Why should I tell AI what to do instead of what not to do?
1 Answer
A common prompting mistake is filling prompts with "don'ts": don't be wordy, don't use jargon, don't make it boring. These instructions can work, but describing what you want instead is usually more effective.
Why do positive instructions work better?
1. They give a clear target. "Don't be wordy" says what to avoid but not what to aim for. "Use short sentences, under 15 words each" gives a concrete target.
2. They avoid drawing attention to the unwanted thing. Mentioning something, even to forbid it, can sometimes make it more prominent in the AI's response. For example, telling an image generator "no clouds" may actually bring clouds into the image.
3. They are easier to check, both for the AI and for you.
Here are some examples of converting negatives into positives:
"Don't use jargon" becomes "Use everyday words a Class 10 student would understand."
"Don't be too formal" becomes "Write in a friendly, conversational tone, like a message to a colleague."
"Don't write long paragraphs" becomes "Keep paragraphs to 2 or 3 sentences."
"Don't make things up" becomes "Use only the information in the document below, and write 'not mentioned' when something is missing."
"Don't use markdown" becomes "Write in plain flowing paragraphs."
"Don't be boring" becomes "Open with a surprising fact and use one real-world example in each section."
Explaining the reason also helps. "Keep it under 50 words because it will be shown on a mobile notification" lets the AI understand the goal and make better choices.
When are negative instructions still useful?
Banned phrases: "Avoid the phrases 'in today's world' and 'game-changer'" is clear and effective.
Hard limits: "Do not include prices" or "Do not mention competitor names."
Safety and policy rules, especially in system prompts for business apps.
Correcting a specific repeated mistake.
The best prompts usually combine a clear positive description of the goal with a few precise negative rules for specific things to avoid.
This principle applies across tools, including chatbots, image generators and AI apps you build. When you catch yourself writing "don't," ask: "What do I want instead?" and write that.