The swift progress in machine learning algorithms, artificial intelligence, and interactive immersive media technologies has led to the introduction of computer-generated imagery on Instagram. This feature, so-called “human-like virtual influencers (VIs)", has revolutionized the way people interact with technology. Using a combination of cutting-edge AI technologies, in a novel application of computer vision algorithms, and large language models to extract the content posted by two popular human-like VIs on Instagram, the present study is the first to categorize and classify types of human-like virtual-influencer-generated content. Quantitative methods, such as partial least squares structural equation modeling (PLS-SEM), were used to examine the impact of human-like virtual-influencer-generated content on consumers' willingness to follow as well as purchase intentions. The information was gathered from 650 Thai customers. The findings showed that consumers' willingness to follow and purchase intentions were significantly influenced by the positive effects of emotional appeal content, which includes relational, entertaining, positive emotion, and negative emotion content. These effects outweighed those of rational appeal content, such as informative and remunerative content, as well as authenticity appeal content. Meanwhile, disclosing sponsored content had no effect on consumers' willingness to follow. The theoretical underpinnings of uses and gratifications (U&G) theory, parasocial relationships and Richins' hierarchical model of emotions are confirmed and expanded upon in this work, and the suggested inclusive approach also significantly advances the expanding corpus of research on VIs. Our research also provides a contribution to the recent literature on human-like VI marketing.