AI influencers catch the eye, but only real people can vouch for anything

1. In short: AI influencers get as much attention as real influencers, but persuade fewer people.

AI influencers get about as much engagement as real influencers, but persuade fewer people to buy and are seen as less credible, according to meta-analyses in the Journal of the Academy of Marketing Science in 2025 and 2026 [1, 2]. The engagement gap is small and not significant (r = 0.06), but the purchase-intention gap is significant (r = 0.15) [1]. Passing off an AI person as a real customer who tried the product is already illegal in Sweden [3, 4]. Our conclusion is that value moves to what AI cannot offer: a real person who was actually there.

The essay tests the thesis that we first want AI influencers and then turn back to people. The research supports half of it. People are needed not because AI fails, but because a recommendation needs a witness.

2. From luxury to background: when TV became everyday, human contact became the luxury.

In 1956 Sweden had 37,000 TV licences, and by 1965 about 75% of households had a TV, according to a government bill from 1966 [5]. Today 95% of Swedes have a TV, but linear viewing has fallen from 166 minutes a day in 2010 to 98 in 2024, according to the media authority Mediemyndigheten [6]. The TV didn’t disappear. It just stopped being special, as Figure 1 shows.

In the New York Times in 2019, Nellie Bowles described how the human became expensive when screens became cheap: “Human contact is becoming a luxury good.” And: “Humans are more expensive, and rich people are willing and able to pay for them” [7]. In the US, time spent with friends in person fell from about 60 to 20 minutes a day between 2003 and 2020, most of all during the pandemic [8].

David Bowie said in 2002: “Music itself is going to become like running water or electricity”, and that the only unique thing left would be touring [9]. Is influencer marketing heading the same way, first to AI creators and then back to people?

3. What are AI influencers? The first wave has already come and gone.

An AI influencer, or virtual influencer, is an invented person created with computer graphics or generative AI who promotes products on social media. The term is also used for AI actors in ads and for AI as a tool for real creators.

The first wave of such characters has already ebbed. Lil Miquela had about 3 million Instagram followers around 2020 and 2.5 million in August 2025, and her engagement sat in the 14th percentile among influencers, according to CreatorDB data in Digiday [10]. The ones that went viral often did so because people thought they were real. The AI band Velvet Sundown had 1.4 million monthly listeners on Spotify before the band admitted it was AI, and about 600,000 a little over a month later [11, 12].

In Sweden, ICA’s virtual baby influencer gathered just under 5,000 followers in three years [13]. Brands’ interest is cooling too. On Collabstr the share willing to use AI creators fell from 86% to 60% between October 2024 and August 2025, and of 27 large advertisers 60% had no plans to use them, mainly for reasons of trust [10].

4. AI vs human influencers: attention can be manufactured, persuading is harder.

Real influencers persuade more people to buy than AI influencers, but don’t get more engagement. That is the finding of a meta-analysis of 135 experiments in the Journal of the Academy of Marketing Science in 2025 [1]. The measure r shows how large a difference is, and 0 means no difference at all. In engagement the gap is small and not significant (r = 0.06), in purchase intention it is significant (r = 0.15). Figure 2 shows both estimates with their uncertainty intervals.

A newer meta-analysis of 210 experiments explains why: virtual influencers spark more curiosity but have lower credibility, and over time novelty counts for less and credibility for more [2]. What drives purchase intention is how human the influencer looks, not the number of followers [14].

One caveat: the experiments measure what people say they want to buy after a single post, and another meta-analysis finds little difference in what they say they will do [15]. No independent study has compared real sales.

5. Can you trust AI influencers? Trust drops when the audience learns that nobody was there.

AI influencers are judged less credible than real influencers [2]. According to EY’s 2026 survey, 70% of Swedes want content made with AI to be clearly labelled [16]. In Sprout Social’s 2026 survey in the US, the UK and Australia, 44% were uncomfortable with brands using AI influencers and 25% comfortable [17].

What is penalised is knowing, not quality. AI-written ad copy was rated better than the professionals’ as long as readers didn’t know who wrote it, which the researchers called “human favoritism, not AI aversion” [18]. An AI label lowers belief even in true news headlines [19], but not always. In a Stanford study the label did not make political messages less persuasive [20].

Here is the core of the essay: an AI creator persuades best when it resembles what it is not. A virtual influencer shown with real people seems more trustworthy [21], and in a study commissioned by Taboola AI ads got about as many clicks as ads made by people. The ones that didn’t look like AI did best [22, 23]. The credibility is borrowed. The AI creator works best when it is taken for a person, which is exactly what it cannot honestly be.

At the same time, trust is low even in the platforms where real influencers are. Only 8% of Swedes have high trust in Instagram and 2% in TikTok, according to Sweden’s Förtroendebarometern 2025 [24].

6. Are AI influencers legal? Yes, but not passing them off as real people.

In Sweden it is legal to use an AI-generated person in marketing, but illegal to let it pass as a real consumer or creator who has used the product. The Swedish Marketing Act makes the EU blacklist Swedish law. The list names practices that are always banned, among them falsely representing oneself as a consumer (point 22) and submitting false reviews or endorsements (point 23c) [3, 4]. Falsely claiming to have tried the product is also misleading (section 10) [3].

Since 2 August 2026 the EU AI Act also requires a deep fake, meaning realistic images, video or audio that could pass as authentic, to be clearly labelled at the latest on first exposure (Article 50(4)–(5)) [25]. The EU’s simplification package, the Digital Omnibus, did not postpone this [26]. It is unclear whether a wholly invented person is covered, and law firms read the Commission’s guidelines differently [27, 28]. The cautious choice is to label.

A label doesn’t make a false “I tried it” true. A breach of the Marketing Act can bring a market disruption fee of between SEK 10,000 and 4% of turnover [3], but there is no Swedish ruling on an AI person yet, and the Swedish Consumer Agency’s 2026 review of influencers did not mention AI [29]. This is not legal advice.

If it comes out

When The Guardian examined ads with AI-generated customers in 2026, the brands removed the images or said it was a test [30]. Guess was criticised for an AI model in Vogue [31], while H&M was open about its digital twins [32]. No case shows a measured drop in sales, and Coca-Cola’s AI-made Christmas ads got top marks in an effectiveness test [33]. Our judgement, not based on measurement, is still that being caught costs more than being open from the start.

7. AI influencers are 10–300 times cheaper per video, but the money buys you a video, not trust.

The strongest argument for AI is price. At vendors’ list prices, a 30-second video costs SEK 5–15 with an AI avatar and SEK 85–110 with an AI actor. A simple UGC video, a customer-style video ordered by the brand, costs SEK 700–3,000 from a Swedish creator [34, 35, 36, 37, 38, 39]. About SEK 100 against 1,000 makes AI roughly 10 times cheaper, and about SEK 10 against 3,000 roughly 300 times (our arithmetic, Figure 3).

A sponsored post with a smaller Swedish creator costs SEK 1,000–25,000 [40], but that price also includes an audience that trusts the creator. For an AI video, the audience has to be bought with ad money.

AI content can also be tailored to every customer. In July 2026 Mark Zuckerberg promised “a whole new and nearly infinite universe of personalized content” [41].

But AI removes the cost of production, not the cost of trust. Instagram’s head Adam Mosseri wrote on New Year’s Eve 2025: “Authenticity is fast becoming a scarce resource, which will in turn drive more demand for creator content, not less.” And: “The bar is going to shift from ‘can you create?’ to ‘can you make something that only you could create?’” [42]. Meanwhile, feeds are filling up. In a 2026 test, 59% of what a new TikTok account was shown was AI slop [43].

8. Will AI replace influencers? No, but the idea of a return holds up less well than it sounds.

AI is unlikely either to replace influencers or to disappear. Value moves, it doesn’t swing back. When recorded music became almost free, US concert prices rose 82% between 1996 and 2003, while inflation was 17% [44]. Even so, the record business came back through streaming, to $29.6 billion in revenue in 2024 [45].

People can also love characters that are artificial. The company behind the VTuber group hololive, whose performers appear as animated characters, had sales of ¥49 billion in the year to March 2026 [46], and more than one in five Swedish AI users see their AI as a person, buddy or friend [47]. But those characters don’t pretend to be anything else, and behind every VTuber there is a person.

What we say differs from what we do. In an experiment, readers spent as much time and money on a story labelled as AI-written, even though almost 40% said afterwards they would have paid less [48]. The average price of a UGC video on Collabstr fell 5.7% in 2025 [49]. The premium for the human shows up in surveys and in the law, but not yet in kronor.

And did we really all want AI influencers? Most large advertisers have no plans to use them [10]. The wave looks more like a novelty that paid off for a short while, before credibility came to matter more [2].

9. Should you choose AI influencers over real people? Not when someone has to vouch for the product.

Influencer marketing doesn’t sell content but testimony: someone like me has tried this. In Sprout Social’s 2026 survey in the US, the UK and Australia, 81% say it matters whether an influencer has actually used the product [17]. AI can create attention, but it can never be a witness.

That is why the line is deception, not artificiality. A character that is openly AI is a perfectly legitimate format. An AI posing as a real customer is the most persuasive version, and it is exactly the one that the law and audiences are putting a stop to. What is valuable is not that someone is human, but that someone was actually there.

So we think the market splits rather than turns back. AI becomes the cheap layer of product videos and ads, and people who were actually there become the expensive one. It resembles music, where copies became cheap and presence dear, more than the candle, which became a niche when electricity came. The open question is whether brands will pay more for real testimony.

Ted Chiang wrote in 2024 that “art is something that results from making a lot of choices”, and of generative AI: “It reduces the amount of intention in the world” [50]. A recommendation is a choice too.

So should you choose AI influencers over real people? For cheap product videos, often yes, if clearly labelled. For recommendations, no. Single studies suggest the human edge is largest for things you feel, taste or wear [51], but a meta-analysis could not confirm that product type decides it [14].

10. About the evidence: many experiments, little knowledge of real sales.

Collabify wrote this essay in October 2026 from published sources, and where we state our own opinions, we say they are ours. The costs are our arithmetic on list prices, assuming USD 1 ≈ SEK 10. What is missing is Swedish data on trust in influencers, independent sales data and a Swedish legal ruling.