Homemade Pasta · Research
US brands will put $44 billion behind creators in 2026. Independent evidence now answers the questions that decide whether that money works: how big a creator to hire, when a brand is ready, what makes the content work, and where AI-generated creators fit.
Why this exists
US creator ad spend reached $29.5 billion in 2024 and IAB projects $44 billion for 2026.18 Most of the numbers used to allocate it come from platforms and influencer tools with a commercial interest in the answer.
Better evidence now exists, and it comes from outside the platforms. Independent studies since 2022 cover hundreds of thousands of real posts and campaigns, including revenue data, and the largest reviews pool 345 controlled experiments on more than 135,000 consumers.1, 2, 7 This page turns that evidence into decisions: who to hire, at what size, with what message, and how to know whether it worked.
Where the evidence is thin or contradicts itself, the page says so. Two of the most repeated claims in the category, that AI creators outperform people and that whitelisted creator ads beat brand ads by a fixed margin, have no independent support. The first is contradicted outright.
Executive summary
r measures the strength of an effect on a scale from −1 to 1. Zero is no effect, 0.1 is small and 0.3 is moderate. A negative value means the comparison did better, here human creators.
Four findings determine how much to spend, on whom, and how to judge the result.
The practical consequence is a change of unit. Buy creators on delivered views or outcomes rather than on follower count, spread budget across more of them, match size to how well the market already knows the product, and design a holdout before launch rather than reconstructing lift afterward.
Recommendations
| Action | Why it comes first | Quantified impact |
|---|---|---|
| 1. Price creators on views or outcomes, not followers | Follower count is the most visible input to creator rates and the weakest predictor of reach. It is also the easiest number to inflate, and inflated counts still raise perceived influence. | At the published average elasticity, doubling followers returns about 7% more views. Moving from 10,000 to 1,000,000 followers returns about 58% more. |
| 2. Default to many small creators | Return per dollar falls as follower count rises, in revenue data as well as in engagement data. | Low-follower targeting returned an order of magnitude more on revenue per follower, revenue per reach and return on spend. |
| 3. Match creator size to brand familiarity | Large creators perform best on familiar products with concrete, demonstration-led messages. New brands and launches do better with smaller creators and recommendation-style messages. | Directional, drawn from 135 experiments. No study isolates company size itself. |
| 4. Run creator content as paid media with a holdout | Paid amplification of creator content is the fastest-growing part of the category and the part that can be tested. A geographic or audience holdout designed into the brief is the only defensible lift number. | Observational methods missed true lift by a factor of three in half of 15 randomized Facebook tests. |
| 5. Use AI creators for attention, people for trust | Virtual influencers match people on engagement and trail them where the sale depends on credibility. Disclosure rules now apply to both. | Engagement gap not significant. Credibility gap r = −0.23, purchase intention gap r = −0.13. |
Part One
How big a creator to hire, why the research appears to disagree, and when a brand is ready to scale up.
1.1 The size question
The cleanest estimate of what a larger following is worth comes from Wharton researchers who tracked more than 500,000 TikTok videos and used causal machine learning to separate the effect of follower count from the effect of content. They define the follower elasticity of impressions: the percentage gain in views for a one percent gain in followers. It is always positive and averages 0.10.3
An elasticity of 0.10 means views scale with the tenth root of followers. A creator with ten times the audience delivers 26% more views (100.10 = 1.26). A hundred times the audience delivers 58% more (1000.10 = 1.58). Creator rates do not scale that slowly.
Revenue data points the same way. Beichert, Bayerl, Goldenberg and Lanz followed direct-to-consumer campaigns from follower counts through reach, engagement and actual revenue, net of what the endorsements cost, using more than 1.8 million purchases and three field studies. Low-follower targeting outperformed high-follower targeting by an order of magnitude on revenue per follower, revenue per reach and return on influencer spend. Engagement explained the gap.2
The largest review of controlled experiments finds that creators above one million followers drive more purchase intention than smaller ones, while smaller and mid-size creators drive more engagement.1 The field data point the other way, because the two kinds of study measure different things.
The experiments measure persuasion per exposure, with the creator's fee absent. The field studies measure return per dollar, with the fee included. A large creator can persuade more per view and still return less per dollar when the rate grows far faster than the reach.
The same authors caution that the experiments lean on stated intention, which overstates effect because several factors intervene between intention and purchase.1 For budgeting, the revenue evidence governs.
An analysis of 802 Instagram campaigns featuring more than 1,700 influencers, backed by eye-tracking and laboratory experiments, found an inverted U: engagement with sponsored content rises with follower count, then falls. A larger count signals broader reach but also a weaker relationship with the audience.4
Two conditions flatten the curve. When influencers write sponsored content in their own style rather than repeating the brand's messaging, small and large creators both gain relative to mid-size ones. When the sponsor is a relatively unknown brand, follower count matters less for engagement.4
On TikTok, engagement concentrated among creators with roughly 7,500 to 10,000 followers, and the return on popularity fell sharply as popularity rose. Branded challenges inviting user videos were the exception: the elasticity held steady across creator sizes, which supports using large creators to seed participation formats.3
A study of 5,835 sponsored posts by 2,412 influencers across 1,256 campaigns estimated the return on influencer budget directly. A 1% increase in spend raised engagement 0.457%. Doubling a budget therefore buys about 37% more engagement (20.457 = 1.37), not twice as much.5
The same study found that larger follower counts raised effectiveness, one of the few field results pointing that way. It measured engagement as reposts on Weibo in 2018 and did not observe revenue. The shape of the spend curve is the transferable finding, not the coefficient.
1.2 When to engage
No published study tests influencer effectiveness by company size. The evidence covers product and brand familiarity, which is what company size usually stands in for. A large company launching a new product is, for this purpose, an unknown brand.
Three independent findings point the same way. Large creators are better matched to familiar products, and the authors of that review recommend smaller creators for new or niche brands.1 Posts announcing new products lower engagement, because followers hesitate to vouch for something their networks do not know.5 And the reputational cost creators pay for sponsored content is smaller when the brand is less well known.10
| Market position | Creator size to favor | Message that fits |
|---|---|---|
| New brand, or any launch | Many small and mid-size creators | Recommendation-led and experiential. Personal narrative over specifications. |
| Recognized in a niche | Mid-size creators with category expertise, plus small creators for volume | Expertise-led for awareness. Hedonic content for trial on social platforms. |
| Familiar, mass-market product | Large creators become defensible | Concrete, rational, usage-based. Demonstrations and comparisons, especially for products buyers can evaluate before purchase. |
The message column comes from the same review. Larger creators do better with concrete, usage-based endorsements and with search products. Smaller creators do better with abstract, recommendation-based messages and with experience products, where a detailed or insistent pitch from a small creator can backfire.1
Creator effectiveness declines with audience age. Younger audiences are the most receptive, and Gen X and Baby Boomer audiences are more skeptical of influencer endorsements and engage less with influencer content.1 For categories selling to older buyers, other endorsement forms deserve a share of the test budget.
Funnel stage interacts with platform. In in-market data from 57 sponsored blogging campaigns, blogger expertise mattered more for awareness than for trial, but expertise did not drive engagement on Facebook, where hedonic content worked better for trial campaigns.11
1.3 Model
Views follow the measured elasticity.3 The rate follows your own rate card. The output is what a larger creator returns per dollar in reach, before any difference in persuasion.
Set the two follower counts and how the larger creator's rate scales.
The larger creator must convert about 6.3 times as well per view to break even on reach. Controlled tests support a persuasion advantage for large creators, not one of that size.
Views multiple = follower multiple0.10. Rate multiple = follower multiplerate elasticity. Views per dollar = views multiple ÷ rate multiple. The spread figure assumes the smaller creators reach distinct audiences. The 0.10 elasticity is the published average across TikTok videos from 2020 and 2021; the curve varies by content type and is often nonlinear, so outputs are directional.
Part Two
What moves results once the creator is chosen, what disclosure costs, and where AI-generated creators earn a place.
2.1 What works
Across 135 experiments, creators beat brand-owned posts on both engagement and purchase intention. They beat celebrities on engagement and match them on purchase intention. Meta-analytic structural modeling finds creators work only indirectly, through two channels: credibility, which mainly carries purchase intention, and attractiveness, which mainly carries engagement.1
| Compared with | Engagement | Purchase intention |
|---|---|---|
| No endorsement | 0.31 | 0.26 |
| Celebrity | 0.24 | No difference |
| Virtual influencer | No difference | 0.15 |
| Brand's own post | 0.11 | 0.09 |
Each figure is r, the strength of the creator’s advantage on a scale from −1 to 1. Zero means no advantage, 0.1 is small and 0.3 is moderate. Creator marketing works, at magnitudes that justify testing rather than faith.1
A second review, covering 251 studies, separates what drives attitude from what drives sales. Follower identification with the creator matters most for attitude and engagement. Informational and hedonic value in the post matter most for purchase intention. The creator's communication matters most for actual purchase behavior.6
A review of 61 studies found that disclosing sponsored content lowers brand attitude, credibility and evaluation of the source, and raises recognition, persuasion knowledge and resistance. Its effect on behavioral intention was not significant.9 The compliance cost of clear disclosure is smaller than its reputation suggests.
Across 85,669 YouTube videos from 861 beauty and lifestyle creators, a sponsored video cost an average 0.19% of subscribers relative to a comparable organic video. The loss was larger for bigger creators, and smaller when the sponsored content fit the creator's usual style or the brand was less well known.10 A creator who runs sponsorship after sponsorship is spending down the audience a brand is paying to reach. Frequency and fit belong in the brief.
2.2 AI influencers
The largest review to date, published in 2026, pools 210 experiments on 91,058 consumers comparing virtual influencers with human ones. Virtual influencers are rated more novel and draw statistically equivalent engagement. They trail human creators on credibility, ad attitude, brand attitude and behavioral intention.7
The advantage is also decaying. Comparing studies published through 2022 with later ones, the persuasive weight of novelty declined as consumers habituated, while credibility grew in importance.7 A virtual creator that wins attention because it is new will need to win trust as the novelty wears off, which is the ground where virtual creators trail human ones.
Design matters. Highly human-like virtual influencers persuade mainly through credibility and suit futuristic brands, new products, rational messages and claims about touch, taste or smell. Stylized, cartoon-like ones persuade through novelty and suit traditional brands, familiar products and emotional messages.7 A second review of 46 experiments finds the persuasion gap concentrates where virtual influencers show low or no follower counts, and becomes negligible for those with large followings.8
The FTC's 2023 Endorsement Guides expanded the definition of an endorser to include anything that appears to be an individual, group or institution, which brings virtual influencers under the same disclosure standard as people.13 In the EU, Article 50 of the AI Act has applied since 2 August 2026. Deployers must label deepfakes, and providers of generative systems must mark synthetic output. The Digital Omnibus deferred the Act's high-risk obligations but left these transparency duties on schedule, apart from a marking grace period to 2 December 2026 for systems already on the market.14
Research on AI-generated labels is split. Some experiments find a label lowers trust in the ad and the advertiser. Others find disclosure raises engagement or purchase intention when consumers rate AI capability highly.16 Nothing settles it yet. Label as the law requires, and test the creative rather than assuming a penalty.
2.3 Evidence ledger
Common claims in the category, graded against the independent evidence.
Revenue field data show it directly. Reach and engagement studies using different data and methods point the same way.2, 3, 4
Significant on engagement and purchase intention across the experimental meta-analysis.1
It lowers attitude and credibility. The effect on behavioral intention was not significant across 61 papers.9
Credibility, attitudes and intention all significantly lower across 210 experiments. Engagement statistically equivalent.7
True for purchase intention per exposure in experiments. Contradicted on return per dollar in revenue data.1, 2
On engagement, yes. On purchase intention there is no significant difference.1
Experiments point in both directions and no meta-analysis has settled it.16
Widely circulated figures claim virtual influencers deliver markedly higher engagement at far lower cost. The two largest reviews find engagement parity at best and deficits on every trust measure.7, 8
Published vendor claims range from 20% better to three times better, with no independent test behind any of them. Treat it as a hypothesis for a controlled test.
Both are observational. Codes miss buyers who never use them and credit buyers who would have purchased anyway.12
Reach barely tracks it, and purchased followers raise perceived influence even when consumers know some are fake.3, 15
Part Three
Why most reported influencer results overstate the effect, and the designs that measure it.
3.1 Measurement
The experimental evidence on influencers measures what people say they will do. The authors of the largest review say relying on purchase intention overstates effectiveness, and call for sales data and field tests.1 Randomized tests of influencer campaigns against sales are close to absent from the published evidence.
Observational shortcuts do not fill the gap. When researchers compared 15 randomized advertising experiments at Facebook, covering 500 million user observations, against the matching and regression methods advertisers commonly use, the observational estimates generally overstated effectiveness, and in half the studies they were off by a factor of three.12
| Design | How it works | When it fits |
|---|---|---|
| Geographic holdout | Run amplified creator content in matched test markets and withhold it from control markets. Compare sales measured outside the platform. | Paid amplification of creator content, any brand with regional sales data. |
| Audience holdout | Withhold amplified creator ads from a randomized share of the target audience using platform lift tooling, then compare conversions. | Brands with enough conversion volume for the platform's power requirements. |
| Staggered creator launch | Randomize the order in which creators go live and read the sales response to each wave. | Organic creator programs, where exposure cannot be withheld from individuals. |
Each needs a control defined before launch. None can be reconstructed afterward.
Whitelisting, which Meta calls partnership ads and TikTok calls Spark Ads, runs a creator's content as paid media through the creator's own handle. It converts an organic post, which reaches whoever the algorithm chooses, into a paid placement with a budget, a target audience and a holdout.
IAB estimates US creator ad spend at $37 billion in 2025 and $44 billion in 2026, and describes brands treating creators as a distinct media channel.18 No independent study yet measures whether amplified creator content outperforms brand creative. That is a question each brand can answer for itself with the designs above.
A 2023 study combining real-world data and controlled experiments found that follower count raises perceived influence even when consumers recognize that some followers are fake. Creator expertise and existing popularity weaken the effect.15 The market rewards the number whether or not it is real, which is the incentive behind purchased followers.
Industry estimates of fraud prevalence come largely from vendors that sell detection tools, and the figures vary widely. The durable defense does not depend on any of them: pay on delivered views and outcomes, which fake followers do not produce.
Appendix
B2B, market sizing, terms and sources.
A.1 Supporting detail
Almost all influencer evidence comes from consumer markets. The B2B work consists mainly of interviews with marketers. One, drawing on 22 interviews with senior marketers, found B2B practitioners prefer the term influential marketing, grounded in trust, expertise, professionalism and industry networks, and reports that empirical research remains scant.17 A second set of 22 interviews identified four distinct strategies B2B firms use to operationalize influencer programs.17
Neither measures performance. The consumer findings most likely to transfer are the ones about credibility: expertise-led creators, concrete messages, and caution with older audiences, who are measurably more skeptical of influencer endorsements.1
eMarketer puts US influencer marketing at $10.52 billion for 2025, defined as payments to creators to promote products on social and video platforms, excluding paid media.18 IAB puts US creator ad spend at $37 billion for 2025 on a broader definition of creator advertising.18
The two measure different things. Figures should never be compared across sources without checking the definition.