Do you know what to watch out for when using AI to write content?

AI tools like ChatGPT, Claude, and Copilot are great for drafting emails, reports, LinkedIn posts, and documentation quickly. They're not always great at sounding human.

Left unchecked, AI-generated text comes across as overly formal, overly dramatic, or just a bit fake.

Watch out for common AI tells

Before hitting send, check the draft for these patterns:

• Overly formal tone - "plays a vital role", "holds significant importance", "a profound heritage". Use the words you'd say in conversation
• Adjectives instead of facts - "continues to captivate", "boasts a...", "seamlessly connecting...". If something is great, show why it's great
• Emojis - 🚀 💡 🔥 at the start of every line looks unpolished in emails, reports, and documents for clients or stakeholders
• Em dashes (—) - AI overuses them. People usually reach for a comma, a hyphen, or parentheses
• Dramatic contrasts - "It's not just X, it's Y" or "The question isn't A, it's B" when there is no real contrast
• Fake or vague analysis - Sentences that sound deep but say nothing. Ask yourself: Is this actually true? Do I have evidence? If not, cut it

❌ Figure: Bad example - Bad example - Emoji, em dash, dramatic contrast, and fluff with no facts

✅ Figure: Good example - Good example - Plain words and concrete facts

Tell the AI how to write before it starts

Fixing the tone after the fact is slower than getting it right the first time. Give the AI your output format up front instead of accepting its default voice.

• Claude - Use Styles to create a custom writing style from a sample of your own writing, then reuse it in every chat
• Claude Code - Use output styles or project instructions (CLAUDE.md) so agents write PR descriptions, docs, and commit messages the way your team does
• ChatGPT and Copilot - Put the same guidance in custom instructions so you don't repeat it in every prompt

✅ Figure: Good example - Good example - A reusable style instruction that prevents the most common AI tells

Run a humanizer pass

The humanizer skill is an open-source Agent Skill for Claude Code, Codex, and other agents. It is based on Wikipedia's Signs of AI writing guide and rewrites the patterns above (and many more) without changing what the text says.

Once installed, run /humanizer on a draft before you send it. This is especially useful when an agent writes on your behalf, such as a PR description or release notes.

Final tips

• Keep the good stuff - AI often produces solid drafts and saves time
• Don't blindly copy and paste - Review tone, phrasing, and accuracy first
• Ask yourself: "Would I say it this way?" - If not, rewrite it