AI tools have collapsed the cost of producing content. The smart response is giving real creators more room to sound like themselves.
Scroll for five minutes on Instagram or TikTok right now, and there’s a decent chance you land on an AI-made video. That’s not an exaggeration: Kapwing‘s research found that 59 percent of videos served to a brand-new TikTok account are AI-generated, nearly three times the rate on a fresh YouTube feed. A friendly face, a sunlit kitchen in the background, walking through “what I noticed after 30 days” of some supplement or serum. It feels personal, which is the point.
The New York Times recently profiled dozens of ads built exactly like that one, except the face wasn’t real. The doctors, healers and everyday women in these videos are AI-generated, built to sell supplements using fabricated before-and-afters. One woman the Times interviewed, a 71-year-old retired truck driver managing an autoimmune condition, bought a supplement after seeing ads like these on Facebook. It was recalled months later after federal regulators linked it to salmonella contamination.
That story is disturbing on its own terms, but the bigger detail is the economics behind it: one ad shop profiled in the piece produces around 1,200 AI videos a day, at roughly $10 each, deliberately styled to look American. Believability, at a certain level of polish, now costs almost nothing to manufacture.
That’s a preview of the fork every brand is standing at right now.
Cheap content, tighter scripts
AI has done the same thing to legitimate creative work that it did to the fraud that was profiled: it dropped the cost of producing something that sounds like a real person talking. Ten hook variations that used to take a team a week now take a prompt and an afternoon. Creative briefs that used to need a full planning cycle can get a first draft in minutes.
The instinct I see brands act on is the wrong one. Faced with tools that can generate more content, faster, many respond by tightening the leash: more required lines and less room for a creator to talk in their own words. It reads like control, but it’s the opposite of what the moment calls for. If the raw material of believable content just got cheaper for everyone, including the people faking it, the one thing that stays expensive is a real person saying something in a way only they would say it. Scripting that out of a creator doesn’t protect the brand. It erases the one asset AI can’t touch.
Consumers are getting sharper, too
That scepticism isn’t limited to obviously fake AI avatars. Research published in the Journal of Consumer Research found that simply disclosing AI involvement can reduce engagement with a piece of content, and audiences are getting sharper at spotting the tells even without one. You can see brands responding to it throughout this year: Aerie’s spring relaunch opens inside an AI interface before cutting to an unretouched real shoot, Equinox ran deliberately ugly AI imagery next to real, un-edited bodies back in January, and Almond Breeze had the Jonas Brothers reject a string of absurd AI concepts on camera before landing on a plain pitch. Three unrelated brands landed on the same insight this summer: the fastest way to earn trust right now is to show your work wasn’t manufactured.
The same logic applies to creator content, even when no AI is involved. Consumers are more sceptical of an AI-generated doctor selling kidney supplements, but they’re also more sceptical of a real creator reciting the tenth version of the same brand-approved line this month. The bar for “does this sound like a real person” is rising everywhere at once, and an overproduced, heavily scripted creator post is starting to fail that test in a smaller version of the same way a synthetic avatar does.
Brief the outcome, not the line
The fix isn’t more polish. It’s less scripting, in the right places.
A brief that hands a creator word-for-word lines is optimizing for control, not belief. A brief that states the outcome (what the audience should understand and feel after watching), the required claims and disclosures, and then gets out of the way, is optimizing for the thing that actually converts: a person who sounds like they’re talking to a friend, because they are.
The test I use with creators is simple. Read the content out loud. Does it sound like something a person would actually say to another person about a problem they both understand? Or does it sound like a brand talking at an audience through someone else’s face? The second one is getting easier to spot, whether the face is AI-generated or just over-directed.
What this looks like in a real brief
Picture two versions of the same assignment. Version one hands a creator a script: say the product cleared up your skin in two weeks, mention the discount code twice, hold the bottle at the fifteen-second mark. It gets filmed exactly as written, and it reads like it.
Version two hands the same creator an outcome (a new customer should understand what the product does and feel like it’s worth trying), the non-negotiables (the discount code, the required disclosure, one specific claim legal has approved), and nothing else. What that creator says, which detail they land on, which demonstration they think to show, is theirs to find. That’s the version an AI-avatar testimonial can only ever fake badly.
Fake believability got cheap but real belief didn’t. AI is going to keep getting better at sounding like a person. The brands that scale in this environment are the ones who never had to fake it in the first place.
Feature image credit: Getty Images