A human creator costs about $198 a video before usage rights add 30 to 50%[1]. An AI tool costs about $11 a video on Arcads’ entry plan[2]. An agency retainer runs $5,000 to $15,000 a month, an in-house team $250,000 to $400,000 a year[3]. But the per-asset comparison is wrong: you are buying tests, and at a hit rate of one in twenty, the number of tests you can afford decides whether you find a winner. AI wins on volume and speed. A person wins the moment the ad needs someone who has used the product. The rules on what a synthetic presenter can say are strict, with penalties starting at $53,088 per violation.
We make AI ads, which is reason to read this with your guard up, and also why we know where the seams are. The cases where a human wins are here because pretending otherwise would cost us a client in month two.
The cost table, sourced
| Option | What it costs | What that leaves out |
|---|---|---|
| Human UGC creator | $150 to $300 per short video, average about $198[1] | Usage rights add 30 to 50%. Whitelisting and paid usage on top. Two to three weeks from brief to file is common. |
| AI UGC tool, self-serve | About $110 a month for 10 videos, $220 for 20 (Arcads, as reported by reviewers)[2] | Roughly $11 a finished video. You write, direct, check and edit it yourself. No public pricing page, so plans change. |
| Creative-only agency retainer | $5,000 to $15,000 a month[3] | Strategy and production included. Integrated agencies with media buying run $10,000 to $40,000. |
| In-house creative function | $250,000 to $400,000 a year[3] | A strategist, an editor, a buyer and a coordinator, fully loaded, before tools and creators. |
Two things the table hides. The AI tool’s $11 excludes the person operating it, the one doing the research, the script and the claim check. The creator’s $198 excludes the coordinator hours to brief, chase and re-shoot, which is the real reason most creator pipelines cap around eight videos a month.
The number that matters is cost per test
Motion’s 2026 benchmark puts the share of ads that become winners at about 5%[4]. Twenty tests produce about one winner. The question is never “what does a video cost,” it is “how many tests can I afford, and how good is each one.”
Twenty human videos is $3,960 before rights, plus a month managing creators. Twenty AI videos is around $220 in software, plus a month doing everything else. In both cases the judgement is the real cost; the software is a rounding error. That is why agencies exist here, ours included: the tools made pixels cheap and made the thinking the whole job.
The tools made the pixels cheap. That made the thinking the whole job.
Where AI wins today
- Demonstration. Product in hand, open it, pour it, use it. A synthetic presenter is very good at showing a thing.
- Hook and format testing at volume. Twenty openings on the same body in a week, which no creator calendar supports.
- The same script from many faces. Five presenters, five rooms, one afternoon.
- Iteration speed. A winner gets re-cut in hours, not in the three weeks a re-brief takes.
- Categories where the product does the talking: consumables, tools, anything with a visible before and after that the product itself performs.
- Brands that cannot get a creator on the schedule, or whose product is too niche for the marketplaces.
The common thread: the product can prove all of this on camera. If the proof is in the product, a synthetic presenter can hold it up.
Where a person still wins, and probably always will
- Genuine testimony. “I use this and here is what happened” is the one line a synthetic presenter can never say.
- Lived credibility. A woman talking about post-partum hair loss, a runner talking about a knee. The story is the product.
- Complex physical demonstration where the technique is the point: a diffuser routine, a makeup look, a workout.
- Founder-led content. The person who built it, saying why. Nothing synthetic gets close.
- Anything where the viewer has to believe this specific person owns the product and would buy it again.
Here the proof is in the person. The longest-running haircare ads in the Meta Ad Library are seven-minute testimonial advertorials, people telling their own story. That is the right format for that promise, and one we will not fake. If your product needs it, hire the human.
The rules, plainly
- Meta applies an “AI info” label to detected AI-generated content and to its own generative tools when the edit is significant or a photorealistic person is involved; Ads Manager carries a disclosure control advertisers are expected to use[5].
- TikTok requires realistic AI-generated content that shows people or scenes to carry its AI-generated content label, and auto-labels via content credentials where it can[6].
- YouTube requires disclosure of realistic altered or synthetic content a viewer could mistake for real[7].
- The FTC’s rule on consumer reviews and testimonials (FTC, 16 CFR Part 465) took effect October 2024. It bans testimonials by people who do not exist or never used the product, naming AI-generated fakes specifically[8], and the Commission sent ten warning letters under it on December 22, 2025, citing penalties of up to $53,088 per violation[9]. A synthetic presenter implying an experience it never had is fabricated evidence, not a creative choice.
What a synthetic presenter can and cannot say
This is the rule every script runs through, written so you can adopt it whether or not you work with us. A gate in our pipeline checks every script against it before anything is produced.
The presenter can
- Show the product, open it, pour it, apply it, hold it to the lens
- Read the claims the brand supplied and can substantiate, in the words it approved
- Ask the question the reviews keep asking, then demonstrate the answer
- Compare formats, sizes and routines the label supports
The presenter never
- Say “I use this”, “this worked for me”, or imply any personal experience
- Describe results, timeframes or a personal before and after
- Play a customer, a doctor, an expert, or a real person
- Run without the AI label the platform asks for
The test for any line is simple. Could a stranger who has never touched the product say this truthfully while holding it? “One pump covers the whole thing” passes. “My hair has never felt better” fails.
Where we stand, and when we will tell you to hire a human
We use synthetic presenters for demonstration and volume, label every ad, and never make testimonials. If a brand needs a real person testifying, we say so on the first call, not in month two. What we do produce is twenty demonstrative concepts a month, built on the reviews and the competitor read, and let the ad account decide. See it before you believe it: we will make one ad for your product, free, to put next to your best creator video.
Questions people ask
Do AI ads perform worse than creator ads?
Nobody has clean public data on that yet, us included. Platforms do not penalise labelled AI content, so performance comes down to the concept and research behind it, same as a human creator. The marginal cost of one AI ad is low enough to just test it against your current best.
Will Meta or TikTok reject an AI-generated ad?
Not for being AI-generated. Meta and TikTok both require disclosure labels for realistic synthetic people; YouTube requires disclosure of realistic synthetic content. What gets an ad rejected is the same thing that always did: a fake testimonial, an unsubstantiated claim, or someone’s likeness without consent.
Can I clone my own face or a creator’s face?
Consent-based likeness cloning exists, and the disclosure duty does not go away because the face is licensed; a cloned creator still cannot testify to an experience the clone never had. We do not use it. Our presenters are described by features and generated fresh, and every one is labelled.
References
- 1Billo, UGC rates 2025; Superscale, UGC pricing 2025: real rates, hidden fees, AI UGC Average about $198 per deliverable; usage rights add 30 to 50%.
- 2Arcads pricing as reported by reviewers in 2026 (eesel AI, Superscale, Creatify) Arcads has no public pricing page; figures are from account-holder reports and may change.
- 3Y'all, creative agency cost guide, 2026
- 4Motion, Creative Benchmarks 2026
- 5Meta, expanding GenAI transparency for Meta’s ads products, February 2025
- 6TikTok, AI-generated content
- 7YouTube, disclosing use of altered or synthetic content, March 2024
- 8FTC, final rule banning fake reviews and testimonials, August 14, 2024
- 9Arnold & Porter, FTC issues warning letters over Consumer Review Rule, January 2026 Ten letters dated December 22, 2025; penalty figure of $53,088 per violation.