GEO (Generative Engine Optimization)

GEO Statistics: Generative Search Research Benchmarks

Explore generative engine optimization research with original GEO benchmark results, study dates and clear limits. Learn what AI visibility scores can and cannot tell you.

Research compilation by Online Marketing HQLast updated: Last verified:
Download the data (CSV) ↗6 source-checked findings

GEO benchmark: position-adjusted word-count scores

Unmodified19.3
Quotation addition27.2
Statistics addition25.2
Source citation24.6
Keyword stuffing17.7
Absolute scores from table 1 of the 2024 paper, averaged across the authors' experimental runs. These are visibility metrics, not traffic growth percentages.

Read our research methodology and data reuse terms. Each source retains its own rights.

The statistics at a glance

Swipe the table to see each value, year and original source.

GEO Statistics: Generative Search Research Benchmarks: values, data periods and original sources.
StatisticValueData periodSource
#01Baseline position-adjusted word-count scoreTable 1 overall score for unmodified sources; not a percentage increase.19.32024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗
#02Quotation-addition position-adjusted word-count scoreTable 1 overall score for this experimental method; not a guaranteed production result.27.22024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗
#03Statistics-addition position-adjusted word-count scoreTable 1 overall score in the research engine; not a measured conversion or traffic lift.25.22024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗
#04Source-citation position-adjusted word-count scoreTable 1 overall score for citing sources within the experimental setup.24.62024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗
#05Keyword-stuffing position-adjusted word-count scoreTable 1 overall score; a result of this experiment, not a universal ranking penalty.17.72024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗
#06Queries in the full GEO-bench datasetSection 3.2: 8,000 training, 1,000 validation and 1,000 test queries; not 10,000 independently evaluated business websites.10,0002024 paper, version 3Aggarwal et al., GEO: Generative Engine Optimization, table 1 / section 3.2 ↗

GEO statistics are research results, not traffic forecasts

Generative engine optimization, or GEO, concerns how content appears in generated answers. The original GEO paper, version 3, offers a controlled benchmark for comparing content changes. Its 2024 experimental results are useful historical evidence, but they do not describe the present behaviour of every commercial AI search product.

The table reports absolute position-adjusted word-count scores from the paper. A score such as 27.2 is not a claim of 27.2% more visitors. The metric captures prominence within an answer, combining cited text and its position. Keep that distinction when using the chart in a presentation or brief.

Turn the research into an editorial test

Start with questions your customers actually ask. Select a manageable set that covers definitions, comparisons and practical decisions. Record the engine, date, wording and location assumptions for each observation. Repeated runs help reveal whether a result is stable or merely one generated answer.

Review the source pages before changing them. Can a reader identify who is making a claim, what evidence supports it and when it was measured? Does a citation link to the underlying study? Is the answer understandable without removing a crucial qualification?

Treat changes as hypotheses. You might test whether a clearer explanation, a sourced table or a more useful comparison helps readers and makes the content easier to reference. Do not add irrelevant quotations or numbers just to imitate an experimental technique.

Keep visibility, visits and business outcomes separate

An AI answer may mention a brand, cite a page or send a visitor. Those are different events. Record them separately, then examine whether referred visitors reach relevant content and complete useful actions.

Our AI search statistics cover observed search behaviour and help explain why a citation should not be treated as a visit. This GEO guide focuses on experimental content visibility so the two evidence types remain distinct.

A practical reporting sheet can track the question, response date, citation URL, accuracy of the cited claim and referral activity when identifiable. Preserve examples so reviewers can check the interpretation later.

Questions about GEO evidence

Does the research prove a guaranteed increase?

No. Results depend on the study's engine, sources, prompts and scoring method. Commercial systems and their retrieval behaviour can change.

Should we publish more statistics to appear in AI answers?

Publish a statistic when it helps answer the reader's question and you can verify it. Unsupported precision makes a page less dependable, even if it looks authoritative.

Is this new 2026 research?

No. The study version used here is from 2024. The page's verification date tells you when we checked the source.

Continue with the GEO source-trail guide, editorial review checklist, or GEO topic hub.

MAKE YOUR SOURCES EASY TO TRACE

How to cite this page

Choose a format and copy the reference. For numerical claims, always check the original linked source.

Online Marketing HQ. (2026, November 18). GEO Statistics: Generative Search Research Benchmarks. https://www.onlinemarketinghq.co/stats/geo-statistics

Chat on WhatsApp