Research · 10 October 2026 · arXiv preprint
What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO)
Selected by Luis Da Silva
Before an AI system can cite a page, it usually has to find it. This recent preprint follows commercial questions over time and separates retrieval, citation and brand mentions. Its results challenge the idea that adding a standard set of page features reliably creates visibility. Matching the questions people ask appears more informative than many generic recommendations. For a small firm, the practical response is to publish specific, useful explanations and observe whether engines retrieve them. The study is observational, limited to English US markets and not yet established through peer review. It offers a useful challenge, rather than a universal recipe.
Study and evidence
A longitudinal observational study of retrieval, citations and brand mentions, with API replays and controlled subexperiments.
How to read it
Recent preprint, 10 October 2026; peer review has not been established. The DOI is marked pending registration. English US commercial questions and API replays do not validate Portuguese sector performance or a causal effect on demand.
Why it matters
Our editorial interpretation: It connects Luis's communication work with a measurable distinction between being known, being found and being cited. For longevity and forestry, develop a small bilingual set of real buyer questions and publish evidence-led answers without treating schema or an llms.txt file as a demand-generation guarantee.
Luis Da Silva, Partner & Chief Marketing Officer
About the publisher
Mahe Chen, Nick Koudas. arXiv preprint.
Useful for
Communication and content teams working across longevity and forestry. The sector applications are our inference and require local testing.
Credit and license Mahe Chen, Nick Koudas. 2026. What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO). arXiv preprint. arXiv:2610.15033v1, submitted 2026-10-10. DOI registration pending: 10.48550/arXiv.2610.15033. Author-version licence: CC BY 4.0, https://creativecommons.org/licenses/by/4.0/. Original study and licence: https://arxiv.org/abs/2610.15033v1. General Value supplies an original editorial summary. Unchanged arXiv v1 preprint, 10 October 2026, hosted under CC BY 4.0. This is not a peer-reviewed proceedings edition; use the arXiv record while DOI registration is pending.
Original research belongs to the credited authors. The bilingual summary and proposed application are General Value's editorial interpretation, not findings tested for our firm.