Why AI content gets flagged
AI writing has patterns. Humans pick up on them quickly, clients pick up on them, and AI detection tools pick up on them. Here are the biggest tells, and they're all things our own guidelines have already identified.
Overused words
The entire banned words list exists because AI defaults to these. Words like "delve," "leverage," "robust," and "tapestry" are almost always AI.
Lists of exactly three
AI loves the rule of three. If every list in a piece has exactly three items, that's a pattern. Vary your list lengths: use two, four, or five items.
Em dashes everywhere
Claude uses em dashes constantly. Redditors and savvy readers treat them as an instant AI tell. Never use them in any output.
"Not only X, but Y" structures
This sentence structure is extremely common in AI copy. It reads as a formula because it is. Rewrite every instance you find.
Beyond those tells, AI copy tends to be generic. It says things like "this innovative solution helps businesses thrive" when what you want is "this saves your manager three hours on Monday mornings." Specificity is the biggest signal of human writing.
The OO five-step AI writing workflow
Follow this every time you use AI for client copy. It takes longer than just hitting send on the first draft, but it's the difference between work that comes back with revisions and work that goes live.
-
Brief Claude well.
Use the CRISP framework from Module 3. Give Claude the context, the role, specific instructions, examples, and parameters. Garbage in, garbage out.
-
Generate a first draft.
Let AI do about 10 to 20% of the work. Think of the first draft as a rough outline with some decent sentences in it, not a final product.
-
Run it through the banned words check.
Open the AI Words to Avoid doc and search the output for every word on the list. Replace all of them. This step alone improves the copy noticeably.
-
Rewrite at least 30% by hand.
Add your own voice, your own examples, your own opinions. Tighten sentences. Make the opening line surprising. Add a specific detail Claude couldn't know. This is what makes it sound human.
-
Read it out loud.
If it doesn't sound like something you'd actually say to a coworker or a client, rewrite it. Your ear knows. Trust it.
When clients use AI to review your work
This is a newer situation and it's becoming more common. A client runs your copy through ChatGPT and sends back feedback that reads like "This content could benefit from a more professional tone and clearer value propositions." That's ChatGPT talking, not the client.
When you notice this happening:
- Reverse-engineer the prompt the client likely used
- Generalize that prompt and incorporate the Communication Codex standards
- Run your content through Claude using that same lens before sending it to the client in the future
- Consider flagging it to the client with a light touch: "We've noticed you've been using AI to review content. We'd love to align on what you're optimizing for so we can get closer on the first pass."
Check out the Notion guide below for a full breakdown on handling this situation. And don't get defensive. Adapt.
Reddit-specific AI writing rules
Reddit has the strictest requirements of any platform because its community is highly sensitive to anything that smells inauthentic. Marketing copy that passes on LinkedIn will get destroyed on Reddit. Follow these rules exactly when writing Reddit content.
Reddit AI Writing Rules
- Act as a Redditor first. A marketer second (or not at all).
- No salutations, no sign-offs, no emojis, no hashtags.
- No em dashes. Ever. Reddit users treat this as an instant AI tell.
- Avoid AI-detectable adjectives: "commendable," "innovative," "meticulous," "thought-provoking."
- On Reddit specifically, lists of three are actually a cultural norm. (This is an exception to our general rule about varying list lengths.)
- Answer the question first. Mention the brand second, if at all.
- Write shorter sentences. Be direct. Be opinionated. Have a point of view.
- Never start a prompt with "Write a Reddit post about X." Give Claude the subreddit, the specific question, the persona, and examples of good responses from that community first.
Exercise: spot the AI, fix the copy
Below is a paragraph of AI-generated marketing copy. Using the OO guidelines, rewrite it in a way that would actually pass. A model rewrite is shown below it.
In today's dynamic landscape, leveraging cutting-edge AI solutions is crucial for businesses looking to elevate their digital presence. Our robust, seamless approach ensures that your brand resonates with diverse audiences while fostering meaningful connections. Not only do we deliver groundbreaking results, but we tailor every strategy to align with your unique ethos and goals.
Running a digital marketing program without AI in 2026 is like managing a client account without a spreadsheet. You can do it. It's just slower and more error-prone. We use AI to handle the repetitive work, speed up drafts, and test more ideas in less time. The strategy, the judgment, and the client relationships? That stays human. That's kind of the whole point.
Common writing mistakes to avoid
- Skipping the banned words check. It takes two minutes and it matters. Make it part of your process.
- Sending the first draft. AI first drafts are not final drafts. Rewrite at least 30%.
- Not reading it out loud. Your ear catches things your eyes miss. If it sounds weird when spoken, it reads weird too.
- Writing Reddit copy like you're writing a LinkedIn caption. These are completely different platforms with completely different rules. Use the Reddit doc, every time.