There is a slide in a lot of agency credentials decks right now, and you have seen it. A glowing node diagram, the word “ideation” in the middle, a little spark icon, and a promise that the agency’s proprietary AI can generate “hundreds of campaign concepts in minutes.” The room nods. Procurement smiles. Somebody asks whether that means the retainer comes down.
It should. Because the thing on that slide is the one part of the job nobody should be paying a machine to do.
That is not a Luddite take from a creative shop with a grudge. It is, as of this week, the position of the people whose literal job title is AI transformation. On November 16, Gabriel Bridger, who leads experience, marketing and AI transformation at Genpact, published a piece through the Fast Company Executive Board with a headline that should be stapled to every pitch deck in America: AI doesn’t belong in the idea. It belongs in everything after it.
Read that again, because of who is saying it. This is not a copywriter fighting for his chair. This is a transformation lead, a person whose career depends on AI working, drawing a hard line around one stage of the process and saying the model stays out.
His map is simple. At intake, AI is a gift: it can pull past performance, brand guidelines, research and competitive scans into a structured brief in minutes instead of days. In production, it is a gift again, because once a human has signed off on the concept, the creative decision is already made and the rest is assets, drafts and formats. In amplification, it is practically mandatory, because one idea now has to live in dozens or hundreds of languages, formats and audience segments. Then the performance data loops back into the next brief.
And in the middle, at the idea itself, he will not compromise. His reasoning is the part every CMO should tattoo somewhere visible: a model-led concept tends toward the generic because the model draws from the average of everything already made.
That is the whole game. That sentence is the business model of the next decade of creative work.
A large language model is, mechanically, a machine for predicting the most probable next thing. That is a beautiful property when you need a product description in 40 languages. It is a catastrophic property when you need an idea, because the most probable idea is, by definition, the one closest to everything that already exists. You are not paying for a concept. You are paying for the centroid.
The industry already has the receipts. When the IPA put out its AI issue in September, it ran a long piece asking whether AI is making every brand look the same, and the most useful line in it came from Brandon Kaplan, COO of McCann New York: “AI doesn’t erode distinctiveness - accepting its first answer does.” The same piece cited Havas research across more than 21,000 AI-powered brand experiences in seven markets, finding that 57% of people think brands use AI to create shortcuts.
Fifty-seven percent. More than half your audience already assumes the shortcut. Which means every concept that smells like a prompt is not just mediocre, it is confirming a suspicion.
Then there was the Super Bowl. According to an Advertising Week analysis in March, 23% of Super Bowl LX commercials either promoted AI or used it to make the ad. The payoff? CivicScience found only 6% of viewers said those AI ads made them more likely to use AI. NielsenIQ’s research described consumers finding AI ads annoying, boring and confusing. The most expensive 30 seconds in American media, and a meaningful slice of it was spent proving that the average of everything is forgettable at scale.
So why does the ideation slide keep showing up in decks?
Because it sells. Speed is legible to a CFO. “Two hundred concepts” sounds like more value than “three concepts.” And because, if we are being honest about our own industry, a lot of agencies have spent years selling volume dressed up as strategy, and AI ideation is the cleanest possible upgrade to that hustle. More output, less thinking, same invoice. That is not innovation. That is bullshit with a progress bar.
Here is the turn, and it is the uncomfortable one.
The fair objection goes like this: humans steal from the average too. Most human-made advertising is derivative. Half the spots on any given NFL Sunday could have been written by a model and nobody would have noticed. So what is the difference between a mediocre human concept and a mediocre machine one?
Accountability. Taste is not some mystical spark; it is a person with a cultural read, a point of view and something to lose, deciding which of ten directions actually says something and then defending that choice in a room full of people who would prefer the safe one. Bridger puts it in plain terms in his own closing: “The idea is human. The scale is AI.” A model cannot be wrong in public. It cannot be embarrassed. It has never stood in line at a Silver Lake taco truck at 1 a.m. and overheard the exact phrase that becomes a campaign. It has never been to the show, read the comments, felt the room go cold. It cannot take the risk because it has no skin to put on the line.
And risk is the job. The ideas that actually move culture are, almost without exception, the ones that looked like a bad idea in the deck. Nobody average-of-everything’d their way into a cultural moment.
There is also a darker economic point that agencies selling AI ideation should sit with. If your differentiator is that a model generates your concepts, your differentiator is available to every competitor and every in-house team with a Copilot license by Friday. You have not built a moat. You have published your own replacement manual and slapped a logo on it.
We use AI every day at BoomHaus, and we are not shy about where. Brief synthesis, research digests, versioning, cutdowns, localization, the grinding production work that used to eat whole weekends. It makes us faster and it makes the work more consistent. But the idea is ours. The cultural read is ours. The decision about which of ten directions is the one worth fighting for is ours. That line is not a marketing position; it is the only thing that makes the rest of the process worth paying for.
Bridger, to his credit, warns that his own framework can calcify into dogma if teams run it on autopilot, and calls it a starting map, not a rulebook. Fine. But the map has one fixed point, and the people who know AI best keep putting it in the same place.
So the next time an agency, a vendor or your own internal team pitches you AI-generated concepts as innovation, ask the only question that matters: if the machine is having the idea, what exactly are we paying the humans for?
If the answer is “speed,” you are buying a faster way to sound like everybody else.
The idea is the one thing you cannot outsource to the average. Everything after it, sure. Go nuts.