JMIR Publications

ABSTRACT

Background:

Human-like virtual profiles (VPs) are AI-generated personas on social media that mimic human appearance and behavior. Prior work on virtual influencers largely predates the advent of hyper-realistic generative AI tools, leaving limited understanding of how design features such as photorealism, identity consistency, cultural alignment, and content format—including AI-generated video produced with face-swapping apps—affect audience engagement and perception.

Objective:

This study investigates how key VP attributes—photorealism, physical and behavioral consistency, cultural alignment, and the inclusion or absence of human co-presence—relate to engagement patterns across different content formats (static images vs. AI-generated videos) on Instagram.

Methods:

A total of 189 human-like VPs were identified on Instagram. Profiles with fewer than 100,000 followers were excluded, and those exhibiting suspicious follower activity were filtered out. Using a scorecard based on follower counts, growth, and engagement metrics, 25 top-performing VPs were selected for detailed analysis. Profiles with engagement rates above 5% were included in a profile-level analysis, while the top 25 image and 25 video posts per account were evaluated for content features and performance. Audience demographics were assessed using third-party analytics, and engagement was calculated as impressions relative to total followers.

Results:

Followers were predominantly male (≈85%), aged 25–34, and concentrated in India and Indonesia. Engagement was highest for VPs displaying photorealism, internal consistency, and cultural congruence with their audience. Posts featuring human–VP co-presence did not outperform VP-only content. Video posts outperformed static images, with top videos achieving engagement rates between 118.46%–414.34%, compared to 27.52%–59.30% for images. Nearly all high-performing videos were produced with mobile face-swapping tools, indicating a shift toward accessible generative video creation.

Conclusions:

Engagement with AI-generated VPs is primarily driven by visual realism, identity coherence, and cultural alignment, with AI-generated video (especially face-swapped formats) emerging as the most engaging medium. The lack of advantage from human co-presence suggests that audiences may form parasocial bonds with VPs themselves rather than through human association. These findings update earlier research by incorporating modern AI capabilities and highlight implications for the design of trustworthy and psychologically safe virtual agents. The results offer insights into how synthetic media can influence digital engagement, emotional response, and well-being, underscoring the need for ethical standards and transparent practices in AI-generated communication.