Document Type
Poster
Publication Date
Spring 2026
Abstract
AI influencers are hyper-realistic computer-generated models designed to look and behave like real social media influencers. This study uses Aspect-Based Sentiment Analysis (ABSA), qualitative content labeling, and quantitative regression modeling to analyze how AI influencers on Instagram generate audience engagement. User comments were classified into key aspects including Human Projection, AI Awareness, Controversy, Parasocial Attachment, and Exploitation/Spam, with each associated with positive, negative, or neutral sentiment. Post content types and caption strategies were structured into analyzable features, while engagement metrics including likes, comments, and engagement rate were examined across these categories. A linear regression model was used to estimate the impact of content types on engagement, while interaction analysis evaluated how combinations of content and messaging influenced performance. Results show that audiences largely respond to AI influencers as if they were human, despite awareness of their artificial nature, reflecting strong parasocial behavior. Content characterized by emotional evocation and controversy generated the highest levels of engagement, especially when paired with emotional caption styles. In contrast, promotional content was largely ineffective in generating engagement. However, the regression model explained only a small portion of the variation in engagement, suggesting that additional factors beyond content type also play a significant role. The findings suggest that AI influencers not only replicate but also optimize human engagement patterns through emotionally driven strategies, raising important implications for digital marketing and ethical design.
Recommended Citation
Aslam, Ayesha, "Hacking Human Attention: AI Influencers Aspect-Based Sentiment Analysis and Behavioral Analytics" (2026). IPHS 484: Senior Seminar. Paper 48.
https://digital.kenyon.edu/dh_iphs_ss/48
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.