My Rules for Working With AI (So It Doesn't Write Like AI)

I use AI in my work constantly, and I am not precious about it. The tools are good, they save real time, and pretending otherwise would be a strange thing for a marketer in 2026 to do. But most AI-assisted writing is instantly recognisable, and it is recognisable because it is bland. So over the last year I have built myself a short set of rules. They are the difference, for me, between a tool that helps and a tool that quietly makes everything sound the same.

Here they are, unpolished.

The rules

• Kill the tells. The em-dash, “dives into,” “leverage,” “seamless,” “comprehensive,” the three-adjective stack. A model reaches for these constantly because its training is full of them. I strip every one. If a phrase shows up in a thousand AI drafts, it cannot be in mine.
• It drafts, I decide. I let it produce a first version to argue with, never a final one to publish. The blank page is the only thing I outsource. Every real decision, what to keep, what is true, what is actually interesting, stays mine.
• Specific beats fluent, always. The model gives you fluent by default. Fluent is worthless now; everyone has it for free. The only thing worth keeping is the specific: the real query, the real number, the real thing that happened. If a paragraph could describe any business, I delete it.
• Never let it assert something I cannot verify. This is the important one. A model will state a fact, a credential, a statistic with complete confidence and no idea whether it is true. Anything it hands me gets checked before it exists anywhere public. If it cannot be verified, it is cut, not softened.
• Opinions about ideas, never about people. The model is happy to write “most marketers get this wrong.” I will not. That starts fights and earns nothing. “Here is what changed my mind” is the register; certainty about other people’s incompetence is not.
• If it sounds wrong said aloud, it is wrong. My last check is to read it as if I were saying it to a friend over coffee. AI writing fails this test constantly. Human writing has a pulse, and the ear catches what the eye misses.

Why I bother

Because the point of these tools is leverage, not replacement, and the moment your writing sounds like the tool, you have handed over the one thing people were actually coming to you for. A human being fully in charge is not a nostalgic preference. It is the product.

So yes, I work with AI. And these are the rules that keep it working for me instead of the other way around.