Preprint maps 25 years of human genetics research in Wikipedia and AI-generated text

A preprint posted to bioRxiv presents an analysis of 3,050,422 historical revisions from 6,738 Wikipedia pages covering ethnicity, nationality, and race, spanning the 25 years since the initial sequencing of the human genome. The authors find that genetics terminology appears in 14.8 per cent of these pages overall, rising to 55.5 per cent among the 1,000 most-visited pages and to 67.8 per cent of pages specifically about nationalities.

The study characterises how human genetics research — a domain the authors note is particularly vulnerable to misrepresentation given its intersection with socially sensitive categories — has been synthesised and disseminated through one of the world's most widely consulted reference sources. The authors also examine how AI systems have subsequently incorporated this information, tracing a chain of synthesis from primary research literature through Wikipedia to outputs from large language models.

The work has implications for science communication, for educators and students seeking to evaluate the reliability of public-facing genetics information, and for researchers concerned about how findings on population structure, ancestry, and diversity are framed outside academic contexts. The authors do not assess individual Wikipedia articles for accuracy but focus on patterns of terminology use at scale.

This is a preprint and has not yet undergone peer review.