How to Create Content With Claude and ChatGPT That Doesn’t Sound Like AI

The banned phrases, system prompts, and workflow I use every week.

I’ve been using AI to write for two years now.

Most AI content is garbage. You know it when you see it. The same phrases. The same structure. That weird confidence about nothing.

But here’s what I’ve learned: the problem isn’t AI. The problem is how people use it.

A good prompt won’t fix bad defaults. You need rules. Permanent ones. Rules that run on every output without you remembering to paste them.

I use a banned phrase list. 47 patterns that make content sound robotic. I block them all. The AI has to find other ways to say things. Those alternatives sound human because they don’t match what readers have learned to spot.

I also use a source verification rule. One line in my config that stops AI from inventing statistics. No more “studies show 73%...” with zero citation.

The result: content that reads like I wrote it. Because I did. AI handled the scaffolding. I handled everything that matters.

Why This Works

AI sounds like AI because it learned from the internet. The internet is now full of AI writing. Each model trains on the last model’s output.

The voice gets flatter with every generation.

You’ve seen the patterns:

• “In today’s fast-paced world...” (means nothing)

• “Studies show that 73%...” (no source, made up)

• “It’s not just about X. It’s about Y.” (every other paragraph)

• “Game-changer.” “Supercharge.” “Delve.” (instant AI detector)

These show up because they were common in training data. Now they’re everywhere. Readers spot them immediately.

One prompt can’t override patterns learned from billions of words. You need rules that apply automatically.

What’s Inside

I’m sharing the full system. The exact config I use.

1. The banned phrase list.

   47 patterns. Filler words, corporate speak, AI clichés, false authority phrases. Block them all and the output changes completely.

2. System prompts for Claude and ChatGPT.

   Copy-paste ready. Put them in your project instructions or custom settings and forget about them.

3. The source verification rule.

   One addition that stops fabricated statistics. Every claim needs a URL or it gets flagged.

4. The 5-step workflow.

   How I actually create content. Research, outline, draft, fact-check, human pass. What AI handles. What I handle.

5. What stays human.

   The parts you should never delegate. Your data, your opinions, your experience. Skip this and your content becomes replaceable.

6. 20-minute setup.

   The minimum viable config. Four things that handle 80% of what matters.

This took me two years to figure out. Hundreds of articles. Constant refinement.

You can copy it in an afternoon:

---

1. The Banned Phrase List