Stop Producing AI Slop - A Guide to Fix It

You can spot AI slop from a mile away.

The em dashes everywhere. The "Here's the thing:" followed by a colon and a subheading. The bullet points that say nothing. The perfect grammar with zero personality. The formatting that screams "I typed one sentence into ChatGPT and published whatever came out."

We've all done it. I've done it. And the moment someone called me out on it, I realized how obvious it was.

This guide is about fixing that. Not by avoiding AI - but by using it properly.

What AI Slop Actually Looks Like

Before you can fix it, you need to recognize it. Here are the telltale signs:

The Formatting Red Flags 🚩

• Em dashes everywhere - "The project - which was already behind schedule - needed a complete overhaul - and fast." Real humans don't write like this.
• Headlines with colons - "Leadership: Why It Matters More Than Ever" - generic, says nothing, reads like a Wikipedia stub
• Excessive bullet points - every paragraph turned into a list because the AI doesn't know what to emphasize
• "Let's dive in" / "Here's the thing" / "In today's fast-paced world" - filler phrases that AI defaults to when it has nothing real to say
• Perfect structure, zero substance - intro, 5 numbered points, conclusion. Every time. About everything.

The Content Red Flags 🚩

• No opinions - the content presents "both sides" of everything because the AI is trained to be neutral. Real experts have a point of view.
• No specific examples - everything is hypothetical. "Imagine a company that..." instead of "I worked with a company that..."
• Generic advice - "focus on your customers" / "create valuable content" / "leverage technology." Things that sound smart but mean nothing.
• No voice - you could swap the author name and nobody would notice. It reads like it could have been written by anyone about anything.

Why Most People Get Bad Output

It's not the AI's fault. It's the prompt.

Here's what most people do:

"Write me a LinkedIn post about leadership."

And ChatGPT does exactly what you asked. It writes a generic post about leadership. No context about who you are, what you've experienced, who you're writing for, or what you actually think about leadership.

Garbage in, garbage out.

The AI is a mirror. Give it nothing and it reflects nothing. Give it your real experience, your voice, your opinions, and your audience - and the output changes completely.

The 70-80-95 Rule

Here's the framework I use every day:

Stage 1: AI Gets You 70-80% There

When you're starting fresh or working on something new, AI does the heavy lifting on structure and first drafts. But you finish the last 20-30% manually.

• AI writes the first draft
• You add your real examples, opinions, and voice
• You cut the filler and AI-speak
• You rewrite the parts that sound generic
• You publish something that's actually yours

This is where most people should start. Accept that AI is a draft machine, not a publishing machine.

Stage 2: AI Gets You 90-95% There

Once AI has enough context about you - your writing style, your business, your audience, your opinions - the output gets dramatically better. The manual editing drops from 30% to 5-10%.

This happens when you: - Build a system prompt or context file that teaches AI who you are - Feed it examples of your best work - Give it your specific audience and their problems - Set clear rules (no em dashes, no filler phrases, no generic advice)

This takes time to build. You don't get here on day one. But once you do, AI becomes a genuine extension of how you think.

The 100%?

There's no 100% with AI. The last 5-10% is always you. Your judgment, your taste, your "this doesn't feel right" instinct. That's what separates content from slop.

How to Write a Professional Prompt

The difference between slop and quality is usually the prompt. Here's the structure I use:

1. Tell AI Who It Is

Bad: "Write a blog post" Good: "You're a senior web consultant who builds revenue systems for service businesses. You write in a direct, no-bullshit tone. Short sentences. Real examples. No corporate speak."

2. Tell It Who the Audience Is

Bad: (nothing) Good: "The reader is a business owner with 10-50 employees. They're busy, skeptical of AI hype, and want practical advice they can use today."

3. Give It Your Constraints

Bad: (nothing) Good: "Never use em dashes. No filler phrases like 'here's the thing' or 'let's dive in.' No bullet lists unless they genuinely help readability. Keep it under 800 words."

4. Give It Context

Bad: "Write about lead generation" Good: "I recently built a system for a client that responds to new leads in under 60 seconds with automated qualification and personalized pricing. Write about why speed-to-lead matters, using this as the example."

5. Tell It What NOT to Do

This is the most underrated part. AI needs guardrails:

• "Don't start with a question"
• "Don't end with 'What do you think?'"
• "Don't use the word 'leverage' or 'synergy'"
• "Don't write more than 3 bullet points in a row"
• "Don't include a conclusion section - just end on the last strong point"

Building Your Context System

The people getting the best AI output aren't writing better prompts each time. They've built a system that gives AI permanent context.

Option 1: A Persona Document

Write a document that describes: - Who you are and what you do - Your writing style (short sentences? long? casual? formal?) - Phrases you use and phrases you never use - Your audience and their problems - Examples of your best writing - Rules and constraints

Paste this at the start of every AI conversation. Or better - save it as a custom instruction / system prompt.

Option 2: Reference Files

Keep a folder of: - Your 5 best pieces of content (the AI learns your voice from examples) - A list of your clients' biggest pain points - Your hot takes and opinions on your industry - Real client stories (anonymized) that AI can reference

Feed these to AI when you need content. The more specific the input, the less generic the output.

Option 3: Built-In Tools

Most AI tools now support persistent context: - ChatGPT has Custom Instructions and memory - Claude has project knowledge and CLAUDE.md files - Cursor and similar tools read your codebase as context

Use these. The 5 minutes you spend setting them up saves hours of fixing bad output.

The Quick-Fix Checklist

Before you publish anything AI helped create, run through this:

1. 1Remove all em dashes - replace with commas, periods, or just restructure
2. 2Kill the filler phrases - search for "here's the thing," "in today's," "let's dive in," "it's worth noting" and delete them
3. 3Check for opinions - does this content have a point of view? If it's neutral on everything, it's slop
4. 4Add one real example - replace at least one hypothetical with something from your actual experience
5. 5Read it out loud - if it doesn't sound like something you'd say to a client, rewrite it
6. 6Check the formatting - if every section is the same length with the same structure, break the pattern
7. 7Remove the conclusion - if the last paragraph starts with "In conclusion" or "To sum up," just delete it. End on the last strong point.

AI Is Your Growth Partner, Not Your Replacement

The fear around AI content is understandable. Nobody wants to sound like a robot. Nobody wants their brand associated with generic garbage.

But avoiding AI entirely is equally stupid. The people producing the best content in 2026 are using AI heavily - they're just using it well.

The difference between slop and quality has never been about the tool. It's about the operator.

Give AI nothing and it gives you nothing back. Give it your expertise, your voice, and your real experience - and it gives you a head start that no amount of manual writing can match.

The businesses that master this now will look back in a year and wonder how they ever did it the old way.