Why I Told a CEO None of His AI-Generated Ideas Would Work

A CEO client handed me his laptop and turned the screen toward me. “AI wrote these in about five minutes,” he said. “Which one do you like?” Ten campaign theme options for a new product he was launching, right there on the screen.

I read through all ten. They were polished. Grammatically clean, well organized, nothing technically wrong with any of them. They also sounded like they could belong to any company in his industry. Two of them could have swapped places with a competitor’s website and nobody would have noticed. I told him that, as directly as I could. He pushed back a little at first, then he agreed.

That’s what AI does when you hand it a prompt with nothing behind it. It doesn’t fail loudly. It produces something polished that says nothing. Ten different ways of sounding like everyone else.

He asked what to do instead. I reminded him of the work we’d done together a month earlier, the actual conversations with his customers, the specific things they’d told us about why they buy and what they’d been missing. He asked me to go build the campaign theme from that instead. I did. It worked, and it didn’t sound like anyone else’s.

That’s the part people miss when they talk about AI and marketing. AI is genuinely bad at writing, in the sense that matters. It can produce clean sentences all day. It can’t tell you what’s true about your customers, what’s distinctive about your business, or what story is actually worth telling. Without insight behind the prompt, AI has nothing to draw on except the same patterns every other ungrounded prompt pulls from. That’s why ten options from ten different companies running the same kind of prompt start to sound like one option, repeated ten times.

The skill that mattered in that scene wasn’t picking the best of the ten. None of them were worth picking. The skill was recognizing that all ten had the same problem, and knowing exactly where to go to fix it. That second part is judgment built on experience: knowing that real insight already existed, knowing where it was sitting, and knowing how to turn it into something worth publishing.

This is the conversation I’m having with more CEOs now: whether their teams have the judgment to recognize generic output when they see it, and the discipline to go find real insight instead of tweaking the prompt and hoping for a different result. AI will hand you ten options in five minutes without blinking. It won’t tell you that all ten sound the same, and it won’t go interview your customers for you.

That’s still a human job and it’s more valuable than ever.