DobieCore Learn

Lesson 1.5

Tell AI What NOT To Do

Purpose

By the end of this lesson, you will understand why telling AI what to avoid is often more powerful than telling it what to do. You will stop getting results that are almost right and start getting results that are actually right.

Use This When

  • Your AI output is close but keeps including one annoying thing you did not want
  • You want to stop AI from using certain words, phrases, or styles
  • You are generating images, writing copy, or creating content and AI keeps drifting in a direction you hate
  • You want the fastest way to improve AI output without rewriting your entire request

Before You Start

Nothing. This lesson is about a mindset shift: stop asking for good things and start blocking bad things.

Why Negative Instructions Work

Positive instructions tell AI what to include. Negative instructions tell AI what to exclude. Most people only use positive instructions. That is like painting a room and only telling the painter which walls to paint, not which ones to avoid.

AI is excellent at following rules. When you give it negative constraints, you remove entire categories of wrong answers before AI even generates them.

How to Write a Negative Instruction

Be specific. Be direct. Do not hint.

Weak: “Please try not to be too formal.” Strong: “Do not use formal language. Write like you are talking to a friend at a coffee shop.”

Weak: “I do not really want a long response.” Strong: “Do not write more than 200 words.”

Weak: “Maybe avoid too many buzzwords.” Strong: “Do not use these words: synergy, leverage, optimization, scalable, best-in-class.”

The Pattern

Instead of: “Make it friendly.” Try: “Do not use corporate language. Do not start sentences with ‘In today’s world.’ Do not use exclamation points.”

Conversation Prompt

Copy and paste this into your AI chat:

“I need you to help me [your task]. Before you start, here is a list of things I do NOT want you to do.

Do not:

  • [Specific thing to avoid 1]
  • [Specific thing to avoid 2]
  • [Specific thing to avoid 3]
  • [Specific thing to avoid 4]
  • [Specific thing to avoid 5]

Now, [your actual request].”

Dobie Tip

Negative prompting is especially powerful with image generation. When I create images of Sabre, my blue Doberman, I cannot just say “blue Doberman Pinscher.” I must be specific: “blue (dark gray) cropped and docked Doberman Pinscher.” And if the AI gets creative and adds a tail or floppy ears, I add: “no tail, no floppy or double ears.” The AI was not being bad. It just needed clearer guardrails. The same thing happens with your business copy.

Follow-Up Questions

  • “What did I forget to tell you not to do? What would you have guessed wrong if I had not given you these negative instructions?”
  • “Which of my negative instructions was most helpful, and which one was unnecessary?”
  • “Give me three more things you almost did that I should add to my do-not list.”
  • “If I remove one of these negative instructions, how does your answer change?”

Expected Result

By the end of this lesson, you should have:

  • ✔ A list of five negative instructions for a real business task

  • ✔ A before-and-after showing how negative instructions changed AI’s output

  • ✔ The habit of adding a “do not” section to every AI request

  • ✔ A personal list of words, phrases, or styles you never want AI to use for your business

Common Mistakes

Avoid these negative instruction traps:

Being vague about what to avoid. (“Do not be weird” is not a useful instruction. “Do not use humor” is.)

Forgetting to pair negative instructions with positive goals. (Telling AI what not to do without telling it what to do leaves it guessing.)

Only using negative instructions when you are frustrated. (They work best when you plan them, not when you are reacting to a bad answer.)

Giving up when AI ignores your negative instruction once. (Usually means you need to be more specific, not that negative prompting does not work.)

Instead, remember:

  • ✔ AI takes instructions literally. If you say “do not use jargon,” it will avoid jargon. If you say,

“try to avoid jargon,” it will ignore you.

  • ✔ The most specific negative instruction always wins over the most general positive request.

  • ✔ A good “do not” list saves more time than a hundred rewrites.

  • ✔ Negative prompting is not complaining. It is engineering.

Why This Matters

Most AI frustration comes from almost-right answers. The blog post is good but too formal. The image is perfect except for one weird detail. The email is clear but sounds like a robot. Negative instructions are how you fix the “almost.” They are the difference between settling and getting exactly what you want.

Save This

Create a running list called “My AI Do-Not List”. Every time AI does something that annoys you, add it. Examples:

  • Do not use exclamation points.
  • Do not start sentences with “In today’s competitive landscape.”
  • Do not make up statistics.
  • Do not use hashtags.
  • Do not include pricing unless I provide it.
  • Do not guess, unless I ask you to.

Update this list as you work with AI. It will become one of your most valuable business assets.

Coming Up Next

Lesson 1.6: Finding Your Brand Voice. Now that you know how to give AI context and guardrails, let us teach it how you actually sound.