The pitch for a virtual influencer almost always starts with the same promise: a character who will never miss a deadline, never insult a sponsor, never get caught on a phone video at the wrong party. For a marketing director who has just spent a quarter cleaning up after a human creator, that sounds like a finished argument. It is not. The controlled persona that makes a virtual influencer attractive to brand safety reviewers is also what makes a single misstep harder to recover from, and the 2025 research on the topic is starting to map that asymmetry in detail.
Why “brand safe” became the headline pitch
Most of the agencies selling virtual influencers in 2026 lead with the same three lines: scriptable, scandal-free, scalable. The argument is simple. A CGI persona has no nightlife, no political feed, no late-night DMs, no friends to leak from. Every word she posts has passed through a writer, an art director, and at minimum one legal review. Compared to a human creator with their own thumbs and their own opinions, that is a real reduction in stochastic risk.
This framing is so embedded in trade press that it has become almost invisible. Open any 2025 or 2026 vendor explainer on AI personas and the same three claims will appear in the first 200 words. They are not wrong. They are incomplete.
Where the “zero risk” frame breaks down
Several real incidents complicate the narrative.
In 2019, Calvin Klein released a campaign in which model Bella Hadid kissed the virtual influencer Lil Miquela. The brand received fast and pointed backlash for what critics called queer-baiting: presenting a same-sex moment for visual interest while featuring a model widely understood to be heterosexual. Calvin Klein issued a public apology, acknowledging that the spot could be read as misrepresenting the LGBTQ+ community. The persona involved was synthetic. The reputational hit landed on the actual brand.
The Shudu Gram case sits in a different category. Shudu, created in 2017 by British photographer Cameron-James Wilson, was marketed as the world’s first digital supermodel. By 2018, writers including Bolu Babalola and Vanessa Angélica Villarreal had publicly challenged the project, arguing that a white male creator profiting from a Black-presenting digital model occupied space that real Black models were already struggling to gain. A 2018 Affinity magazine poll, with more than 23,000 votes, found 83.5% of respondents opposed to the model’s premise. Brands that have hired Shudu have hired into that critique each time, and the critique has not aged out.
Cases like these do not contradict the “scriptable, no nightlife” pitch. They show that the harder risk for a virtual influencer is not the creator going off-message. It is the persona itself being read as off-message the moment it appears in a campaign.
The 2025 research on accountability transfer
Until recently, this was mostly anecdote. In February 2025, Northeastern University associate professor Sian Joel-Edgar and colleagues published a study in the Journal of Business Research, summarized by Northeastern, looking at how consumers assign blame when a campaign goes wrong.
The result has implications for any brand counting on “zero risk.” When a human influencer promotes a faulty product or says something incorrect, consumers tend to attribute the error to human fallibility. The blame disperses. When a virtual influencer is involved in the same scenario, blame concentrates on the brand. The persona is perceived as less responsible because it is perceived as less of an agent, so the company behind it absorbs more of the trust damage.
In other words, the controlled persona that reduces the probability of a public scandal can amplify the consequences if a scandal still occurs. Both effects are real, but they are usually marketed as one.
A separate 2025 study published in Frontiers in Communication tested whether virtual influencers can act as effective crisis spokespersons. The researchers ran an experiment with 169 participants comparing human spokespeople with humanized and animated virtual ones. Their finding was nuanced: virtual influencers can help repair brand image, but only when the persona is built with enough mental human-likeness and narrative richness to feel like a credible communicator. A thinly written CGI face delivering a corporate apology does not land.
Five risk categories brands actually face
Pulling these threads together, the practical risk surface for a virtual influencer is broader than the standard pitch suggests. It includes:
The first is creative misread. Campaigns that work as drawings can read very differently as social posts (the Calvin Klein case). With a human model, the actor’s biography and personality often act as soft commentary on the image. A virtual character has no biography that audiences trust, so the image speaks alone.
The second is representational politics. A persona’s race, gender presentation, body type, and accent are all authored choices. Choosing them, and choosing who is behind the keyboard, becomes a campaign question that human casting answers more naturally.
The third is the disclosure surface. Audiences are increasingly aware that the entity replying to comments is not the character. Brands that lean hard on a “real friend” framing risk a backlash when that frame collapses, which is part of why our piece on virtual influencer disclosure treats it as a baseline practice, not a nice-to-have.
The fourth is the team behind the persona. A virtual influencer is typically operated by a studio of writers, 3D artists, and social managers. If anyone in that studio acts in a way that becomes public, the persona inherits it. The persona is “safe”; the humans behind it are still human.
The fifth is the accountability transfer documented in the Northeastern work. When something does go wrong, consumers route blame to the brand more directly than they would with a human creator.
What “brand safe” actually looks like in practice
None of this means virtual influencers should be avoided. The reduction in stochastic risk is real, and for many campaigns it is the right trade. The fix is to treat brand safety as a process, not a property of the persona.
Three habits separate the campaigns that hold up:
Treat the persona as a published character, not as a deniable account. Write a backstory the brand would defend in a press response. Test the character against the same scenarios you would test a human spokesperson against: misuse on social, a defective product, a controversial moment in the wider culture.
Disclose early and consistently. Audiences who learn the persona is synthetic from a hostile post react differently than audiences who knew from the first frame. The 2025 disclosure research is consistent on this point.
Plan the failure mode. Have a crisis communication script that can be voiced by the persona, by the brand directly, or by a human spokesperson, depending on the severity. The Frontiers researchers found that virtual spokespersons can repair brand image, but only if they have been built with the depth to feel credible in that moment.
These are familiar practices for any communications team that runs a virtual influencer marketing program. The point is that they are also necessary.
A note on framing
At Vinfluencer.ai, we work on AI conversational companions, not on brand campaigns. The reason this topic matters here is that the same trust dynamics that show up in brand-safety research show up in companionship, just at a different scale. A persona that listens to a user about their week is not selling a product, but it is asking for some of the same trust that a brand asks for when it puts a synthetic face on an ad. Honesty about who is behind the character, what the character can and cannot do, and what happens when the system gets something wrong is part of that bargain in both contexts.
FAQ
Is a virtual influencer actually safer for brands than a human creator?
For the failure modes that involve a creator’s private life, yes. Virtual personas do not have nightlife, political feeds, or off-platform behavior to manage. For the failure modes that involve campaign content, representation, and product issues, the picture is more mixed: 2025 research from Northeastern’s Sian Joel-Edgar suggests blame transfers to the brand more directly when a virtual persona is involved.
Have virtual influencers ever caused real brand controversies?
Yes. The 2019 Calvin Klein campaign with Lil Miquela and Bella Hadid drew queer-baiting accusations and a public apology from the brand. Shudu Gram has faced sustained criticism since 2018 over the appropriateness of a white male creator profiting from a Black-presenting digital supermodel. Both have followed brands that have worked with the personas.
Can a virtual influencer deliver a crisis apology effectively?
Partially. A 2025 Frontiers in Communication study found that virtual spokespeople can help repair brand image after a crisis, but only when the persona is rich enough (in writing, voice, and narrative depth) to feel like a credible communicator. Thinly written CGI characters tend not to land in serious moments.
What does a “brand safe” virtual influencer program look like in practice?
Treat the persona as a published character with a defensible backstory, disclose its synthetic nature early and consistently, and plan the crisis path before the crisis happens (including who delivers the apology, the persona, the brand, or a human spokesperson, depending on severity).