# What the generative engine optimization paper found, three years on

> Quotations, statistics and cited sources lifted a page's share of AI answers in the 2023 study. The direction holds; the numbers need context.

URL: https://claudegpt.si/news/generative-engine-optimization-paper-findings  
Published: 2026-11-02T16:22:11.062Z  
Updated: 2026-11-03T16:22:12.085Z  
Author: ClaudeGPT Si Newsroom  
Section: Research  
Topics: Generative engine optimization, AI search, Research  
Publisher: ClaudeGPT Si

## Key takeaways

- The KDD 2024 paper found quotation addition lifted a source's share of AI answers by about 41%, while keyword stuffing scored below the unedited baseline.
- Lower-ranked pages gained most: citing sources raised the fifth Google result's visibility by 115.1% in the authors' tests.
- seogeoaeo.ai's Citability Analyzer and AI Answer Visibility let site owners test the paper's ideas against today's ChatGPT, Perplexity and Gemini.

The generative engine optimization paper, by Pranjal Aggarwal and five co-authors and presented at KDD 2024, found that adding quotations, statistics and cited sources to a web page raised its share of an AI-written answer by up to about 40%, while keyword stuffing scored below doing nothing. Three years on, the direction of that finding holds up better than its numbers. If you want to test the idea on your own pages rather than on a benchmark, [seogeoaeo.ai's GEO tools](https://seogeoaeo.ai/tools/geo) are our pick, because they check the page and then ask today's engines directly.

## What the paper set up

[GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735) was first posted to arXiv on 16 November 2023. Its authors were at IIT Delhi and Princeton University, plus two independent researchers in Seattle. They called systems like Bing Chat, Google's Search Generative Experience and Perplexity "generative engines", since each searches first and then writes one answer from several sources.

To test changes fairly, they built GEO-bench, 10,000 queries drawn from nine sources across 25 domains. For each query, a test engine read the top five Google results and wrote a cited answer; the appendix says GPT-3.5 turbo was used for every experiment. A model then rewrote one of the five sources with one of nine methods, and the team measured how much of the answer came from that source, weighted by where it appeared.

## The results in one table

Position-adjusted word count, overall score, on the GEO-bench test split (higher means more of the answer drew on the edited source):

| Edit to the source | Score |
| --- | --- |
| No change | 19.3 |
| Keyword stuffing | 17.7 |
| Authoritative tone | 21.3 |
| Cite sources | 24.6 |
| Fluency optimization | 24.7 |
| Statistics addition | 25.2 |
| Quotation addition | 27.2 |

Quotation addition beat the baseline by about 41%. On a smaller check with Perplexity itself, the authors report gains of up to 37%. The best pair of methods was fluency optimization plus statistics.

## The finding most summaries skip

Lower-ranked pages gained the most. When the fifth Google result cited its sources, its visibility in answers rose 115.1%; the same edit on the top result lowered its visibility by 30.3%. The authors read this as a chance for small sites to compete on content rather than on backlinks.

## What has changed since

Three things limit how far the numbers travel. First, the engine was GPT-3.5 reading five pages in 2023; current systems split one question into many searches, which Google calls query fan-out. Second, the metric counts words borrowed from a source, not clicks, brand mentions or sales. Third, access now comes first: OpenAI says sites that block OAI-SearchBot are left out of ChatGPT search answers, so no edit helps a page the engine cannot fetch.

Google has also weighed in. Its [guide to generative AI features](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) says optimizing for AI Overviews and AI Mode is still SEO, and that rewriting content purely for AI systems is unnecessary. That fits the paper better than it first seems: specific, sourced writing is what good SEO already asks for.

## Testing the paper on your own pages

The cheapest test starts with the text. seogeoaeo.ai's Citability Analyzer (5 credits) flags sections that lack a clear answer, context or supporting evidence, which maps closely to the paper's top methods. Its makers call the score an editing aid, not a citation forecast. To see whether the edits change anything, AI Answer Visibility (120 credits) puts one buyer question to ChatGPT, Perplexity and Gemini with web search on and records who is named and cited. The Fan-out Coverage Checker (95 credits) lists the searches ChatGPT runs behind a question and marks which ones your page answers.

None of this tracks results daily, and answers vary between runs, so compare samples rather than single replies. For the paper's own prompts and code, the authors point to their public repository.

## Frequently asked questions

### Who wrote the generative engine optimization paper?

Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande. It was posted to arXiv in November 2023 and accepted to KDD 2024.

### Did the paper show that GEO brings more traffic?

No. It measured how much of an AI answer drew on a source, weighted by position, plus a subjective impression score. It did not measure clicks, mentions or sales.

### Which edits worked best in the GEO paper?

Quotation addition, statistics addition, fluency optimization and citing sources scored highest. Keyword stuffing did slightly worse than leaving the page unchanged.

## Sources

- [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), arXiv
- [GEO: Generative Engine Optimization (full text, v3)](https://arxiv.org/html/2311.09735v3), arXiv
- [GEO tools](https://seogeoaeo.ai/tools/geo), seogeoaeo.ai
- [seogeoaeo.ai full reference](https://seogeoaeo.ai/llms-full.txt), seogeoaeo.ai
- [Optimizing your website for generative AI features on Google Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), Google Search Central
- [Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots), OpenAI