What GEO research actually supports, what the famous “40%” result means, and how to run a defensible visibility study.
What the “up to 40%” result actually showed
The foundational GEO paper tested strategies such as adding quotations, statistics, citations, and improving fluency across GEO-bench. Its widely repeated “up to 40%” result refers to relative improvement on a generative visibility metric within the study’s experimental setting.
The evidence is meaningful but bounded. Relevant documents had already been supplied within a fixed source context. The result does not establish that a page will receive 40% more organic retrieval, that cross-platform traffic will grow by 40%, or that conversions will rise by the same amount.
Always ask which stage of the visibility pipeline a result actually measures.
Separate strong evidence from plausible practice
Martinez’s critical survey concludes that the strongest causal evidence currently concerns content already present in a generative engine’s context. Relevance and context position are among the more reproducible factors. Extractable evidence—definitions, verified statistics, comparisons, quotations, and procedural detail—often correlates with or improves use in particular settings.
What is not yet strongly established is a universal intervention that creates durable, longitudinal, cross-platform gains in organic AI discoverability, traffic, and conversion. Platform behavior changes, source overlap can be low, and repeated runs can vary. Responsible GEO strategy should label observational patterns as observational and avoid turning correlations into guaranteed tactics.
A defensible first measurement exercise
Select one commercial category and write five unbranded prompts: broad discovery, problem/solution, use case, comparison, and decision criteria. Run them on at least two AI search systems. Record the engine, exact prompt, date, search activation, brands mentioned, first recommendation, citation count, cited domains, own-domain sources, third-party sources, mention position, and any obvious factual errors.
Repeat the prompts and use close paraphrases. Then compare: Do the same brands appear? Do the same domains appear? Does the most-cited brand also receive the first recommendation? Does a more specific use case change the shortlist?
Measure the system before attempting to optimize it.
A baseline makes later changes interpretable. Without one, apparent improvement may be ordinary run-to-run variation.
Citation is not traffic, and traffic is not conversion
Answer participation, citation, referral traffic, and conversion belong to different stages. A citation may influence a user without receiving a click. A click may produce engagement without revenue. A recommendation may cause a branded search later rather than an immediate referral.
Measure each outcome directly when possible. Keep prompt-level evidence for AI visibility, analytics for referrals and engagement, and business systems for leads, subscriptions, or sales. The arrows between those systems are hypotheses until observed.
Questions about this topic
Does GEO improve visibility by 40%?+
The foundational study found up to roughly 40% relative improvement on specific visibility metrics in its experimental setup. It did not prove a universal 40% lift in retrieval, traffic, recommendations, or sales.
How many times should an AI visibility prompt be tested?+
There is no universal number, but repeated runs and prompt paraphrases are important because generative responses vary. Record the model, mode, date, geography, and exact wording where possible.
Are FAQ sections proven to increase AI citations?+
No universal causal rule has been established. FAQs are useful when they answer real reader questions clearly, but the format itself should not be treated as a guaranteed GEO tactic.
What is the most important GEO metric?+
It depends on the objective. Unknown brands may prioritize unbranded discovery; publishers may examine citation and referrals; commercial teams may care most about accurate recommendation and conversion. Keep stages separate.
References
Sources are listed in APA 7 style. Preprints are identified as such and should not be treated as peer-reviewed findings unless separately published.
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., & Narasimhan, K. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5–16). Association for Computing Machinery. https://doi.org/10.1145/3637528.3671900
- Martinez, O. (2026). Optimizing visibility in generative engines: A critical survey of generative engine optimization (2023–2026) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.14035
- Zhang, K., He, X., & Yao, J. (2026). From citation selection to citation absorption: A measurement framework for generative engine optimization across AI search platforms [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2604.25707
- Kumar, P. (2026). Generative engine optimization at scale: Measuring brand visibility across AI search engines [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2606.20065