Is There Academic Backing for Tactics to Get Cited by AI?—Reading GEO Research in a Medical Context

Regarding "tactics to make it easier to get cited by AI," academic research grounded in actual measurement already exists. In this article, we organize the findings of two representative studies and consider how they can be read in the context of information delivery for medical institutions.

GEO (Generative Engine Optimization) is the optimization concept that aims for one's own information to be accurately cited and referenced when a generative AI composes its answers.

Known as the first systematic study is the research that proposed the concept of GEO (Aggarwal et al., KDD 2024). This study reported through experiments that adding clear citations, statistics, and sources to a web page can improve visibility on generative engines, while conventional SEO techniques such as keyword stuffing have little effect.

A newer large-scale comparative study (Koudas, 2025) compared how AI search and conventional Google search select sources across multiple industries and languages, and reports the following differences. First, AI search systematically prioritizes earned media—authoritative mentions by third parties—over a brand's own site or social media. Second, AI search services differ greatly in the diversity of referenced domains, the freshness of information, cross-language stability, and sensitivity to the phrasing of the question. Third, there is a bias favoring major brands, and niche operators need a strategy to overcome it.

Translating these findings into the context of medical institutions, they can be read as follows.

The finding that pages with statistics and sources are more likely to be cited means, in medicine, "making public data and primary information explicit." Presenting verifiable facts with sources attached, rather than claims of effectiveness or superiority, also aligns in direction with the thinking of the Medical Advertising Guidelines.

The prioritization of earned media suggests that enriching your own clinic's site alone is not enough. What is questioned is whether your clinic's information is accurately described in the rosters of academic societies and public institutions, in regional medical information, and on third-party media.

The differences between engines and between languages, and the sensitivity to phrasing, mean that you cannot judge "how you look to AI" from a single question. Observation across multiple models, multiple question patterns, and different points in time is necessary. This is also the thinking that this site takes as the premise of its verification.

What remains for niche operators as a counter to the major-brand bias is the primary data and specialization that the majors do not have. The accumulation of verifiable information rooted in a region and a clinical field is thought to be one of the few paths that compensates for the disadvantage of scale.

Finally, we also state the limits. These studies are chiefly experiments in English-speaking environments, and whether the same effect size holds in the Japanese-language medical domain has not been verified. That verification is precisely the domain this site will fill going forward through actual measurement.