative engine optimization (GEO) paper was GEO: Generative Engine Optimization by Aggarwal et al., with 128 Semantic Scholar citations, followed by Generative Engine Optimization: How to Dominate AI Search (49) and What Generative Search Engines Like and How to Optimize Web Content Cooperatively (22).
The field is three years old and almost all of it is from 2026. Of the 59 arXiv papers whose abstract uses the phrase "generative engine optimization", 53 were posted in 2026, five in 2025 and one in 2023. The 2023 paper that named the field holds 34% of all the citations the group has received. The median paper has 2 citations, and 21 have none yet.
What are the most cited GEO papers?
| Rank | Paper | What it contributes | Citations | Influential | Posted |
|---|---|---|---|---|---|
| 1 | GEO: Generative Engine Optimization | Names the field; proposes GEO and the GEO-bench benchmark | 128 | 26 | Nov 2023 |
| 2 | Generative Engine Optimization: How to Dominate AI Search | Compares AI search with Google across verticals and languages | 49 | 5 | Sep 2025 |
| 3 | What Generative Search Engines Like and How to Optimize Web Content Cooperatively | AutoGEO: learns engine preference rules, then rewrites content | 22 | 7 | Oct 2025 |
| 4 | E-GEO: A Testbed for Generative Engine Optimization in E-Commerce | 13,747 shopping queries, each paired with 10 Amazon listings | 16 | 1 | Nov 2025 |
| 5 | SAGEO Arena | End-to-end environment covering retrieval and reranking, not just generation | 10 | 0 | Feb 2026 |
| 5 | AgenticGEO | Self-evolving agent that adapts its optimisation strategy to the content | 10 | 1 | Mar 2026 |
| 5 | How Generative AI Disrupts Search | 11,500 queries compared across Google, AI Overviews and Gemini | 10 | 5 | Apr 2026 |
| 8 | From Citation Selection to Citation Absorption | Separates being cited from actually shaping the answer | 9 | 1 | Apr 2026 |
| 9 | Beyond SEO: A Transformer-Based Approach for Reinventing Web Content Optimisation | Fine-tuned BART rewriter on 1,905 travel-site content pairs | 8 | 0 | Jul 2025 |
| 9 | Multimodal Generative Engine Optimization | Image and text perturbations that move vision-language rankers | 8 | 0 | Jan 2026 |
| 9 | Diagnosing and Repairing Citation Failures in Generative Engine Optimization | Taxonomy of why a page is not cited; AgentGEO repairs it | 8 | 1 | Mar 2026 |
| 9 | Structural Feature Engineering for Generative Engine Optimization | How document, chunk and emphasis structure change citation | 8 | 1 | Mar 2026 |
Citation counts were read from Semantic Scholar on 3 November 2026. Papers with the same count share a rank, which is why this list has twelve entries: four papers are tied at 8. "Influential" is Semantic Scholar's count of citations that build substantially on the paper.
Where can you verify each paper's numbers?
- GEO: Generative Engine Optimization — arXiv · Semantic Scholar
- Generative Engine Optimization: How to Dominate AI Search — arXiv · Semantic Scholar
- What Generative Search Engines Like — arXiv · Semantic Scholar
- E-GEO — arXiv · Semantic Scholar
- SAGEO Arena — arXiv · Semantic Scholar
- AgenticGEO — arXiv · Semantic Scholar
- How Generative AI Disrupts Search — arXiv · Semantic Scholar
- From Citation Selection to Citation Absorption — arXiv · Semantic Scholar
- Beyond SEO — arXiv · Semantic Scholar
- Multimodal Generative Engine Optimization — arXiv · Semantic Scholar
- Diagnosing and Repairing Citation Failures — arXiv · Semantic Scholar
- Structural Feature Engineering for GEO — arXiv · Semantic Scholar
How were these papers selected?
One mechanical rule: the paper's arXiv abstract contains the exact phrase "generative engine optimization". That search returned 59 papers, all of which resolved to a Semantic Scholar record. A 60th, posted on 31 October 2026, appeared in arXiv's search index only later the same day; its Semantic Scholar record could not be read on 3 November, so it is not ranked or counted below. They were then ordered by citation count. Nothing was added or removed by hand.
A phrase rule is narrow on purpose. It keeps out papers that are about AI search in general, and it makes the list reproducible: anyone running the same arXiv search gets the same population.
Why is "Evaluating Verifiability in Generative Search Engines" not on the list?
With 497 citations it is cited far more than any paper above, and it is often read alongside them. But it audits whether generative search engines' citations support their answers; it is about the engines, not about optimising content for them, and its abstract never uses the phrase. The same goes for Manipulating Large Language Models to Increase Product Visibility (52 citations), which studies the same effect under a different name.
What does each paper contribute?
1. GEO: Generative Engine Optimization — 128 citations
The paper that named the field, by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan and colleagues, published at KDD. It frames AI answer engines as "generative engines" and argues that content creators have little control over when and how their content appears in them. It introduces GEO as a black-box optimisation framework and GEO-bench, a benchmark of user queries across domains, and reports that GEO can raise visibility in generative engine responses by up to 40%. AutoGEO, third on this list, evaluates on GEO-bench.
2. Generative Engine Optimization: How to Dominate AI Search — 49 citations
Mahe Chen, Xiaoxuan Wang, Kaiwen Chen and Nick Koudas compare AI search (ChatGPT, Perplexity, Gemini) with Google across verticals, languages and query paraphrases. Their central finding is that AI search shows a systematic bias towards earned media — third-party, authoritative sources — over brand-owned and social content, and that the AI engines differ from each other in domain diversity, freshness and sensitivity to phrasing.
3. What Generative Search Engines Like and How to Optimize Web Content Cooperatively — 22 citations
AutoGEO, from Yujiang Wu, Shanshan Zhong, Yubin Kim and Chenyan Xiong. It prompts frontier models to explain which retrieved content they prefer, distils those explanations into preference rules, and uses the rules both as a prompt-based rewriter and as rewards to train a smaller rewriting model.
4. E-GEO: A Testbed for Generative Engine Optimization in E-Commerce — 16 citations
The first dataset built for GEO in shopping: 13,747 realistic consumer product queries, each paired with 10 retrieved Amazon listings. The authors use it to compare five generative engines, seven LLM rewriters and fifteen hand-crafted rewriting heuristics.
5. SAGEO Arena — 10 citations
Argues that existing GEO benchmarks start from a fixed set of candidate documents and so skip retrieval and reranking, the stages that decide whether a page is seen at all. SAGEO Arena evaluates optimisation end to end and keeps structural signals such as schema markup.
5. AgenticGEO — 10 citations
Treats GEO as a control problem rather than a fixed rewriting rule: a self-evolving agent keeps an archive of strategies and picks one conditioned on the content, so it can adapt as black-box engines change.
5. How Generative AI Disrupts Search — 10 citations
A SIGIR paper with a public dataset of 11,500 user queries, each run through Google search, Google's AI Overview and Gemini. AI Overviews appeared for 51.5% of representative queries, and the sources each system retrieves barely overlap (average Jaccard similarity below 0.2).
8. From Citation Selection to Citation Absorption — 9 citations
Splits GEO into two measurable stages: whether a platform chooses a page as a source, and whether that page then shapes the answer. Across 602 prompts on ChatGPT, Google AI Overview/Gemini and Perplexity, it finds that ChatGPT cites fewer sources than the other two but each cited page carries more influence.
9. Beyond SEO — 8 citations
Fine-tunes a BART-base model on 1,905 pairs of raw and optimised travel-website text, then tests whether the rewrites are more visible in a large model's answers.
9. Multimodal Generative Engine Optimization — 8 citations
Shows that vision-language models used to rank products can be steered by small image perturbations combined with fluent text, more than by either alone.
9. Diagnosing and Repairing Citation Failures — 8 citations
Asks why a specific page is not cited instead of applying one rewrite to everything. It proposes a taxonomy of citation failures and an agent, AgentGEO, that diagnoses and repairs them, reporting over 40% relative improvement in citation rate while changing about 5% of the content.
9. Structural Feature Engineering for Generative Engine Optimization — 8 citations
Looks at structure instead of wording: document architecture, how information is chunked, and visual emphasis, and how each changes citation probability across six generative engines.
What are the limits of this ranking?
- Citations lag. Most of these papers are months old; a paper posted in September 2026 has had almost no time to be cited, so this list favours the earliest work.
- One phrase defines the field here. Work that uses other names — answer engine optimization, LLM visibility, ranking manipulation — is excluded even when the subject is the same.
- arXiv only. Papers published without an arXiv version are not in the search.
- Citations measure attention, not correctness. Several of these papers report effects measured on their own benchmarks.
How is this different from being cited by an AI search engine?
These are academic citations: how often researchers reference a paper. Being cited by an AI search engine is the thing most of these papers study — whether an assistant names or links your page when someone asks a question. That is measured across a set of real queries, by which sources and brands get cited. For the practical side, start with what AI visibility is or how to track brand mentions in AI search.
Frequently asked questions
GEO: Generative Engine Optimization by Aggarwal et al., posted in November 2023 and published at KDD, with 128 Semantic Scholar citations on 3 November 2026. It introduced the term and the GEO-bench benchmark.
59 arXiv papers with a Semantic Scholar record use the exact phrase in their abstract as of 3 November 2026, plus one posted on 31 October that could not yet be ranked. Of the 59, 53 were posted in 2026, five in 2025 and one in 2023.
Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan and colleagues, in the 2023 paper GEO: Generative Engine Optimization.
Generative Engine Optimization: How to Dominate AI Search reports a systematic bias towards earned media, meaning third-party authoritative sources, over brand-owned and social content.
Several. GEO-bench came with the original paper; E-GEO covers e-commerce with 13,747 queries; SAGEO Arena evaluates retrieval and reranking end to end.
The short version
On 3 November 2026 the most cited generative engine optimization papers were the 2023 paper that named the field (128 citations), a large comparison of AI search with Google (49) and AutoGEO (22), followed by E-GEO, SAGEO Arena, AgenticGEO and a SIGIR study of AI Overviews. Of the 59 papers that use the phrase, 53 are from 2026, so the ranking will move quickly.