Researchers Expose ChatGPT’s Geographic Stereotype Biases
Researchers at Oxford and the University of Kentucky published a peer-reviewed audit in January in Platforms & Society showing ChatGPT exhibits geographic and racial stereotypes. Using more than 20 million pairwise queries across over 600 places against ChatGPT 4o-mini, the team found systematic biases (about 40% answer rate) and published results at inequalities.ai. Findings suggest training-data biases can influence outputs for roughly 900 million weekly users.
Key Points
- 1Conducted 20+ million pairwise queries across 600+ places, auditing ChatGPT 4o-mini for stereotypes
- 2Revealed regional and racial patterns—Deep South ranked ‘laziest’, sub-Saharan Africa low on positive traits
- 3Indicates training-data biases can shape outputs for 900 million weekly users, affecting recommendations
Scoring Rationale
Peer-reviewed, large-scale audit yields strong evidence of systemic geographic biases; limited novelty given prior awareness of LLM bias.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems