AI-generated creative doesn’t look nearly as weird as it used to. Sure, there’s still the occasional extra hand or mangled logo, but most of the time, it looks completely normal.
And that’s almost weirder. Everything can be technically right and still feel slightly…off. Maybe everyone is smiling a little too much, or the room looks more like the idea of a kitchen than somewhere a person has actually cooked. Sometimes the copy just sounds exactly like something a brand would say, which is sort of the problem. Part of taste is recognizing when something feels off. Putting your finger on why is harder.
Taste is one of those words marketers use constantly. The same goes for “good” creative. Good can mean original, funny, beautiful, memorable, or impossible to scroll past. It can also mean the ad got someone to buy something, which tends to become a fairly persuasive definition once money is involved.
An ad can feel right to the people making it and fall flat with the people it needs to reach. AI gives marketers more ways to see how the work is likely to land before it goes live. That can be reassuring right up until the prediction disagrees with you.
Defining Creative Effectiveness in an AI-Driven Environment
Creative effectiveness is how well an ad or piece of creative produces its intended response, whether that’s attention, recall, engagement, brand lift, or conversion.
Marketers have plenty of ways to measure those outcomes once an ad is in market. AI can use data from past campaigns to help evaluate the next one before it launches. By analyzing creative alongside audience and performance data, it can find patterns across messages, visuals, and formats, then estimate which combinations are most likely to work for a particular audience.
Those estimates give the creative team another way to weigh its options. There will still be strong opinions about typography and requests to give the layout more room to breathe. The team also has evidence to help judge how those choices might land with the intended audience. Suddenly, “I just like this one better” has some competition.
How Audience Intelligence Changes What Works
Any definition of “good” gets more complicated once you account for the audience. People bring different interests, expectations, and reasons for paying attention.
At RAD, our AI-driven platform uses audience intelligence to identify the different groups within a brand’s audience and what they’re likely to respond to. And those groups don’t necessarily care about a product for the same reasons. What looks clever to one group can look painfully forced to another. Seeing those differences can challenge the team’s assumptions before the creative goes live.
The people making the ad and the people watching it can also see the work very differently. Creative teams bring trained eyes, professional standards, and strong preferences. Audiences bring their own associations with the brand. Unfortunately, they don’t always find the same things brilliant.
Coca-Cola’s AI Holiday Ad: Critics vs. Consumer Testing
Coca-Cola put that tension on display in 2024, when it remade its famous “Holidays Are Coming” campaign using generative AI.
The familiar ingredients were all there: red trucks, snowy scenes, the music and a holly-jolly amount of festive cheer. People had spent years associating those sights and sounds with the start of Christmas. That nostalgia was part of what the remake needed to preserve.
The critics, to put it mildly, were not feeling warm and fuzzy. They called the ad creepy, cheap, and “soulless,” picking apart the artificial imagery in articles and comment threads. A familiar holiday ritual had become an argument about whether AI had taken the warmth out of it.
Consumers in System1’s testing, however, responded very differently. System1, which uses consumer responses to predict an ad’s long-term brand-building potential, gave the AI version 5.9 stars, the highest score its model awards. Participants weren’t told the ad was AI-generated or asked to compare it with the original.
System1 pointed to decades of familiarity with the trucks, the song, and the story as reasons viewers still connected with the ad. Whatever critics thought the new production had lost, enough of that familiar feeling remained to resonate with consumers.
Coca-Cola returned with another AI-generated version in 2025, which also earned 5.9 stars. For a remake like this, preserving the emotional connection has to be part of the brief. Whatever tools make the new version, it still has to bring back the feeling people loved in the first place.
How AI-Generated Ads Perform With Consumers
That emotional connection is also where other AI-generated ads have struggled. And viewers don’t necessarily have to recognize how an ad was made to respond differently.
In research published in May 2026, Ipsos and Syracuse University’s Newhouse School tested 20 ads with 3,000 U.S. consumers. Ten were existing human-made ads and ten were AI-generated alternatives. Only 25% of viewers who saw an AI ad were at least somewhat confident it was AI-made.
Even so, the AI ads scored lower on Ipsos’ measures of predicted short- and long-term effectiveness. The researchers found particular weaknesses in emotional storytelling. Viewers’ difficulty identifying AI hadn’t erased the differences in how they responded.
It’s a good reminder that a lot can get lost in seemingly small creative choices.
A person’s expression can be slightly wrong for what they’re saying. A joke can make sense and still be excruciating coming from a particular brand. A line can check every box in the brief and sound ridiculous once somebody says it out loud. Everything can look right and still land wrong.
Where Human Judgment Still Matters
Taken together, the two examples show how often testing and taste can tell different stories, and how easily either one can miss part of the picture. Coke’s critics objected to the ad’s artificial look, while tested viewers still connected with familiar holiday imagery. In the Ipsos study, viewers struggled to identify AI-made work, yet the human-made ads prompted stronger responses.
If the creative you love is predicted to underperform, the next step is to understand why. The analysis may have caught something the team missed, or it may be missing something about the brand, audience, or context. The same goes for criticism. If people keep describing the work as artificial, flat, or strangely emotionless, there may be something useful buried in the comments, however irritating the delivery.
Human judgment still has to sort through all of it. AI predictions and audience criticism are both evidence to investigate, not verdicts to follow. Marketers may look at the prediction, the test results, and the criticism and still choose the other ad. They should be able to explain why, what they expect it to do, and what would make them reconsider. After all, nobody said having more information would make the decision easier.
