48% of ad buyers think viewers reject AI creative. (Buzzer.) They reject the label.

Four researchers at NYU Stern and Emory ran the experiment the advertising industry keeps arguing about instead of testing. Fully AI generated visual ads outperformed ads built by human experts, lifting click through by as much as 19%. Then the same ads were disclosed as AI made. Click through fell 31.5%.

The number the industry reacted to this summer was a softer one. FreeWheel, Comcast's ad tech arm, found that 48% of media buyers assumed viewers would find AI generated creative off putting, while only about 10% of viewers who saw one said they did. The trade read was that buyers have lost touch with their own audience. The data does not support that reading. Line up the three studies now on the table and the audience is perfectly consistent. It is answering three different questions.

Nineteen up, thirty one point five down, same creative

Anindya Ghose, Hyesoo Lee, Vilma Todri and Panagiotis Adamopoulos compared three kinds of ad: human expert work, human work modified by generative AI, and fully machine generated. The middle option, human work touched up by AI, produced no significant improvement. The fully generated ads won. Disclosure then took away a third of the result.

The performance is real and the penalty is real, and they attach to the same asset. Between the two results the creative is identical. The only variable is whether anybody said the word.

Gallup and FreeWheel fielded a month apart and got opposite answers

Gallup surveyed 3,270 US adults on its probability panel between May 4 and 11, margin of error 2.4 points. It found 62% consider it unacceptable for a business to use AI to create people or voices, rising to 73% among adults under 30, with 49% viewing AI in advertising negatively. FreeWheel's viewer sample, fielded by Dynata in April with 2,496 respondents, put dislike at 10%.

Both are sound, and they measure different things. Gallup asked people to judge a practice. FreeWheel asked people to judge an ad. A practice gets a verdict; an ad gets a shrug.

FreeWheel is not a neutral instrument here. It launched AI agent infrastructure for buying and selling premium video in March, so a study from that house reporting viewers relaxed about AI creative arrives with a commercial interest attached. Gallup has nothing to sell. And the NYU work measured behavior rather than opinion, which makes it the only one of the three that watched what people did instead of what they say.

The industry had this in June and bought humanity anyway

FreeWheel released the report at Cannes and VideoWeek wrote it up on July 9. Through the rest of the summer the humanity campaigns kept arriving: Anthropic's 90 second Claude spot, Meta's late July film, a Garage Beer and Liquid Death musical satire, with Etsy, TD Bank, Polaroid, Pinterest and Heineken working the same seam. Andy Pearson, VP of creative for Liquid Death, told Marketing Brew that having to prove you are human is "sort of like the real-life version of the Captcha stuff."

He is right, and the Captcha is the whole problem. The audience liked the machine's work right up until somebody told them a machine made it. That is not a finding about advertising. That is the business model of every magic show.

A claim you cannot survive having labeled

The IAB knows the price. Version 2 of its AI Transparency and Disclosure Framework, published August 18, drops blanket labeling for a risk based model, cites the 31.5% figure in support of doing so, and asks advertisers to name an AI Disclosure Lead within 60 days on a 24 month roadmap. Read plainly, the trade body's position is that disclosure is expensive and should be rationed to the cases where it prevents real deception.

Now read a humanity campaign against that. The brands spending 2026 advertising how human they are have chosen the one claim whose value collapses the moment it is audited, in a market that has just appointed compliance officers to do the auditing.

When LinkedIn shipped a button to filter machine written posts in August, we argued that a platform had repriced authenticity overnight. This is that repricing with a number attached. The same mechanism sat under our reporting on reach that buys trial and not a second purchase, and under last week's finding that the four most effective prospecting channels still require a specific human to spend a specific hour. Authenticity now behaves less like a creative asset and more like a disclosure liability, and the liability is priced at 31.5%.

Every brand that bought a humanity campaign this year holds an unhedged claim, and the first label to land beside it takes a third of the response with it. Two brands in that same summer roundup are already hedging with media rather than copy, Pinterest through offline experiential work and Polaroid through a pro analog push. Money spent putting actual people in front of actual buyers carries no disclosure exposure, because there is nothing left to disclose. One of those two budgets gets repriced in 2027, and it is not the one with the handshakes in it.