A new academic study reveals that audiences engaging with virtual influencers — digitally synthetic social media personas — exhibit measurably different psychological and emotional patterns in their discourse compared to audiences of human influencers, suggesting that artificiality itself reshapes not merely what people say online, but the underlying structure of how they say it.

The research, posted to the preprint server arXiv under the identifier 2603.24410, was conducted by analyzing YouTube comments across three matched pairs of virtual and human influencers. Researchers employed a two-layer analytical framework combining Formal Concept Analysis (FCA) and association rule mining to map the co-occurrence of sentiment, Big Five personality cues, and topic tags in weekly comment data.

The findings present a striking structural divergence. Human influencer discourse, the study found, tends to consolidate around a single emotionally regulated mode — anchored by low neuroticism and consistent positivity, which researchers describe as a 'stability-centred regime.' Virtual influencer audiences, by contrast, operate across three structurally distinct discourse modes. One of these, an appearance-focused discourse cluster, was largely absent from human influencer comment sections despite appearing at roughly equal frequency in raw terms — a distinction the authors argue would be invisible to conventional frequency-table or survey-based analysis.

Perhaps most notable for broader cultural conversations around digital wellness, the study found that virtual influencer contexts were associated with negative sentiment in psychologically sensitive domains, specifically mental health, body image, and what the researchers term 'artificial identity.' Human influencer contexts did not exhibit the same pattern in those topic areas.

'Virtuality reshapes not just what audiences say, but the underlying grammar of how signals co-occur in their reactions,' the authors wrote in the abstract.

The research positions FCA — a mathematical technique from knowledge representation — as a viable tool for nuanced social media discourse analysis, an application that has not been widely explored in influencer studies. Previous research in this space has largely relied on surveys or aggregate engagement statistics, methods the authors argue capture surface-level content but miss structural co-occurrence patterns.

The implications touch on ongoing debates in media studies and platform governance about the responsibilities of brands and platforms deploying virtual influencers, particularly when those synthetic personas attract audiences who may be processing questions of identity, appearance, and mental health differently than they would around human creators.

The preprint has not yet undergone peer review. Source: https://arxiv.org/abs/2603.24410