OpenAI Business CMO Colin Fleming Says Everyone Can Do Marketing Now. So Who Gets to Say No?

Every marketing team has a version of the meeting Colin Fleming describes. Someone brings an idea. Then come the questions. Who builds it? How long does it take? What are we killing to make room? Is there budget?

The idea doesn't get rejected…It just never gets scheduled.

Fleming, chief marketing officer for business at OpenAI, wrote about that meeting this week, and his argument is that it's over. AI has collapsed the distance between having an idea and finding out whether it's any good. A marketer can build the prototype. A salesperson can build the campaign. Nobody has to sit in a specialist's queue to get started.

He's right, and the piece is a lot less breathless than most things a CMO writes about their own company's product. He's blunt that AI makes it easy to churn out what he calls "perfectly competent garbage," and he holds a real line on quality: if you could swap another company's logo onto the work and nothing would change, the work is wrong.

The trouble is the constraint he picks. Fleming says the thing holding marketing back was access, meaning permission to start. It was cost. And cost was quietly doing a second job that nobody ever assigned it.

Nobody chose this filter, but it worked

Production expense wasn't only a tax on marketing. It was a filter. A dumb one, frequently an unfair one, but it caught things.

It's why companies ran the same visual system for four years instead of refreshing it every other quarter. It's why plenty of forgettable ideas died in a resourcing conversation rather than in front of a customer. The repetition that builds brand memory got manufactured almost by accident, because making something new cost more than reusing what was already sitting in the drive. Nobody had to win an argument for consistency. The budget won it for them.

Take the cost out and that stops happening by itself. Consistency turns into a call someone has to make, and then defend, against an endless supply of alternatives that are competent, on brief and free. Almost nothing in a marketing org rewards making that call. Saying no is the only marketing activity that leaves no artifact behind.

Fleming gets close to this. He writes that marketing moves from controlling who gets to participate to setting the standard for what deserves to ship. In a lot of companies that's going to play out as a demotion wearing a promotion's clothes, because a standard nobody is obliged to obey is just a senior person's opinion. The CMO keeps full accountability for whether the brand hangs together. Every other function is shipping customer-facing work, and none of them report to marketing.

So refusal becomes the scarce thing, and it has gotten much harder to do, because the work you'd be killing took someone forty minutes and cost the company nothing. That reduces the case for shipping it to a shrug. Why not? Brands don't usually die of bad work. They die of a great deal of fine work that never stacks into anything you'd recognize across a room.

The editor in chief turning up on marketing org charts is an early attempt at a fix. That job isn't there to add output. It's there because somebody needs the standing to spike things.

The synthetic audience is the risky part

Fleming mentions, almost in passing, that OpenAI is testing synthetic audiences built from real customer research, and that he ran a draft of his article past a synthetic CMO. It told him to sharpen the business case.

He hedges the right way, noting that a synthetic audience is only as good as what real customers told you in the first place. But quality isn't the danger here. Availability is. A synthetic panel is sitting there at eleven on a Tuesday night, endlessly patient, and it will never tell you the whole premise is broken, because the premise is what you fed it.

Customer research is slow for the same reason it's useful. Real people are a nuisance. They misread positioning that took a quarter to write. They shrug at the feature the roadmap is built around. They raise things nobody thought to ask about, which has always been the only input that actually changes a strategy. The friction cost money. It was also the only dependable way to make contact with something outside the building.

Run your own research back through a model and you get the shape of what you already had, smoothed toward the average of everything else it has read. Ask for feedback and it will suggest tightening the business case, because that's what feedback sounds like.

Useful for pressure-testing a message before you spend against it. Not much use for discovering you're wrong, and most brand positions worth owning came out of somebody discovering they were wrong.

There's a category-level version of this that should bother anyone in a crowded market. If four competitors all run their positioning past models trained on roughly the same public internet, they get nudged toward the same defensible middle. Ask all four to describe themselves and you'll get one answer.

A finding that cuts both ways

Fleming cites OpenAI research showing marketing tasks turning up in other people's jobs more than tasks from any other field in the study. His favorite detail is that salespeople sent more ChatGPT messages about marketing than about sales.

He takes that as evidence marketing is universal, that storytelling is a skill rather than a title. There's a less flattering read available. Marketing tasks spread furthest because they were the easiest to get at. No license required, no compliance sign-off, no decade of knowledge living in somebody's hands. Work that can be described in a paragraph is work that travels, and if marketing topped the list, some of what got measured is how much of the job was already describable.

That doesn't sink the argument. It raises the cost of getting it wrong. The describable half of the job now belongs to the entire company. The half left over is the part that mattered anyway: knowing the customer well enough to tell when a room full of executives is talking itself into something, and having the nerve to sit on a position for three years while it does its slow work.

One number deserves a caveat. Weekly active enterprise Codex users on marketing teams up 26 times since February, with 39% of messages involving code and 12% involving writing, says something real about direction. It also describes the marketers who went and adopted a coding tool. Of course the usage skews toward code. That's who showed up.

The bit at the end is the actual story

A few lines near the close deal with advertising inside ChatGPT, and the prospect of a company getting found while a buyer is still weighing options and working out who to trust. It reads like a coda. It's the most important thing in the piece, and it's the reason the rest of the piece matters.

Brand used to be what a company said, plus what customers experienced, plus what the trade press wrote. A growing chunk of it is now whatever a model says when a buyer asks it to compare three vendors. That answer gets assembled out of sources the company didn't write, weighted in ways nobody outside can inspect, and handed over as one paragraph with no blue links to work on and no obligation to show its reasoning.

The traffic numbers have pointed this way for a while. Fewer than one in three Google searches sends a click anywhere now, per SparkToro's 2026 read of Similarweb data. AI referral visits more than tripled year over year, and when ChatGPT started linking directly to brands inside its answers in May, referral traffic jumped more than 150% in a week.

Somebody else is holding the brand now, which is a bigger development than a change to the funnel. The strange part is how conventional the response has to be. Models summarize toward the middle, so any position loose enough to be flattened into a category description will get flattened. The claims that survive are narrow, specific and repeated in more than one place: on the company's own site, in coverage it didn't pay for, and out of the mouths of customers willing to attach their names. Say a small thing, say it everywhere, say it for years. None of that is new advice.

The takeaway

Fleming's main call is sound, and any marketing team that answers all this by guarding the perimeter, locking down tool access or insisting the department owns every touchpoint deserves what's coming. Good ideas have never cared about reporting lines.

The essay just stops a beat early. Once everyone can make the thing, the person you can't replace is the one who looks at work that's competent, on brief and free, says no to it, and doesn't get overruled by lunchtime. Most marketing organizations have no idea whether they have that person. For thirty years the budget did the job for them.