Most respondents mistake AI videos for real footage in new survey

An online survey conducted from October 15 to 19, 2026, revealed that 64.2% of 1,000 respondents had mistaken AI-generated videos for real footage, while 80.0% demanded mandatory disclosure for news and reporting content.

A survey of 1,000 men and women aged 20–59 who watched online videos within the past three months found that 64.2% of respondents had experienced mistaking an AI-generated video for real footage, only to discover later it was made by AI. Only 45.7% of respondents said they felt confident in their ability to distinguish between AI videos and real footage. The online survey was conducted from October 15 to 19, 2026.

Regarding recent consumption, 80.8% of respondents had watched AI videos within the last three months. The most common type encountered was SNS videos for fun or entertainment at 46.2%, followed by product or service advertisements at 34.4%, and dramas or movies at 29.4%. Confidence in distinguishing AI was highest among those in their 20s at 55.6% and lowest among those in their 50s at 41.2%. However, the experience of mistaking AI for real footage exceeded 60% across all age groups: 68.0% for those in their 20s, 63.2% for those in their 30s, 63.2% for those in their 40s, and 62.4% for those in their 50s.

Among the 642 respondents who had mistaken AI for real footage, 42.8% said they subsequently began to suspect whether other videos were also made by AI. Additionally, 37.2% felt the experience was novel or interesting, while 26.0% reported a decrease in trust toward the channels or accounts that posted the videos. These results suggest that, regardless of confidence in identifying AI, experiences of being misled can influence the judgment of authenticity for other content.

Acceptance of AI video varied depending on the content type. For SNS videos for fun or entertainment, 60.5% said they could accept AI production; 51.5% for educational or informational videos; and 48.1% for advertising videos. The acceptance rate was lowest for news and reporting videos at 32.4%. Conversely, the demand for mandatory disclosure of AI production was highest for news and reporting at 80.0%, followed by educational or informational videos at 75.2%, and advertising videos at 73.8%. In the case of educational or informational videos, where acceptance was over half but demand for disclosure was high, it appears that the permission to use AI and the requirement to disclose its use function as separate criteria for judgment.

Regarding corporate AI advertising, 46.3% responded that it would be acceptable if the fact of AI production was disclosed, while 25.8% stated that even with disclosure, they would trust such ads less than those featuring real footage. Among the latter group, 30.6% were women and 21.0% were men. Acceptance of advertising videos was 53.6% for those in their 50s, 51.6% for those in their 40s, and 42.4% for those in their 20s. While 51.6% responded that AI videos could offer new and creative expressions, 60.2% responded that they lack emotion or authenticity. Furthermore, 56.3% replied that as AI videos increase, the value of videos filmed by humans will rise.

The primary concerns regarding the spread of AI videos were fake news and public opinion manipulation at 52.3%, followed by exploitation for financial fraud and voice phishing at 38.3%, difficulty in distinguishing AI from real footage at 37.3%, and exploitation for sex crimes or defamation at 34.9%. These figures represent the combined totals of the top three concerns cited by respondents.

Jo Min-hee, CEO of PMI, stated that there is a coexistence of positive perceptions regarding the potential use of AI video and concerns regarding authenticity and reliability, noting the need to transparently disclose production facts according to the purpose of the content. As the survey showed a clear difference in acceptance between entertainment and news, she suggested that standards for announcing AI production should be established by considering the nature of the content and the level of information reliability expected by users.

Read the original (Korean)