AI creators (virtual influencers): what they are and why brands use them in 2026

What does the category’s history look like, from Hatsune Miku to Mia Zelu?

The category is not new. The timeline shows that more than fifteen years passed between the first commercially significant virtual celebrity and mainstream brand adoption — and almost all of that time was spent making the technology operational, not on the audience “getting used to it.”

2007. Hatsune Miku. A Japanese Vocaloid character by Crypton Future Media — the first commercially successful fully artificial artist. Hologram concerts, album sales, merch. The “virtual artist” category was born here, before social networks in the modern sense.

2015. Any Malu (Brazil). An animated virtual character who grew out of YouTube. 280M+ views, a show of her own on Cartoon Network. The case shows that a non-photorealistic stylized character can scale through a strong personality, not through photorealism (Socially Powerful).

2016. Lil Miquela. Created by the studio Brud — the first commercially significant Instagram-native virtual influencer. 2.6M followers on Instagram and a press-reported rate of around $10,000 per sponsored post. Brud raised $20–30M in January 2019 at a pre-money valuation of at least $125M, led by Spark Capital (TechCrunch, 14.01.2019), and was acquired by Dapper Labs in October 2021 in an all-equity deal with undisclosed terms (TechCrunch, 04.10.2021). Campaigns with Prada (Milan Fashion Week takeover 2018 — noted as a strong adoption signal, with no confirmed public engagement metric), Calvin Klein (2019, with Bella Hadid, +150% mentions), Pacsun, Samsung (creative.salon).

2017. Shudu Gram. “The first digital supermodel,” by Cameron-James Wilson. The Balmain Fall 2018 “Virtual Army” campaign was the first time a major fashion house put a virtual model at the center of a campaign, not as an experiment (BuzzFeed).

2018. Imma (Japan) and Noonoouri (Germany). Imma, by Aww Inc., is Japan’s leading fashion influencer; the IKEA Harajuku 2020 campaign (a 72-hour live stream from a physical storefront with LED panels, with real time-of-day adjusted via color temperature), plus partnerships with Porsche, Coach, PUMA (Muse by Clios). Noonoouri, by Joerg Zuber, has worked with Dior, Balenciaga, Valentino; in August 2023 she signed the first record deal ever for a digital artist with Warner Music and released the track “Dominoes” (Hypebeast, September 2023).

2019. KFC virtual Colonel Sanders. Relaunching the mascot as a streetwear influencer drove +23% engagement among Gen Z (creative.salon).

2020. Rozy (South Korea). Sidus Studio X launched Korea’s first virtual influencer. By 2022 she had brought her creators roughly $2M in revenue, crossing $1M in her first year. Partnerships: Chevrolet, Calvin Klein, Hera (LG Household), dozens of local beauty brands. She released a music album and founded her own skincare brand, OHROZY (CNN).

2022. Kyra (India). FUTR Studios, 263k followers, campaigns with L’Oréal, Realme, boAt, American Tourister. In 2024 the creators pitched on Shark Tank India S3, which cemented the category in Indian mainstream awareness (The Swaddle).

2023. Aitana López (Spain). From The Clueless agency in Barcelona. Earnings up to €10,000/month, averaging €3,000, with a single ad rate just over €1,000, and 343,000+ followers in 18 months. Brand case: Big (sports nutrition) (Euronews, 27.12.2024). The agency was founded after the studio grew tired of the unpredictability of real human models (Fortune, 23.11.2023).

2024. BMW iX2 × Lil Miquela. The “Make It Real” campaign by the agency Monks — the first time a premium automaker gave a virtual character the lead role in a product launch, putting the iX2 in front of Miquela’s audience of more than 8.2 million. The campaign won 1 CLIO, 1 Epica, and 3 Lovie awards (Monks case study). Engagement figures often quoted for this campaign in trade posts (8.4% ER, 1.5M impressions) do not appear in any public primary source, so we don’t cite them.

2025. Mia Zelu. A generated photo model from the Zelu House agency, who grew to 169k followers off “attending” Wimbledon 2025 — most viewers didn’t realize she was AI, even though “AI influencer” sat in her bio. The case was picked up by Northeastern’s D’Amore-McKim Business School (D’Amore-McKim) and UTS Sydney (UTS) as an illustration of digital trust erosion.

Geography shows the category is not local: the US (Miquela), Japan (Imma), Korea (Rozy), Germany (Noonoouri), Spain (Aitana), Brazil (Any Malu), India (Kyra), the UK (Shudu). That is a structural fact — an AI creator is technically untied to geo, unlike a human influencer who monetizes through a local market.

What market sits behind the category in 2026?

The global virtual influencer market is valued at $6.06B in 2024 with a forecast up to $45.88B by 2030 and a CAGR of 40.8% (Grand View Research, 2025). Alternative estimates diverge by multiples, which is in itself a signal that the category is young and methodologies have not settled:

Table 1. Virtual influencer market estimates

Sources: Grand View Research, Straits Research, Market.us, Market Research Future, SNS Insider.

The most likely explanation for the 2–5× spread across similar horizons — a hypothesis, not a settled fact — is that some analysts count “pure AI” characters while others count the entire stack (CGI avatars for metahuman campaigns, UGC avatars from HeyGen/Synthesia, AI tooling for tagging and discovery inside classic influencer marketing). In a marketer pitch, the safer citation is Grand View ($45B by 2030) as a conservative upper bound on the mainstream.

US influencer marketing spending is projected at $10.52B in 2025 (eMarketer, 13.03.2025) and $13.7B by 2027 (eMarketer, 23.06.2025). Against that backdrop, Ogilvy’s “Future of Social” report (March 2024) projects that virtual influencers could account for up to 30% of influencer budgets by 2026 (Mediaweek, 26.03.2024) — an upper-bound scenario, not a committed forecast. (This projection circulates online misattributed to Gartner; we could not find any Gartner publication behind it. The source is Ogilvy.)

Why are brands buying AI creators right now?

Four pressures aligned in phase between 2023 and 2026 — and together they moved the category from experimental to operational.

The technology turned operational

Through 2023–2025, generative video moved out of demo mode into production tooling. Veo 3.1 (Google), Kling 3.0 (Kuaishou), and Runway Gen-4.5 all ship public APIs. Across the 138 channels on the Cifratar pipeline, generation costs run under $0.40/sec on flagship models and under $0.10/sec on mass-market models (internal production data, July 2026). The tool layer itself churns — OpenAI’s Sora 2 was discontinued as a consumer product in April 2026, with the API sunsetting in September — but the pipeline logic persists regardless of which model sits in each slot. ElevenLabs voice synthesis covers 30+ languages in one click, starting at $11/mo on the Creator plan (ElevenLabs). This means localization into 30 markets now costs around $100/mo, instead of hundreds of thousands of dollars on a production team. We break the full stack down in our pipeline guide for AI-native advertising.

Control and predictability

A virtual character won’t run into a scandal, post a political opinion, miss a deadline, or get more expensive after going viral. That’s literally the reason The Clueless gave when launching Aitana López: the studio was tired of the “unpredictability” of working with live models (Fortune). From a brand manager’s view this isn’t a rejection of the human factor — it’s swapping one class of risk (human behavior) for another (the narrative the managers write), and the second class is process-manageable.

The engagement gap

HypeAuditor’s November 2019 Instagram analysis found engagement rates on virtual influencers roughly 3× those of humans; in the >1M follower tier, 2.89% for virtuals against 0.7% for humans (HypeAuditor data as reported by WeRSM, December 2019; HypeAuditor’s 2021 follow-up repeated the “almost three times higher” finding without republishing the tier percentages). No public primary source confirms updated figures for 2024–2025. Use 3× as a narrative anchor, not a precise metric.

In trust-sensitive verticals (parenting, finance, health), our working hypothesis is that the advantage flips toward humans — audiences there need parasocial trust, which a virtual character delivers worse (consistent with the academic findings below). We have not found a reliable public number for the size of that flip, so we won’t quote one. For the marketer the takeaway is: an AI creator is not “always better.” It has its own efficiency curve across the funnel and across verticals.

What does the academic literature say?

The category is mature enough to have its own empirical field. A bibliometric review in ScienceDirect (2025) tracks the growth in publications on the topic from single digits in 2016 to hundreds per year by 2024 (ScienceDirect).

A systematic literature review in Acta Psychologica (2025), titled “Authenticity, ethics, and transparency in virtual influencer marketing,” states the category’s central paradox: consumers perceive virtual influencers as “authentically fake” and accept them precisely in that status. Curated flaws (programmed biographical “imperfections”) and self-justification (open acknowledgment of AI nature) soften the uncanny-valley effect (Acta Psychologica).

Wiley’s Psychology & Marketing (Gutuleac, 2024) reports a counterintuitive result: highly anthropomorphized virtual influencers produce a greater sense of uncanniness than moderately stylized ones (Wiley).

For the marketer, the practical read is this: the race for photorealism can work against the brand in trust-sensitive verticals. A stylized character (in the Noonoouri or Any Malu mode) can be strategically better in categories where the audience needs to “forgive” the artificial. That’s a design lever worth holding consciously, not as a consequence of technical limits.

Regulatory context 2026

By the end of 2026, watermarking and disclosure become mandatory in every major market, and that moves “marking AI content” from best practice into compliance.

• YouTube. Disclosure mandatory since March 2024, enforcement since early 2025 (YouTube Blog).
• Meta (Facebook/Instagram/Threads). “AI Info” label since May 2024, tied to C2PA Content Credentials (Meta).
• TikTok. C2PA since January 2025 — the first platform with automatic detection; in the first year it labeled 1.3B AI videos (TikTok Newsroom).
• FTC. Endorsement rules apply to virtual endorsers the same as to humans, plus the endorser’s non-human nature must be disclosed. The 2024 Rule on Consumer Reviews and Testimonials (16 CFR Part 465) explicitly covers AI-generated fake reviews and fake indicators of social media influence (FTC, 14.08.2024). Civil penalties run up to $53,088 per violation as of the 2025 inflation adjustment (Federal Register, 17.01.2025); the figure at the rule’s October 2024 enactment was $51,744.
• EU AI Act Article 50. Full enforcement August 2, 2026 (artificialintelligenceact.eu).

The practical takeaway: not labeling by the end of 2026 is an active choice of risk, not a passive time saving. For a brand, it’s cheaper to wire C2PA and SynthID into the production pipeline from day one than to rebuild the stack after the first fine. How platform detection actually treats AI content is a topic of its own — we cover it in Algorithms and AI content.

Where does it work, and where doesn’t it?

To avoid buying the category as a “universal solution,” it helps to fix up front where its strength and weakness sit on the funnel.

Works (top-funnel awareness and engagement): fashion, lifestyle, entertainment, tech, beauty, food. The Aitana, Imma, Rozy, and KFC cases all show a steady engagement gap in favor of virtuals.

Parity (mid-funnel consideration): B2B SaaS, retail, automotive. BMW iX2 × Lil Miquela is the indicative case: a premium brand earns relevant impressions, but direct conversion still needs a second touch.

Weaker than humans (bottom-funnel conversion in trust-sensitive verticals): parenting, finance, health. Here audiences need parasocial trust, which a virtual character still delivers worse (Acta Psychologica).

For the brand the takeaway is: an AI creator is not a replacement for the existing pool of human influencers — it’s a different tool in the portfolio. Top-funnel and brand-building on AI, performance bridge on humans, mid-funnel wherever the economics works out.

Frequently asked questions

How is an AI creator different from a regular avatar in HeyGen or Synthesia?

A stock avatar from HeyGen or Synthesia is “a face without a personality”: a set of photorealistic models you can use to voice over any script. An AI creator is a character with a biography, a consistent visual style, a vocal identity, and a posting history, who has an audience. One is used as a production tool; the other is a media asset with its own capital value.

Is it legal to run ads from an AI character?

Yes, with explicit disclosure. The FTC, EU AI Act Article 50, YouTube, Meta, and TikTok all require labeling AI nature — somewhere via a toggle at publish, somewhere via C2PA metadata. Hiding the AI nature is prohibited: FTC civil penalties run up to $53,088 per violation (2025 figure), on top of platform sanctions.

In which verticals is the AI creator already working, and where isn’t it yet?

See the “Where does it work, and where doesn’t it?” section above for the full breakdown. Short version: works in fashion, lifestyle, entertainment, tech, beauty, food; parity in B2B SaaS, retail, automotive; weaker in parenting, finance, health.

Does the audience really believe an AI influencer is a “real person”?

No, and that’s no longer the goal. Current research finds that audiences accept the AI character precisely as “authentically fake” — sincerely-fake, if openly declared as such. Curated flaws (biographical imperfections) strengthen engagement; trying to mask AI nature backfires (see the Mia Zelu Wimbledon 2025 case).

What replaces a human influencer if the budget is capped at $5k?

With $5k you can either buy 1–2 posts from a mid-tier micro influencer ($500–$2,500 per post), or launch your own AI creator on a 2D stack: $1.5k setup plus $1.5k–$2.4k operational for the first month = roughly 30–90 content units instead of two posts. The decision turns on whether you need a one-time reach spike or a systemic channel for 6+ months.

What changes on the 2026–2027 horizon?

Three trends. Auto-detection of AI content becomes the default on every major platform (TikTok has already labeled 1.3B videos). Watermarking is embedded into tools at the model level (Google’s Veo 3.1 writes SynthID by default). The category shifts from explaining “what it is” to selling “how it’s measured” — the client expects unit economics and cases, not an intro to the technology.

What should a brand do next week?

Before launching: lock in the vertical your brand operates in and check against the table above which funnel zone it lands in. If it’s in “works” — the next useful step is to look at how a “channel factory” unit economics compares against the standard “one character — one project” model (full comparison here). If it’s in “weaker” — the category isn’t closed off, but it requires a stylized approach and a realistic KPI bar.