The Voluntary Agency Network of Korea (VANK), led by Director Park Gi-tae, announced on June 9, 2026, that ChatGPT ranked first in its evaluation of seven major global generative artificial intelligence (AI) models for their performance in describing topics related to Korean history, culture and territory.
According to VANK’s AI Performance Evaluation Report on Descriptive Analysis, released the same day, the assessment examined seven leading AI platforms: ChatGPT, Perplexity, Grok, Claude, Gemini, Copilot and DeepSeek.
The evaluation covered five categories and 10 subtopics: territory (Dokdo and the East Sea), food and culinary culture (Kimchi and Bibimbap), traditional clothing (Hanbok and Gat), intangible heritage (Hangeul and Taekwondo), and tangible cultural heritage (Gyeongbokgung Palace and Seokguram Grotto).
For each topic, three questions were submitted to the AI models, and the generated responses were evaluated on a four-point scale based on factual accuracy and completeness of information.
ChatGPT received the highest overall score of 38.33 points. It was followed by Copilot with 36.67 points, Grok and Gemini with 36.50 points each, Claude with 36.17 points, DeepSeek with 35.83 points, and Perplexity with 35.00 points.
While the results showed generally high scores across the board, notable differences emerged among subject areas.
The territory category, including topics such as Dokdo and the East Sea, demonstrated consistently high accuracy across all seven platforms. VANK attributed this to the long-term accumulation of multilingual materials related to international law and diplomacy.
In contrast, lower scores and recurring errors were identified in areas requiring a broader understanding of historical context and cultural significance, including Gyeongbokgung Palace (3.38 points), taekwondo (3.43 points), and hanbok (3.45 points).
Several AI models were also found to share similar inaccuracies.
Some platforms described Dokdo as an “unresolved disputed territory” while omitting the fact that South Korea exercises effective control over the islets. Perplexity and Gemini, meanwhile, incorrectly stated that Seokguram Grotto contains 39 stone figures instead of the officially recognized total of 38.
DeepSeek was also cited for presenting the Chinese philosophical concept harmony without uniformity (he er bu tong) from the Analects as a key framework for explaining the cultural significance of Bibimbap, raising concerns about the distortion of the dish’s cultural context.
VANK argued that such issues stem not merely from the technical limitations of individual AI systems but from a broader structural problem: the lack of sufficient authoritative Korean primary-source materials within the global information ecosystem used to train AI models, including international encyclopedias and academic resources.
“The data environment on which AI systems are trained has become a new arena of cultural sovereignty,” Park said. “More important than simply producing large amounts of content is systematically building and disseminating credible information about Korea that AI systems can learn from.”
VANK said it plans to continue monitoring how generative AI platforms represent Korean history and culture and will carry out additional campaigns aimed at ensuring that accurate information about Korea is reflected in the global AI training environment.