Content Strategy

    Statistics, Quotes and Citations: What the GEO Research Found

    A plain-English summary of the GEO research paper: which content changes improved visibility in AI-generated answers, and how to apply the findings.

    Saarthak Srivastav

    Saarthak Srivastav

    · 4 min read

    Key takeaways

    • The GEO paper (Aggarwal et al., KDD 2024) tested nine ways of editing content to see which made it more visible in generative engine answers.
    • Adding quotations, statistics and citations to sources were among the most effective changes, improving visibility by up to roughly 40% in the authors' benchmark.
    • Keyword stuffing, the classic SEO shortcut, did not help.
    • Effects varied by topic, and lower-ranked sources gained the most from some changes.
    • Treat it as directional evidence for substance and sourcing, not a formula.

    The most cited piece of research on optimizing for AI answers is the GEO paper, and its main finding is reassuringly practical: content with quotations, statistics and citations to sources was more visible in AI-generated answers, while keyword stuffing didn't help. This post summarizes what the researchers did, what they found and how to apply it without over-reading a lab result.

    The paper in brief

    • Title: GEO: Generative Engine Optimization
    • Authors: Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande
    • Published: first posted in November 2023; presented at KDD 2024
    • Contribution: defined "generative engines" (systems that retrieve sources and generate answers), proposed ways to measure a source's visibility in those answers, built a benchmark of queries across many domains (GEO-bench), and tested content-editing methods.

    The term "generative engine optimization" comes from this paper. For how it relates to AEO, see AEO vs GEO vs LLMO.

    What they tested

    The researchers took web sources used to answer queries, rewrote them using different methods, and measured how much more or less visible each source became in the generated answers. The methods included:

    1. Authoritative: making the tone more confident and authoritative
    2. Keyword stuffing: adding more query keywords
    3. Statistics addition: adding relevant statistics
    4. Cite sources: adding citations to credible sources
    5. Quotation addition: adding quotes from relevant experts or sources
    6. Easy-to-understand: simplifying language
    7. Fluency optimization: improving how smoothly the text reads
    8. Unique words: adding distinctive vocabulary
    9. Technical terms: adding domain-specific terminology

    What they found

    • Quotation addition, statistics addition and citing sources were among the top performers. The authors report visibility improvements of up to around 40% in their benchmark for the best methods.
    • Fluency and readability improvements also helped, though less.
    • Keyword stuffing performed poorly, with little or no benefit.
    • Effects varied by domain. Some methods helped more for certain topics than others. For example, the paper reports statistics being particularly useful in some fact-heavy domains and quotations in others.
    • Lower-ranked sources gained more from some methods, suggesting these edits can help content that isn't already at the top of search results.

    The researchers also ran experiments on Perplexity, a real-world answer engine, and reported similar directional improvements.

    How to read these results

    The study is useful, but keep three caveats in mind:

    1. It's a research setup. Real AI products change frequently and use different retrieval and ranking.
    2. "Visibility" was measured in a specific way: roughly, how much of the generated answer drew on a given source and how prominently.
    3. It tested edits to existing sources, not whole content strategies, third-party coverage or brand signals.

    So treat it as directional evidence: substance and sourcing help; keyword tricks don't. That matches what we see in practice.

    How to apply it

    Add statistics, honestly

    • Use numbers that answer the reader's question: ranges, averages, thresholds, changes over time.
    • Attribute every number to a named source and date.
    • Prefer primary sources. Better still, publish your own data. See original research as an AEO asset.
    • Never invent figures. A fabricated statistic that AI repeats is a liability.

    Add quotations

    • Quote named practitioners, customers or experts with relevant experience.
    • Use quotes that add perspective or evidence, not filler praise.
    • Attribute clearly with name and role.

    Cite your sources

    • Link to the primary source for claims, standards and data.
    • Cite official documentation for product and platform facts.
    • Citations make your passage more trustworthy to quote.

    Improve clarity

    Skip keyword stuffing

    Write naturally in your buyers' language. Repeating phrases doesn't help AI answers and makes content worse for readers.

    Before and after

    Before:

    Email marketing is very effective for ecommerce brands and delivers strong results. Many businesses see great returns.

    After:

    Email remains one of the highest-ROI channels for ecommerce. In [Source]'s [year] benchmark of [n] online stores, automated flows such as abandoned-cart and welcome series generated [X]% of email revenue from [Y]% of sends. "Flows do the heavy lifting; campaigns keep you top of mind," says [Name], [role] at [Company].

    The second version has a statistic with a source, a quotation with attribution and a specific claim. It's more useful to readers and, according to the GEO research, more likely to be used in AI answers. (Fill the brackets with real figures and sources; never invent them.)

    The bottom line

    The GEO research confirms what good editors already know: specific, well-sourced content beats keyword-padded content. Add relevant statistics with sources, quote real experts, cite primary references and write clearly. Treat the paper's numbers as directional, and focus on making every page more informative than the alternatives. Our guide to writing content AI assistants cite puts it all into a checklist.

    FAQ

    Frequently asked questions

    What is the GEO paper?

    GEO: Generative Engine Optimization is a research paper by Pranjal Aggarwal and colleagues, first posted in 2023 and presented at the KDD conference in 2024. It introduced the term generative engine optimization and a benchmark for testing how content changes affect visibility in AI-generated answers.

    What did the GEO research find?

    That adding quotations, statistics and citations to authoritative sources were among the most effective ways to increase a source's visibility in generative engine answers, while keyword stuffing was ineffective. Effects varied by domain.

    Does the GEO research apply to ChatGPT and Perplexity?

    The experiments used generative engine setups built for research, plus a test on Perplexity. Real products change often and work differently, so treat the findings as directional rather than exact.

    Should I add statistics to every page?

    Add them where they genuinely inform the reader and come from credible, named sources. Fabricated or irrelevant statistics damage trust and can be repeated incorrectly by AI.

    • GEO
    • research
    • statistics
    • quotations
    • citations
    Saarthak Srivastav

    Written by

    Saarthak Srivastav

    AEO & GEO, AEO-Rank

    Saarthak Srivastav writes and reviews articles on the AEO-Rank blog, drawing on what the AEO-Rank team sees tracking AI answers across ChatGPT, Gemini, Claude, Perplexity and Grok.