A 2021 study found AI endorsers seen as less distinct

The persona

This entry covers peer-reviewed research about audiences generally, not a single named persona, though the studies it examines used AI-generated or virtual endorsers as their experimental stimuli rather than any one site-tracked character.

What the documents establish

Veronica Thomas and Kendra Fowler's paper, 'Close Encounters of the AI Kind: Use of AI Influencers As Brand Endorsers', appeared in the January 2021 issue of the Journal of Advertising after being posted online in September 2020. Its abstract, as indexed by the journal's own metadata registry, describes two studies finding that AI influencers can produce brand benefits similar to human celebrity endorsers, but that consumers are less likely to see an AI influencer as a unique entity, so a transgression by one AI influencer is more likely to be generalised to all AI influencers than a human celebrity's transgression would be generalised to other celebrities. A related 2022 study, Jan-Philipp Stein, Priska Breves and Nora Anders's 'Parasocial interactions with real and virtual influencers' in New Media & Society, ran a preregistered experiment with 179 participants and found parasocial response did not differ significantly between a human and a virtual influencer overall, though participants rated the virtual influencer as less human-like and less similar to themselves, an effect that suppressed a separate direct advantage the virtual influencer showed.

Character versus company

Both papers study how audiences react to a category of endorser, not to any operator's specific business claims; neither paper is evidence about a particular persona's actual follower engagement or a specific company's earnings. Thomas and Fowler's transgression finding is about how a company might manage reputational risk if 'its' AI influencer misbehaves, a distinct question from whether any single studio's persona has, in fact, committed a transgression; that would require its own documented incident, which neither paper supplies.

What to watch

This entry was compiled from the papers' publisher-registered bibliographic data and abstracts rather than the full paywalled text of either article, so any claim about statistical detail beyond what is quoted here should be checked against the original journal version before being repeated as established fact.

• Does the paper's experimental stimulus resemble the kind of persona a reader is evaluating, or a different format entirely?
• Is a finding about 'AI influencers' as a category being stretched to describe one specific operator's persona?
• Has either finding been replicated in a study this entry has not reviewed?

Read together, these two studies suggest AI and virtual endorsers can match human ones on some measures while remaining perceptibly less human-like on others, a nuanced finding that resists collapsing into either 'audiences can't tell the difference' or 'audiences always know,' and one this entry treats as bounded by each study's own sample and design.