The brief that landed at 4 p.m. on a Wednesday

A client needed headline options for a launch campaign, fast. Their internal deadline had moved up, ours moved with it, and the brief that usually gets a day of quiet thinking got about ninety minutes. I could have put a writer on it and asked for the impossible. Instead I opened ChatGPT, pasted in the brand voice notes and the campaign brief, and asked it to generate twelve headline directions.

I want to be honest about why. Not because I thought the tool would write something better than my team could. Because I needed raw material fast, and raw material is exactly what a large language model is built to produce. It doesn’t get tired. It doesn’t need lunch. It will hand you twelve options in the time it takes to make coffee, and every one of them will be technically correct and confidently written. That’s the whole trick of these tools. They’re confident even when they’re wrong, and confident even when they’re just boring.

So I had twelve headlines on my screen in about four minutes. Then the actual work started.

Eleven headlines went in the trash

Here’s what I read, more or less. A handful leaned on the exact language the client’s last three competitors had already used, the safe, expected phrasing that shows up whenever you ask a machine to sound like everyone else in a category. A couple were grammatically clean and completely forgettable, the kind of line that reads fine and does nothing. One tried so hard to sound clever it forgot to say what the product actually did. This is the part people miss when they either fear these tools or worship them: the output isn’t good or bad on its own. It’s raw. It reflects the average of everything it’s been trained on, and average was never the standard we build to.

One headline, though, had something real in it. It found an angle the brief had buried on page three, a line about the product solving a problem the client’s customers actually complained about, not the problem the client assumed they had. I didn’t use it as written. I cut six words, changed the verb, and rebuilt the rhythm so it sounded like a person said it out loud instead of typed it into a form. What shipped to the client an hour later was maybe forty percent the machine’s and sixty percent mine, and I couldn’t tell you the exact percentage because that’s not really how editing works. It’s not a blend you can measure. It’s a decision, made line by line, about what earns its place and what doesn’t.

That’s the part worth sitting with. A tool generated eleven headlines that weren’t good enough and one that had a spark in it a person still had to find, shape, and finish.

“You can do content generation at scale, but infinite content doesn’t imply infinite creativity.”

Jensen Huang, Founder and CEO, NVIDIA, remarks at the Cannes Lions International Festival of Creativity, June 2023

The question that mattered more than the headline

Here’s where it got interesting for me, and it wasn’t the headline. It was the conversation I had with myself afterward about whether to tell the client an AI tool had been in the room.

The easy answer would have been to say nothing. Nobody would have asked. The line I shipped read like it came from a person, because by the time it shipped it had gone through a person, several times over. I could have taken full credit and no one would have blinked.

I didn’t, because that’s not the standard we hold here. We call it the Picasso Standard internally, and the idea is simple: the issue was never the tool. Picasso used whatever was in front of him, and nobody questions whether the work was his. The issue is misrepresenting how something got made. Claiming a machine’s raw output as untouched human skill is the one move we don’t make, ever, because it’s a lie about the craft, and the craft is the entire reason a client hires us instead of generating their own headlines for free.

So I told them. Plainly, without making it a bigger deal than it was: we used an AI tool to generate a wide set of directions fast, under a compressed deadline, and then my team did what my team does, which is read everything with a sharper eye than the tool has and decide what actually deserved to survive. The client’s response, more or less, was a shrug and a "good, that’s what I’d assume you’d do." Which told me something. The disclosure wasn’t the risk I’d built it up to be in my head. The risk was in not saying it and getting caught pretending a shortcut was a finished skill.

What this one project taught me about the whole position

Adobe’s research from earlier this year found that eighty-nine percent of marketing and customer experience professionals expect generative AI will help them create more and better content (Adobe). I believe that number. I also think it undersells the harder truth sitting underneath it, which is that "more content, faster" and "better content" are not automatically the same outcome. More only becomes better when a person with real judgment sits between the draft and the client. Strip that person out and you just get more, faster, of whatever the average sounds like.

McKinsey’s most recent survey found that a third of organizations are now using generative AI regularly in at least one part of their business (McKinsey & Company). That number is going to keep climbing, and I’m not interested in pretending otherwise or wringing my hands about it. What I’m interested in is what the other two-thirds of that sentence looks like once the tool is normal and boring, once nobody’s writing thinkpieces about it anymore. What does the work look like when a machine can generate twelve headlines for anyone, in seconds, for free?

I think it looks exactly like what happened on that Wednesday. The tool gets more common. The judgment gets more valuable, not less. Twelve options is easy now. Knowing which one is actually worth a person’s name on it, and being honest about how it got there, is still the hard part. It was always going to be the hard part. That’s the part we get paid for.

“AI sits at our table. It doesn’t run the haus.”

David Keyes, Founder & CEO, HAUS XXIV