When Virtual Influencers Work—and When They Don’t
The answer to when virtual influencers work is conditional, but it is not vague. They are most defensible when the campaign needs the kind of influence their design can plausibly provide: a stylized persona for distinctiveness and attention, or a more human-like persona for credibility when the surrounding context supports that interpretation. This final judgment connects outcome, mechanism, fit, category maturity, and governance so readers can decide whether to use a virtual influencer, test one narrowly, or choose another format.
Start with the job, not the avatar
A virtual influencer should not be selected because it is novel, realistic, controllable, or fashionable in isolation. Those are properties of the communication system, not the campaign’s objective. The first question is what the audience must do next. Is the immediate task to notice an unfamiliar idea, interact with a brand, form a favorable attitude, trust a functional claim, or take a consequential action? Different answers make different weaknesses visible.
The pooled comparative evidence gives a useful baseline. Across 210 independent experimental studies, 110 papers, 643 effect sizes, and 91,058 consumers, virtual influencers were more novel than their comparison conditions. The average engagement difference was not statistically reliable. On average, the virtual condition was lower on credibility, brand attitude, advertisement attitude, behavioral intention, and attractiveness. These results, reported in the meta-analytic review of virtual influencer marketing, do not describe a universal winner or loser. They describe a recurring separation between being noticed and being believed.
That separation changes the decision. If a campaign’s immediate purpose is to make a new visual identity stand out in a crowded feed, novelty may be a legitimate value. A low-anthropomorphism character can make the artificiality of the source part of the signal. It does not need to imitate a human expert to be memorable. The audience may respond to an unusual world, a distinctive character, or a creative association with the brand. The relevant test is whether the attention is useful for the next stage of the customer relationship, not whether the character attracts attention in the abstract.
If the campaign asks the persona to explain, recommend, or validate something that depends on expertise, the bottleneck changes. A high-anthropomorphism design may make the source appear more capable of intentional action and more suitable for credibility-based evaluation. The meta-analytic structural model finds that the credibility disadvantage is less negative for high anthropomorphism than for low anthropomorphism, and that credibility has a stronger relationship with attitude under high anthropomorphism. That is a relative improvement, not a declaration that a synthetic source has become as credible as a human endorser.
The residual disadvantage matters. Human likeness raises the audience’s expectations about the persona’s behavior, knowledge, and experience. If the account makes claims that the persona cannot plausibly support, greater realism can make the mismatch more conspicuous. A human-like digital identity that appears to speak from personal experience may invite more scrutiny of who actually had that experience. Realism can support a credibility route, but it also makes credibility a more explicit question.
The two choices can be illustrated without turning them into universal recipes. Consider a campaign whose primary task is to make a familiar consumer product feel newly distinctive. A stylized character may add the needed contrast without pretending to be a product expert. Now consider a new technology offering whose value depends on explaining capabilities and signaling modernity. A highly human-like persona may be a more congruent source, especially if its autonomy and message are presented consistently with that role. In either case, the scenario describes a hypothesis to test. It does not guarantee that the audience will interpret the cues as intended.
This is why engagement must remain separate from conversion. A comment can signal curiosity, amusement, or disagreement. A share can signal social currency rather than product interest. A favorable attitude can still fail to produce behavior when price, risk, need, or product fit intervenes. The final outcome should be measured at the distance the campaign claims to influence. A novelty objective needs a novelty and attention test; a credibility objective needs source and message evaluations; an action objective needs a relevant behavioral measure and a comparison that can distinguish identity effects from creative effects.
A human influencer is not automatically the answer to every credibility problem. Human creators can be sponsored, inconsistent, or poorly matched to a product.