The Princeton GEO Study: How Statistics, Citations & Quotes Boost AI Visibility

Published August 7, 2026 · Adam Rock

The study in one paragraph

"GEO: Generative Engine Optimization" (Aggarwal et al., KDD 2024; read it at https://arxiv.org/abs/2311.09735) built GEO-bench, a benchmark of 10,000 queries drawn from nine source datasets, and evaluated nine content optimization methods against generative AI engines. It measured how much each method changed a source's position-adjusted share of the generated answer. Headline result: these methods can boost visibility by up to 40%. Which methods work, and when, matters more than the headline.

The nine methods, as the paper groups them

Top performers

Cite Sources, Quotation Addition and Statistics Addition. These three achieved a relative improvement of 30-40% on the Position-Adjusted Word Count metric and 15-30% on the Subjective Impression metric.

Other high-performing methods

Fluency Optimization, Easy-to-Understand language, Authoritative tone and Technical Terms.

Non-performing

Keyword Stuffing and Unique Words. Adding keywords did not win answers; in the deployed test it scored below the no-optimization baseline.

The domain matters

The paper's Table 3 shows the winning method depends on the topic:

Authoritative tone wins in debate, history and science.
Quotation Addition wins in people and society, explanation and history.
Statistics Addition wins in law and government.
Cite Sources wins across statement and fact queries, law and government, and business, science and health.

There is no single universal tactic. The right intervention depends on what you are writing about.

The caveat to read before quoting "40%"

40% is a headline ceiling, not an expectation. The paper reports the top three methods at 30-40% on Position-Adjusted Word Count, and 15-30% on Subjective Impression. In its deployed-engine experiment on Perplexity.ai, the best-performing methods improved over baseline by 22% on Position-Adjusted Word Count and 37% on Subjective Impression.

One scope note that matters for our category: the study measures a source's share of an answer it already appears in. It does not promise that any AI platform will cite you.

What this means in practice

Add real statistics: original numbers, benchmarks, percentages.
Cite primary sources: link the paper or the dataset, not a summary of it.
Quote named experts.
Match the method to the topic. There is no universal winner.
Do not keyword-stuff. The paper found it non-performing.
Read it yourself: https://arxiv.org/abs/2311.09735

FAQ

Does adding statistics help AI cite my content?

Yes, per the paper. Statistics Addition is one of the top three methods (with Cite Sources and Quotation Addition), achieving a 30-40% relative improvement on the Position-Adjusted Word Count metric.

What is the single most effective GEO tactic?

There is not one. The paper found the winning method depends on the domain: authoritative tone in debate and science, quotations in people and society, statistics in law and government.

Is the "40%" visibility boost a guarantee?

No. It is a headline ceiling from the benchmark. The top three methods measured 30-40% on Position-Adjusted Word Count, and the deployed-engine experiment on Perplexity.ai showed a 22% improvement for the best methods. Treat it as a ceiling, not a promise.

Frequently Asked Questions

Does adding statistics help AI cite my content?

Per the paper, yes. Statistics Addition is one of the top three methods (with Cite Sources and Quotation Addition), achieving a 30-40% relative improvement on the Position-Adjusted Word Count metric.

What are the most effective GEO techniques?

There is no single winner. The paper found the best method depends on the domain: authoritative tone in debate and science, quotations in people and society, statistics in law and government. Keyword Stuffing and Unique Words were non-performing.