AI gave everyone fake taste. It gave no one experience.
If you're a founder or CEO thinking AI means you can run your marketing through an app, let me save you some money: it won't work. AI is one of the best tools marketers have ever had, and it's making most marketing worse. Disagree? Name the last campaign made only by AI that you remember for the right reasons.
I use AI every day, and I'm not giving it back. But back in 2018, I wrote that marketing is an iceberg: people judge it by the shiny tip and forget what's under the water. AI just made everything above the water free. Anyone can now produce a beautiful deck or a polished campaign in an afternoon. What it can't produce is the part under the water: the judgment to know whether the work is true, right for your customer and worth shipping at all.
That's why the first thing people ask is what prompt you used. There isn't one. You prompt, you read, you prompt again, and you ask where that suspiciously convenient number came from. Then you keep about half, toss the rest, and write the part that actually needed a human. It's faster than doing it all yourself, and about as magical as doing your taxes. Nobody makes TikToks about that part. They sell you the magic prompt instead, and if one worked, there wouldn't be ten thousand new ones every week. Marketers have been sold this magic before. It was the guerrilla stunt, then the viral video, then the growth hack. This year it's a prompt.
Why should you care? When good enough costs nothing, it's worth nothing. Marketing that's only the tip of the iceberg sounds exactly like your competitor's, and some of it is simply wrong, with your name on it. Ask AI for a report and you'll have one in seconds. Confident, beautifully formatted, and maybe seventy percent right. The problem is it won't tell you which seventy percent.
And that's with the best tools on the market. Stanford's 2026 AI Index put twenty-six top models through a new accuracy benchmark, and their hallucination rates ran from twenty-two to ninety-four percent. If you can't tell which parts are wrong, AI is actually making you worse at your job. One study found that the more people trusted these tools at work, the less critical thinking they applied. And if your plan is to wait for the next version to fix it, bring a book: computer scientists have published a formal proof that hallucination can't be fully engineered out.
The AI companies say it themselves. OpenAI's Greg Brockman posted this year that "taste is a new core skill," and Sam Altman wrote that the best research teams are built through "context, taste and a real feel for where the field is headed next." In April, Anthropic told Axios that the taste in its new Claude model meant what "an experienced practitioner in that domain would consider great work." The companies building these tools hire people for taste and grade their models against experienced practitioners. That should tell you something.
The real magic of AI is knowing what to say no to, and it starts with the question behind the prompt. Ask a lazy one and you get an eloquent average of everything already said on the subject. Ask a specific one, with a real customer in mind and a reason for them to care, and you get something worth editing.
When your AI tool spits out its confident answer, read it like a reference written by someone's mother. Is that quote real? Is that your brand talking, or the internet? Would a customer believe it, or roll their eyes? Every so often there's a line arguing against your own point, in perfect grammar. Most of the work is making the tool justify itself.
A bland first draft is usually a review of your brief. Plenty of people try a tool once, get something generic, and then either ship it or announce on LinkedIn that AI is overhyped. More often than not, the problem is the person typing.
The part of the iceberg under the water comes from experience, and experience is scar tissue. It's what's left after you burn yourself, again and again: the website that read like a perfect fundraising pitch and left customers asking what's in it for them, the subject line you loved that nobody opened, the proposal that came back with "thanks, but we're going in a different direction." You get it by being wrong, paying for it and working out why. The machine has read about every one of those burns and felt none of them, which is why it can't tell when something is slightly off.
So, founders and CEOs: keep the app. Then put an experienced marketer in charge of what ships. AI will happily hand you a thousand versions. Knowing which one to keep takes experience, and experience doesn't come with a free trial.
