🙅 The master of no

Somewhere in the second week of rebuilding my positioning, I caught myself typing the same thing for the umpteenth time that day. “No.” “Back up.” “Let’s try that again.”

Essentially, I was serving as an AI tamer. (insert whip crack sound here)

Some of those no’s were small. Claude had developed an unexpected crush on the word “shape” and kept sliding it into every other paragraph, much to my annoyance.

The no’s that mattered were a different animal. The AI had taken my messaging somewhere that sounded confident and clean and almost right. There was just one problem: it wasn’t mine. Adjacent to my brand. Not it. That’s the drift that costs you, because by the time you notice, you’ve half-agreed with it.

Looking back across recent work, the pattern held in other projects, too. Most of my job came down to steering the thing back on course.

Working with AI, I’ve realized, means keeping a veto card in your pocket and being ready to play it, again and again. I’ll admit that wears on you. But that pushback is the work.

Why “good enough” keeps showing up

This kind of drift, away from what you actually meant and toward the generic middle, is easy to miss. The AI fills the gaps you leave with something competent. Not wrong, not obviously bad, just the most reasonable, most common version of what you might have meant.

And it won’t stop to ask, unless the gap is glaring or you’ve told it to. There’s a simple reason: asking is friction, and every consumer product is built to remove friction. A model that interrogates you before it answers feels worse to most people than one that just hands you something usable.

But the default is what fills your work with assumptions you never made.

The smarter it gets, the quieter the drift

You’d think better models would fix this. I think they make it worse, and I’ve got receipts.

When Fable 5 first came out, before all the controversy, I tested it with both lazy prompts and fully specified ones. On the detailed briefs, the gap between it and Opus was marginal. Where Fable 5 pulled ahead was the vague prompt. It built a scaffold of reasoning on top of my half-formed request, worked out what I probably meant, and delivered a polished answer without asking a thing.

Its headline skill is inferring your intent from an underspecified ask and proceeding with confidence, which is the exact situation where drift is born.

Better reasoning takes a weak prompt and dresses it up. The output looks more finished and more sure of itself while wandering further from what you needed. A visibly mediocre result at least warns you. A confident, competent, wrong one doesn’t.

You can only say no when you know what “yes” looks like

So the job becomes saying no, well and often. And you can only reject something when you know what you were reaching for.

I learned this the boring way. Early on, I assumed Claude could pick up my quality standards and my brand voice from the writing I’d fed it, and work out on its own which messages I wanted to make. It couldn’t. What I got back was plausible for my field and generic for me. A size too big, a shade too meh. The fix was handing over the details I’d been keeping in my head: the voice guideline, the messaging map, the quality criteria I actually use. Once those were on the table, every no got easier. I finally had something to measure against.

That doesn’t mean every task needs a rigid spec. If you’re brainstorming your way toward something new, a clear objective and a sense of the destination can be enough. But you have to be clear on that much.

What sinks you is having nothing to steer toward. Then you can’t tell drift from a good idea, and the AI is very persuasive about its suggestions.

No’s can be mindful or automated

Doing this deliberately every single time would be exhausting, and you don’t have to. The “no” scales with the stakes.

Nailing down your core content angles, the handful of ideas you want to be known for, takes every ounce of willpower you have, every time, because the cost of drift there is your whole story going slightly false. A daily Threads post doesn’t deserve the rigor. There you can encode the “no”: an automatic review step can carry your voice and quality bar well enough, holding the line on the small stuff so you spend your judgment where it matters.

***

Last week I wrote about the content marketer becoming an architect, owning the whole chain instead of passing it between hands. This is the other half of that job. When one person holds the thread the whole way through, what keeps the work theirs is the willingness to keep saying no. To the confident average, to the word AI suddenly fell in love with, to the messaging that’s almost right.

The next time the AI hands you something that looks finished, ask what you set out to make, and whether this is it. If you can’t answer quickly, the gap is yours to close. And it’s worth closing before you type another word.

“No” matters,

👌 Handpicked stories for you

Here are some interesting things I stumbled upon this week while reading, writing, researching, and being a generally curious marketing person on the internet:

💼 Three for work

5 Takeaways from Google’s GEO Guidelines | If you want one level-headed read on Google’s new GEO guidelines, make it this one. Moz’s Dr. Pete Meyers cuts through the acronym soup (GEO, AEO, or, in Google’s telling, still-just-SEO) to the parts that actually change your work: query fan-out means a single page now has to answer half a dozen related searches at once, and structured data isn’t the lever some people are selling.

The best response to AI slop, infinite advice, and online noise is from Robin Williams | Jay Acunzo makes a great point: your lived experience, the specific way you see things, is the one input a model can’t pull from its training data. When the machine drifts toward the average, the way back is the perspective only you have, delivered with enough craft to actually land.

What Three Months of Fractional Storytelling Looks Like | A useful behind-the-scenes look at embedding a product narrative into a company. Getting the story approved is the easy part, and the real test is whether teams actually use it afterward. A narrative that stays in a doc has failed; one that people start saying it out loud on their own has worked. If you’ve ever watched a beautiful positioning deck die in a shared drive, you’ll recognize the problem.

👀 Two for fun

This visual overview of depth | “How low can you go?” is, apparently, a question comic artist XKCD kept asking for a while.

This incredible place on Earth | The above link made me dive into the rabbit hole that is Mexico’s Cave of Crystals. I feel amazing knowing I live in a world where that is possible.

🐈 Weekly cat

The highest high.