You can spot it in two sentences. The post that opens by announcing that the world moves faster than ever. The newsletter that promises to “transform your business” without saying how. The social caption with three adjectives where one would do. That’s slop, and your customers can smell it as fast as you can.
We run a lot of content on AI. A newsletter engine. Social calendars. Three blogs. A comic strip that goes out every week. None of it is supposed to read like a machine wrote it, and the reason it doesn’t has very little to do with which AI tool we use.
It has to do with rules, and a person who enforces them.
Why AI writing sounds like AI writing
AI models learn from an enormous amount of text, and a lot of that text is marketing copy. So when you ask for “a blog post about kitchen remodeling,” you get the average of every kitchen remodeling post ever written. Safe. Smooth. Forgettable.
The tells are structural more than they are about vocabulary. Swap “utilize” for “use” and the post still reads wrong. Here’s what actually gives it away:
- Everything comes in threes. “Fast, reliable, and affordable.” Once you notice it you’ll see it everywhere.
- Every sentence is the same length. Medium, medium, medium. People don’t write like that. People burst.
- Sentences end with a restatement. “The team finished the project early, highlighting their commitment to quality.” That last part adds nothing.
- Both sides get equal weight. “While X is important, Y also matters.” It sounds balanced. It says nothing.
- The punctuation gives it away. A client once flagged a certain long dash as an AI tell. Once someone sees it, they see it everywhere. So it’s banned in our client-facing writing, full stop.
None of these are grammar mistakes. That’s why they survive editing. The writing is technically fine and still dead on the page.
The kill-list
Our fix is a written list of structural rules that runs on every single piece of content before it ships. We call it the kill-list. It’s the most important document in our content operation, more important than any prompt.
A few of the rules, in plain terms:
- Default to two items, or one. Break the rhythm if a list really needs three.
- Vary sentence length on purpose. Short ones next to long ones.
- No trailing phrases that re-say the sentence.
- One qualifier per paragraph. Say the thing.
- Casual transitions or none. “And.” “But.” “So.” Never “furthermore.”
Then there’s a banned-word list, and it’s long. We won’t reprint it here, because we’d rather not have those words on this page either. If a word shows up in every agency’s pitch deck, it doesn’t show up in ours.
The list gets applied twice. Once when AI drafts, because the rules are part of the instructions. And again at review, because AI follows instructions most of the time, not all of the time.
Voice comes from somewhere
A kill-list stops the bad stuff. It doesn’t create a voice. For that you need source material.
For every client whose content we write, we build a voice file. What their customers call things. How formal they are. Whether they’re funny, and what kind of funny. Real sentences they’ve written, in emails or on the phone, that sound like them. A welding shop and a counseling practice do not sound alike, and the content shouldn’t pretend they do.
That voice file is why a newsletter can go out in a business owner’s register instead of in a generic brand voice. AI is very good at matching a pattern once you show it one. It’s terrible at inventing a personality from nothing.
A person reads every piece
This is the part that doesn’t scale, on purpose. AI drafts. A human principal reads every output before it goes anywhere, and signs off or sends it back.
That review is where most of the quality lives. It catches the claim that isn’t true. The joke that doesn’t land for this audience. The sentence that technically follows the rules and still sounds off. It catches the fact AI made up with total confidence, which does happen, and which is the single biggest risk of publishing AI content without a person in the loop.
We lay out where that human step sits in every build on our How It Works page. It’s not a formality. It’s the model.
What this looks like in practice
Take the weekly comic strip. Thirteen strips so far. The pipeline goes script, then art, then publish, and AI does real work at every stage. But the script gets read before the art starts. The art gets checked before it posts. If a strip isn’t funny, it doesn’t go out. That’s a judgment call, and a machine doesn’t get to make it. We wrote the full breakdown in A Comic Strip Every Week.
The newsletter engine works the same way. The engine drafts from the month’s real news and work, in the right voice, against the kill-list. Then a person reads it, cuts what’s weak, and approves the send. We covered that one in The Newsletter Engine.
Should you use AI for your content?
Probably, yes. Just not the way most people start.
Most businesses open a chatbot, type “write me a blog post,” and publish what comes back. That’s where slop comes from. If you’re going to do it yourselves, here’s the short version of what works:
- Write your rules down first. Words you never use. Things you always say. How you talk to customers.
- Give it real material. Your own emails, your own notes, the actual question a customer asked last week.
- Treat the draft as a draft. Read it out loud. If you wouldn’t say it to a customer’s face, cut it.
- Check every fact. Dates, prices, names, claims. Every one.
Content that sounds like you and says something true will outperform a hundred generic posts. It always has. AI just makes it possible to do that every week instead of every few months.
The short version
Slop isn’t an AI problem. It’s a standards problem. AI without rules produces the average of the internet. AI with a kill-list, a voice file, and a person who reads everything produces content you’d actually put your name on.
If you’d like content that sounds like your business and not like a template, let’s talk about it.