Scroll through Instagram long enough and you’ll bump into someone whose abs seem a little too perfect, whose skin catches the light in a way that feels slightly off. Look closer and you may realize she isn’t a person at all. Virtual influencers, computer-generated characters with millions of followers, are increasingly showing up in fashion campaigns, beauty tutorials, and now fitness feeds. The question for brands and their audiences is a simple one: can a fictional character actually convince you to change your behavior?
A pair of studies published in Psychology & Marketing takes on that question by looking at something surprisingly small: the pronouns these digital characters use when they talk to their followers.
Why pronouns matter for fake humans
Karina Sokolova and her colleagues at Paris School of Business set out to understand how virtual influencers position themselves in their posts, and whether saying “I” versus “you” changes how audiences respond. Human influencers routinely share personal stories in the first person, and that kind of self-disclosure tends to build trust. But virtual influencers occupy an odd space. They look like people, sometimes act like people, but they don’t actually have personal experiences. When a computer-generated character says “my journey,” is it believable?
The researchers drew on positioning theory, a framework from narrative psychology suggesting that speakers construct identity through the way they position themselves relative to their audience. Pronouns are one of the most direct tools for doing that. “I” places the speaker at the center. “We” implies belonging to a group. “You” turns the spotlight onto the listener.
What virtual influencers actually say
The first study cataloged the real-world communication habits of 391 virtual influencers active on X (formerly Twitter), analyzing more than 24,000 of their posts. Two coders rated each character’s profile picture on a five-point scale of anthropomorphism, running from abstract logos and objects up through humanized animals, anime-style figures, and finally hyper-realistic human portraits.
The researchers then sorted every post into one of four communication styles: agent-centered (“I”), group-centered (“we”), audience-centered (“you”), or purely informational, with no personal pronouns at all. Informational posts turned out to be the most common, showing up in the feeds of 91 percent of the accounts studied. But the interesting pattern involved the other styles.
Highly human-looking virtual influencers were significantly more likely to use “I” language, casting themselves as protagonists with personal experiences. Less human-looking agents, the abstract logos and cartoon animals, leaned toward “we” language, presenting themselves as part of a collective. Audience-centered “you” messages, meanwhile, appeared across all levels of realism.
The researchers interpret this as evidence that developers make deliberate design choices, matching how their virtual characters look with how they speak. A hyper-realistic character gets a hyper-realistic first-person voice. A logo speaks for a community.
Testing what actually works
The second study asked whether these choices make a difference to audiences. The team built an experiment using Midjourney to generate images of a fictional 25-year-old fitness influencer named Lea. Depending on which version they were shown, participants were told Lea was either a real human on Instagram or a computer-made virtual character. They also saw one of two versions of her caption:
The self-centered version began, “My journey to feeling my best has been full of ups and downs, but I’ve learned to love every part of it.” The audience-centered version reframed the same message: “Your journey to feeling your best will have its ups and downs, but learning to love every part of it is key.”
A total of 263 women recruited through the Prolific platform were randomly assigned to one of the four combinations. They then rated the influencer on credibility, how human she seemed, how confident they felt about following her workouts, and whether they’d actually consider exercising with her videos.
The surprising role of “you”
The researchers expected self-centered messages to make virtual influencers feel more human, since first-person storytelling tends to build intimacy. The data pointed the other direction. Audience-centered messages, the ones using “you,” were linked to higher perceptions of humanness for the virtual influencer. The effect was not statistically significant for the human version of Lea.
Sokolova and her colleagues interpret this as a kind of perspective-taking effect. When a virtual character directly addresses her audience and acknowledges their experience, she comes across as more attentive and sincere, which in turn makes her feel more alive. When the same character talks about her own “journey,” the fictional nature of that journey may quietly undermine the effect.
The perception of humanness then set off a chain reaction. Feeling that the influencer was more human was linked to higher credibility ratings and higher self-efficacy, meaning participants felt more capable of doing the workouts she might recommend. Those two perceptions, in turn, were linked to stronger intentions to actually exercise using her content.
The pronoun choice didn’t affect credibility or exercise intentions directly. Its influence traveled through the audience’s sense of the influencer as a human being.
What this means for brands and creators
For marketers working with virtual influencers, the research suggests a counterintuitive playbook. The instinct to have digital characters mimic human influencers by sharing personal stories may be less effective than shifting the frame toward the audience. “Your journey” appears to work better than “my journey” when the speaker doesn’t technically have one.
The findings also suggest that virtual influencers can motivate behavior beyond shopping. Fitness is one obvious application, but the same logic could extend to public health campaigns, wellness content, or any domain where influencers currently help shape lifestyle choices. The authors point out that a virtual character delivering encouragement about exercise produced measurable effects on intentions to work out, comparable in structure to what human fitness influencers achieve.
There are limits to what the studies can say. Both used English-language content and American audiences, and the experiment was restricted to female participants viewing a female fitness influencer, a choice the researchers explain by pointing to the female-skewing nature of the fitness influencer market. Pronoun choice explained only a small portion of the variance in how human the virtual character seemed, meaning other factors like voice, movement, and backstory almost certainly matter as well. And the study focused on hyper-realistic virtual characters; the results may not extend to cartoon-style or clearly non-human avatars, which the first study showed operate under different communication norms.
One finding worth sitting with: when participants saw a human-realistic image, the overall persuasion process looked broadly similar whether they were told the character was real or virtual. The researchers note that as generative AI tools make virtual characters visually indistinguishable from real people, the line between the two is becoming less meaningful to audiences than the line between characters who talk about themselves and those who talk to you.
