We accidentally removed the waiting period
There used to be a useful safety feature in advertising: making things was annoying.
A bad idea had to survive a meeting, a budget, a storyboard, casting, production, editing, legal, and at least one person quietly asking, “Are we sure?”
The process was expensive and slow. It was also a filter.
Bad ideas still escaped, obviously. Advertising managed to make terrible work long before anyone could type /imagine. But bad ideas used to require commitment. They had to remain bad for several weeks before the public was allowed to see them.
AI has changed that.
Now a bad idea can become a script before coffee, a storyboard before lunch and a disturbingly plausible video before somebody has had time to develop doubts.
AI has not removed the bad idea. It has removed the waiting period.
AI is making production cheaper, faster and more accessible. The scarce resource is shifting from the ability to make something to the ability to decide whether it deserves to be made.
Production is becoming abundant
Stanford’s 2025 AI Index reported that the cost of running a model at roughly GPT-3.5-level performance fell more than 280-fold between November 2022 and October 2024. In a field experiment involving 758 Boston Consulting Group consultants, GPT-4 made people more than 25% faster on tasks inside the technology’s capability frontier.
The production advantage is real.
The problem begins one step earlier.
AI can generate twenty campaign routes in the time it once took to schedule the meeting about the campaign routes. Unfortunately, it can also generate twenty versions of an idea that should have died in that meeting.
We spent years treating production capacity as the bottleneck.
Increasingly, it is judgment.
Toys “R” Us discovered that Sora cannot save the story
In 2024, Toys “R” Us released a brand film made with OpenAI’s Sora. It dramatized the childhood of founder Charles Lazarus, complete with a young boy, a toy-store dream and a glowing fantasy world.
The technology was the headline. The reaction was less futuristic.
Fast Company called the result creepy and argued that the problem was not simply Sora. The central story itself was not especially interesting.
That distinction matters.
When an AI character has strange fingers, we know where to look. When the entire premise is dull, the prompt cannot be arrested for it.
AI can improve execution. It cannot retroactively make the brief worth answering.
Google automated the only part that mattered
Google’s 2024 Olympics commercial “Dear Sydney” showed a father using Gemini to help his daughter write a fan letter to Olympic champion Sydney McLaughlin-Levrone.
Technically, this demonstrated the product.
Strategically, it demonstrated something else.
The meaningful part of a child writing a letter to her hero is that the child writes it. The spelling mistakes are part of the product. The awkward sentence is the feature. Nobody opens a fan letter hoping for optimized copy.
The ad drew criticism for using AI to shortcut genuine human expression, and Google pulled it from its Olympics rotation.
Someone saw friction and removed it.
Unfortunately, the friction was the point.
Coca-Cola made nostalgia faster
Coca-Cola’s 2024 AI-generated holiday commercials revisited the visual world of its famous “Holidays Are Coming” advertising.
The response included mockery over uncanny movement, distorted details and the feeling that something human had been removed from advertising built around warmth and nostalgia.
Coca-Cola continued using AI in 2025. The Verge reported that the newer production involved around 100 people, five AI specialists and more than 70,000 generated clips. The company said the process could cut production from roughly a year to about a month.
That is an extraordinary increase in speed.
It also creates a useful modern question:
We generated 70,000 clips, but did we generate one reason to care?
Saving eleven months is valuable only if the destination is worth reaching.
A faster train does not improve the town.
McDonald’s found a quicker route to “please remove this”
McDonald’s Netherlands supplied an even cleaner example in 2025.
Its AI-generated Christmas commercial called the festive season “the most terrible time of the year” and showed holiday disasters before suggesting people retreat to McDonald’s until January.
The campaign attracted criticism online and was removed.
Again, AI was visible in the execution, but the larger problem was conceptual. “Christmas can be stressful” is a recognizable human observation. “Christmas is terrible, hide in McDonald’s” is what happens when an observation is stretched until the strategy snaps.
The technology made the commercial possible.
Judgment made it public.
We removed production friction. We may discover that some of that friction was governance wearing an ugly shirt.
Heinz shows why this is not an anti-AI argument
There is a temptation to conclude that audiences simply dislike AI advertising.
That is too easy.
Heinz’s “AI Ketchup” campaign used DALL-E to ask a better question: when AI is asked to draw ketchup, does it produce something that looks like Heinz?
Here the technology was part of the idea, not a substitute for one. The strange images became evidence for a clear brand proposition: Heinz is so synonymous with ketchup that even a machine appears to recognize it.
The campaign later received a D&AD Graphite Pencil.
Same broad technology. Very different judgment.
AI does not make advertising bad. It makes whatever sits upstream of production more consequential.
AI raises the value of the person who says no
The most interesting implication for companies is not that creative departments need more prompt engineers.
They may need better editors.
When production was expensive, organizations naturally limited what they made. AI changes the question from “What can we produce?” to “What should we allow through?”
The BCG experiment offers a warning. On tasks where GPT-4 was well suited, participants performed better. But on a business problem-solving task outside the model’s capability frontier, people using AI performed 23% worse than those without it. The researchers also found less diversity in the ideas produced by AI-assisted participants.
The danger is not simply that AI gives bad answers.
It is that AI can give bad answers fluently, quickly and in the correct aspect ratio.
That makes judgment less visible at precisely the moment it becomes more valuable.
The future creative advantage may belong to organizations with the strongest filters: people who understand the audience, the culture, the brand and the timing, and who can still ask the uncomfortable question after the tenth beautiful render appears on screen.
Should this exist?
Not “Can we make it?”
Not “Can we make six versions by 4 p.m.?”
Just: should it exist?
For a long time, the expensive part of a bad idea was making it.
Now the expensive part may be realizing it was bad after everyone has already seen it.