The Center Point: AI Cannot Have Your Idea
By Jake the WizardAI is not a creative. It is an aggregate. It hands back the most likely answer, the most likely answer is the average, and that is why 86% of marketers have already seen AI output that resembles a competitor’s. It cannot have your idea. It can only build it.
A podcast host asked me what AI was going to do to marketing and I gave him a line I have not had to change since: when you use AI, you’re accepting the center of the bell curve, and it takes a human to connect with humans. That was a hunch with a nice ring to it. In March 2026 two researchers published the receipt.
Why does every brand’s AI creative look the same?
Because the model hands back the most likely answer, and the most likely answer is the average of everything it has read. That is prediction, not thought. It does not originate. It recombines.
Emily Wenger and Yoed Kenett ran 22 large language models against more than 100 people on three standard creativity tasks: list every use you can think of for a fork, produce nouns as distant from one another as possible, follow a chain of word associations. Individual models sometimes beat individual humans. As a population the machines collapsed inward. Model answers looked far more like other model answers than people’s answers looked like other people’s. The title of the paper says it without decoration: large language models are homogeneously creative, PNAS Nexus, March 2026.
Now move that into an ad account. Smartly surveyed 450+ marketers for its 2026 Digital Advertising Trends report. 46% use AI to scale creative. 86% have already seen AI output that resembles a competitor’s. Three in four worry the category is starting to look and sound identical. That is the Duke study happening live, in real budgets, with real money behind it.
Can a better prompt fix the sameness?
No. The researchers pushed the models toward novelty and variability moved a little. Humans still won. A better prompt gets you a better average. It does not get you a new thought.
This is the part the prompt-library crowd keeps skipping. You can steer a model closer to your taste. You cannot hand it taste. The steering wheel is real and the engine is still an averaging machine.
The trap is that the average looks creative
Take one unoriginal idea, wrap it in another, and the result reads like a leap. Average two things that already exist and you get something that feels new to anyone who has not seen both. That is a costume, not an idea. Fluent, competent, forgettable.
Feed the same model the same brief as your competitor and you get the same ad. Not because you share a strategy. Because you share a machine, and the machine has one center of gravity. Nothing about running it harder pulls you off that center. The tool sold to help you stand out is built to walk you back toward everyone else.
So what is AI actually good for?
Building. It is the best builder I have ever had and a weak originator, and both are true at the same time. Once your idea is set, hand it the static work and let it move fast.
| Bring it yourself | Hand it to the machine |
|---|---|
| The idea nobody else has | The forty variants of it |
| The emotion you are leading with | Resizes, cutdowns, reformatting |
| The stance, including who it annoys | Research summaries and rough drafts of the dull parts |
| The final no on what ships | The first 90% of the fill |
Look at the right column. Every job there is one where the average is the correct answer. A resize does not need a point of view. A headline variant of a line you already wrote is labor, not authorship. That work used to eat the week that should have gone to thinking, and now it takes an afternoon. Capability went free, which is why the effort and taste behind the work became the product.
The left column is one input long, and no vendor sells it.
How do you know if you were the center point?
Run this on your last campaign. Delete every line the machine produced. Look at what is left on the page.
If an idea is sitting there, one thing that came out of your head and could not have come from a training set, you were the center point and the tool built underneath you. If the page is blank, you were the operator. The machine handed you the middle of the bell curve with your logo on it, and it will hand your competitor the same middle next Tuesday.
That idea goes on top as the load-bearing wall, not as decoration over the machine’s output. Same logic that runs the Eight Dominoes: the first move decides what every move after it is allowed to be. Get the order backwards and the volume underneath does not save you.
The falsifiable version
The claim, stated so it can be wrong. Two competitors in one category, same model, same brief, no human idea on either side, will produce work a customer cannot tell apart. If those two campaigns come out different in any way a buyer would notice, I am wrong and the sameness problem is a skill issue.
I do not expect to be wrong on this one. 86% of marketers have already run the experiment by accident and reported the result.
The tool is identical for everyone. Your head is not. That is the whole tiebreaker, and it does not come in a prompt pack. No spell here. Bring the idea, put it at the center, and let the machine build underneath it.
Straight Answers
Why does AI-generated creative look the same across different brands?
Because the model returns the most probable answer, and the most probable answer is the average of everything it has read. Two competitors using the same model on the same brief are drawing from the same center of gravity. The sameness is not a prompt failure. It is what a prediction engine does.
Can better prompting make AI output more original?
No. Researchers at Duke prompted 22 models toward higher creativity and variability barely moved. Human responses still varied more. A better prompt gets you a better average, not a new thought. You can steer a model closer to your taste. You cannot hand it taste.
Is AI useless for creative work?
No. It is a strong builder and a poor originator. Point it at work where the average is the correct answer: variants, resizes, cutdowns, the fill around an idea that already has a shape. Point it at the idea itself and you get competent work with no fingerprint on it.
What does being the center point actually mean?
You supply the idea, the machine builds underneath it. The idea comes from the one place a training set cannot reach, which is your own head, and it goes on top as the load-bearing wall. Everything the tool produces gets cut to that line instead of replacing it.
Does using AI hurt brand differentiation?
It does when the tool decides the idea. Smartly surveyed 450+ marketers for its 2026 trends report: 46% use AI to scale creative, 86% have seen output that resembles a competitor’s, and three in four worry brands are converging. The same tool with a human idea on top does the opposite.