McDonald's Netherlands spent five weeks on an AI-generated Christmas ad. Ten people, full-time, according to the production company's own since-deleted comment about it: blood, sweat, tears, and an honestly ridiculous amount of coaxing to get the models to behave and to honor the creative brief shot by shot.
It still got pulled within days. Viewers called it creepy, soulless, and depressing. McDonald's pulled the ad just three days after publishing it.
Here's the part that should actually concern anyone using AI for design work: this wasn't a team that skipped the work. Five weeks of coaxing is not "we typed a prompt and shipped whatever came out." And it still failed in public, badly enough that McDonald's told BBC News the ad was an important learning for the company's understanding of how to use AI effectively, which is corporate for "we got this wrong and we know it."
So if effort wasn't the problem, what was?
The tool isn't the variable. The brief is.
I think the McDonald's ad failed for the same reason a junior designer's first unsupervised project fails, not from lack of effort, but from lack of a real creative spine to work from. A brand guideline isn't a style preference. It's a set of constraints that exist specifically to stop a piece of work from drifting into whatever's easiest or most generic for the tool to produce. Take that constraint away, human or AI, and the work drifts toward the path of least resistance every time.
With a human designer, that drift looks like generic stock-photo energy, safe choices, nothing offensive but nothing intentional either. With AI, the drift looked like the uncanny valley: characters that moved awkwardly, had inconsistent facial features, and mismatched props in the background, creating an eerie effect. Different failure mode, same root cause. Nobody told the system what it absolutely could not become.
Five weeks of coaxing a model to behave isn't creative direction. It's wrestling with the tool after the fact instead of constraining it from the start. That's the actual mistake, and it's the same mistake whether you're directing a person or a model: react to bad output instead of defining good output before you begin.
This isn't really about AI
It's worth saying plainly: the brands that catch the most public heat for AI slop usually aren't the brands using AI carelessly behind the scenes. They're the brands using it as the headline. Coca-Cola, after its first AI-generated holiday ad was criticized in 2024, didn't retreat. It released a second one, which tells you the failure wasn't really about the technology. It was about treating "made with AI" as the pitch instead of treating the actual creative output as the pitch.
What this actually means for design work
If a brand's visual identity, tone, layout logic, and creative boundaries exist somewhere real, written down, specific, not just feel free guidance, AI can work inside that system the same way a contractor works inside an architect's plans. It can move faster without drifting, because the constraints are doing the work a human's taste would normally do silently.
If those guidelines don't exist, AI doesn't fail loudly the way a junior designer's rough draft fails loudly in a review. It fails quietly, confidently, and at scale, because nothing in the system is telling it where the edges are. You don't find out until it's already live and the internet is calling it slop.
The lesson isn't don't use AI for design. It's the same lesson that's been true since long before AI existed: undirected creative work drifts toward generic, no matter how much effort goes into it, and no amount of effort after the fact fixes a constraint that was never defined at the start.