The problem is that most people use AI to generate "average" content. If an article follows the classic AI trajectory—broad introduction, three generic points with bulleted lists, and a summarizing conclusion—it feels hollow. This is where the "AI smell" comes from. When we talk about prompt engineering, we often focus on the output's accuracy, but we forget about the "humanity" of the prose. Real expertise usually comes with contradictions, strong opinions, and specific, messy anecdotes that LLMs are trained to smooth over.
To make AI content actually readable, you have to move past the basic prompt. I've found that treating the AI as a drafting tool rather than a finished-product generator is the only way to maintain quality. For those trying to build a real AI workflow for content, here is a practical approach to stripping away the robotic feel:
-
Feed it raw data, not topics. Instead of asking for "an article about LLM agents," feed it a transcript of your own voice notes or a rough list of your specific findings.
-
Ban the "AI vocabulary." I explicitly tell my prompts to avoid words like "delve," "comprehensive," "unlocking," and "testament to."
-
Force a perspective. Tell the AI to take a contrarian stance or to argue against a common industry belief. This prevents the "balanced but boring" tone that screams AI.
-
Inject "ugly" details. Add specific version numbers, error codes, or weird edge cases that only someone who actually ran the code would know.
If you're using these tools for a deep dive or a hands-on guide, the technical accuracy is the primary currency. If a step-by-step tutorial helps a developer fix a bug in five minutes, they rarely care if an AI helped structure the explanation. However, if the article is an opinion piece, the lack of a "soul" becomes glaring. Ultimately, the "AI-written" stigma only applies to low-effort content. When an AI is used to synthesize complex information into a clear, concise format, it's a productivity win. When it's used to fake expertise, it's just noise. The goal shouldn't be to hide the AI, but to ensure the AI is augmenting a real human insight rather than replacing it.
Claude watermarks won't be visible to the eye but they will 6h ago
Stop letting LLMs write your content in that predictable 7h ago
[Is jailing anti-AI protesters actually a sign that we're losing 8h ago](/en/news/6570/)
[Stop expecting LLMs to be databases because they are 15h ago](/en/news/6543/)
[Most teenagers aren't actually obsessed with AI 1d ago](/en/news/6392/)
Anthropic investors are betting on a $2 trillion valuation for 3d ago
[Next Stripe might be spending $7 billion to grab OpenRouter →](/en/news/6600/)
[an AI side-hustle playbook](https://tanyan888.com/), with plenty of directly applicable cases.