MOSCOW, 20 January 2026. The VCIOM Analytical Center presents the findings of a survey conducted among online users to understand whether they can differentiate between AI-generated and human-generated content.

Neural networks anywhere

Over the recent five years neural networks have been developing extremely fast: AI-generated content is actively introduced into business processes and into everyday life of the online users (based on “VCIOM” data, 73% of respondents have used neutral networks).

Loss of uniqueness

Artificial intelligence not only changes the role of intellect but also blurs the line between machine and human in creating various content (image, video, audio, text). Authorship, as well as responsibility, is not clear, and the content is often used by malicious users for personal gain. As deep fakes keep growing it is important to learn to critically perceive any content in the digital world. It is very likely that marking AI-generated content, which is being discussed at a national level, will soon become legally mandatory. 

AI or real?

One can hardly differentiate real content and AI-generated. On average, in six out of ten cases online users correctly identified the AI-generated content. Only 4% could detect the author when presented with six pairs of images; 3-5 images were identified correctly by 70% of users; 1-2 images, by 25%.  Out of two pairs of videos being tested the authorship of both videos was spotted correctly by 59% of respondents; every tenth respondent was wrong (9%). The most difficult type of content is text (four-line poems were presented); one-third of respondents gave correct answers (34%); 18% gave wrong answers.

Russian nationwide online “VCIOM-Online” survey was conducted December 18-19, 2025. A total of 1,625 respondents aged 18 and older took part in the survey. Survey method: formalized questionnaire, probability-based online panel. Panel participants are recruited through “Sputnik” daily CATI telephone survey where a sample phone is selected by RDD from a complete list of telephone numbers in use in Russia. The data are weighted for social-demographic characteristics. The margin of error at a 95% confidence level does not exceed 3.1%. In addition to sampling error, minor changes to the wording of questions and different circumstances arising during the fieldwork can introduce bias into the survey.