How to Avoid AI Slop: The CO Framework

I know you keep hearing about AI slop. What is it?

AI slop is a product of generating an output using AI where there is no real effort put into making it valuable or high quality. It's the LinkedIn post that reads like every other LinkedIn post. The blog that says nothing in 800 words. The image with six fingers nobody bothered to check.

In one line: AI slop is low-effort AI output shipped without real context or a defined outcome. The fix is the CO framework, context and outcome, done properly, every time.

It's not a niche problem. Merriam-Webster named "slop" its Word of the Year for 2025, defined as digital content of low quality produced in quantity by AI. Ahrefs looked at 900,000 pages published in April 2025 and found AI-generated text in 74.2% of them. YouTube's CEO called managing AI slop a top priority for 2026. This is the water we're all swimming in now.

Why it's everywhere

I don't blame anyone whose AI output turns out low quality. The whole promise of AI is that it hands you back your time, lets you produce something in a fraction of the hours it used to take. That promise is real.

But here's the trap. To get a genuinely high-quality output, people end up spending hours crafting the perfect prompt, rewriting it, adding more context, more instructions, more examples. If your saved time gets eaten by prompting, you haven't gained anything. You've just moved the labor from writing to prompt-engineering.

I hit this myself. I spent hours building a master prompt for my website, seven parts, persona, inputs, task, the works. It's good, I still use it, but most people don't have the patience or the reps to get there. So I went looking for the smaller version. The thing that actually matters once you strip everything else away.

It comes down to two things: context and outcome. I call it the CO framework.

Context: tell it what it actually needs to know

Most slop starts here. People type one line and expect the model to fill in everything else, your voice, your facts, your standards, out of thin air. It can't. It'll guess. And a guess dressed up in confident, well-formatted sentences is exactly what slop looks like.

Context means giving the model the real material it needs. Not a vague topic. The actual facts, the actual numbers, a sample of how you actually write. If you don't hand it the real thing, it invents one. That invention is where slop comes from, not from the model being "bad."

Outcome: define the finish line before you start

Clearly define the outcome you want. Always. And defining an outcome doesn't mean describing a topic, it means naming the actual shape the answer should take.

Here's what that looks like in practice. Ask a model to "list everything wrong with my pricing model" and you'll get a list. Fifteen bullet points, several of them saying the same thing three different ways, no sense of which ones actually matter or how they relate to each other. That's an exhaustive list. It looks thorough. It isn't structured, and it isn't an outcome, it's a topic with more words.

Now ask for the same thing using a framework instead. MECE, mutually exclusive, collectively exhaustive, is the one I reach for most. Tell the model: group the root causes into buckets that don't overlap, and make sure every real cause fits into one of the buckets, nothing left over. You stop getting fifteen overlapping bullets and start getting four or five real categories. Priced on cost instead of value. No competitor benchmark. No tiering. Underpriced relative to willingness to pay. Each bucket distinct, nothing double counted, nothing missing.

That's the actual difference between defining an outcome and describing a topic. "List everything" produces slop dressed up as thoroughness, because an exhaustive list rewards volume, not structure. "Structure this MECE" produces something you can act on, because you told the model what shape a good answer has before it started writing.

And always pair the outcome with guardrails, or you're giving the model permission to invent or hallucinate whatever fills the gap. A guardrail is simple: what should this NOT do. Don't invent stats. Don't use a client name I haven't given you. Don't go over the word count. Say it out loud, every time.

Show it what "good" looks like

This is the step people skip most. If you have an example of the outcome you're after, give it to the model. Not to copy word for word, but to anchor the shape, tone, and length. A model with an example to react to gets closer to what you meant on the first try. A model with no example is guessing at your taste, and that guess is exactly how you get generic output that reads like everyone else's.

Give it a persona, every time

The last piece: give the model a persona that tells it how to think, what to produce, and what skill set to borrow from. "Write me a post" gets you a generalist's guess. "You are a copywriter who writes the way I actually talk, short sentences, no hedging" gets you something closer to your voice on the first pass.

If you don't know which persona fits your context, ask the model directly. "What kind of expert should be writing this, given what I just told you?" It's a fair question and it usually gets you a useful answer.

The actual takeaway

Slop isn't a model problem. It's an effort problem wearing AI as a disguise. The fix isn't a longer prompt or a fancier one. It's four things, every time: real context, a defined outcome, explicit guardrails, and a persona that knows what it's doing.

You don't need seven parts and two hours to avoid slop. You need those four, filled in, instead of one line and hoping.

What's the one step you skip most when you're in a hurry? That's usually the one costing you the most quality.

If you want help building this into how your team actually works with AI, book time with me here.