Telling AI What NOT to Do
3.6 — Telling AI What NOT to Do
Estimated reading time: 10 minutes
Most prompt engineering advice focuses on what to tell the AI to do. But some of the most powerful prompting happens when you tell the AI what not to do.
This might sound strange. Why would you spend words on things you don't want? Because AI models have strong defaults — ways they tend to respond when not given specific guidance. Some of those defaults are useful. Many of them are deeply annoying.
You've seen them. The unnecessary preamble: "Great question! I'd be happy to help with that." The endless bullet points when you wanted prose. The hedging and caveating that turns a crisp answer into a wall of qualifications. The overly formal tone when you wanted something casual. The five-paragraph essay when you wanted two sentences.
These aren't bugs in the AI. They're defaults — trained behaviors that the model falls back on when it doesn't know exactly what you want. The solution is simple: tell it exactly what you don't want, and those defaults disappear.
Why Negative Constraints Work
When you give an AI a positive instruction ("be concise"), you're giving it a direction to aim for. When you give it a negative constraint ("don't use bullet points"), you're giving it a hard boundary to stay within.
Negative constraints are often clearer and more enforceable than positive ones. "Be concise" is subjective — concise to you might mean 50 words; to the AI, it might mean 200. "Under 100 words" is unambiguous. "Don't start with 'I'" is unambiguous. "No more than three sentences" is unambiguous.
Both positive instructions and negative constraints are valuable. The best prompts typically use both.
Key Takeaway: Negative constraints eliminate AI defaults you don't want. They're often more precise than positive instructions because they define hard limits rather than fuzzy targets.
The Most Useful "Don'ts"
Here's a practical catalog of the most common AI behaviors worth switching off, and exactly how to phrase the constraint.
Format Constraints
Don't use bullet points:
Write this as flowing prose. Do not use bullet points or numbered lists.
Don't use headers:
No headers or section titles. Write as continuous paragraphs.
Don't pad the length:
Do not pad this with filler content. If you can say it in fewer words, do.
Specific length limits:
Do not exceed 150 words under any circumstances.
Do not go beyond 5 bullet points.
Keep each section to one paragraph only.
Opening and Closing Constraints
One of the most universally disliked AI habits is the sycophantic opener and the unnecessary summary at the end.
No preamble:
Do not start with "Great question," "Certainly," "Of course," "Absolutely,"
or any variation of affirming the request. Start directly with the answer.
No meta-commentary:
Do not start by explaining what you're about to do. Just do it.
No summary at the end:
Do not add a summary or conclusion paragraph restating what you just said.
No sign-off:
Do not end with "I hope this helps," "Feel free to ask," or similar phrases.
Tone Constraints
No corporate language:
Do not use corporate buzzwords like "leverage," "synergy," "best-in-class,"
"robust," "scalable," or "moving the needle."
No hedging:
Do not hedge every statement with "this may vary" or "it depends."
Give me a direct answer based on the most likely scenario.
No excessive positivity:
Do not soften the critique. I want honest feedback, not encouragement
wrapped in criticism.
Don't write like a textbook:
Do not write in an academic or formal register.
Write like an intelligent person talking to a friend.
Content Constraints
Don't give generic advice:
Do not give generic advice that could apply to any business.
Everything should be specific to the situation I described.
Don't explain what I already know:
Assume I understand [TOPIC]. Do not explain foundational concepts
unless they're directly relevant to my specific question.
Don't include disclaimers:
Do not add legal disclaimers or "consult a professional" caveats at the end.
I understand this is general information.
Don't give me options when I want a recommendation:
Do not give me a list of options and tell me to decide.
Make a recommendation and tell me why.
Before and After: The Full Constraint Comparison
Here's a prompt without constraints, followed by the same request with well-targeted "don'ts."
Before (no constraints):
Explain the main risks of starting a business.
What you typically get: A 500-word response with an intro paragraph, five or six bullet points each with a sub-explanation, and a closing paragraph that says "starting a business is challenging but rewarding." All accurate. All completely generic.
After (with constraints):
Explain the 3 most underestimated risks of starting a business.
Do not include obvious risks like "you might fail" or "it's competitive." Focus on things first-time founders overlook. Do not use bullet points — write it as three short paragraphs. No intro or conclusion. Start directly with the first risk. Total length: under 250 words.
What you get: Three specific, non-obvious risks written in a format you can actually use — maybe for a presentation, a blog post, or just to think with. No fluff. No preamble. Exactly what you asked for.
Style-Specific Constraints: Building Your Personal "Don'ts List"
The most valuable application of negative constraints is building a personal style guide — a set of instructions you add to prompts whenever you want the output to sound like you.
Think about the writing you do regularly. What AI defaults drive you crazy? Build a list.
Example personal "don'ts" list:
- Do not use the word "delve"
- Do not use rhetorical questions as section openers
- Do not use em dashes for every list item — write full sentences
- Do not use "in conclusion" or "to summarize"
- Do not write more than 3 sentences per paragraph
- Do not use passive voice unless necessary
- Do not start consecutive sentences with the same word
Once you have your list, you can paste relevant items into any prompt where you're generating written content. Over time, this becomes one of the most efficient ways to get AI outputs that fit your style without heavy editing.
The Magic Phrase: "Don't Just... Instead..."
One of the most effective constraint structures is pairing a "don't" with a "instead do this":
Don't give me a list of options. Instead, make one clear recommendation
and explain your reasoning.
Don't use technical jargon. Instead, explain it the way you would
to a smart non-specialist.
Don't hedge with "it depends." Instead, answer for the most common scenario
and flag when different circumstances would change the advice.
This structure is powerful because it doesn't just eliminate the behavior you don't want — it actively redirects the AI toward what you do want.
A Complete Prompt Using Constraints
Here's a full prompt that uses negative constraints throughout:
Act as a plain-speaking financial advisor.
I'm 34, earn $85K/year, have $12K in credit card debt at 19% APR,
and $8K in a savings account. I want to know what to do with my money first.
Give me your honest recommendation — not a list of options.
Do not:
- Start with "Great question" or any affirmation
- Use financial jargon without immediately defining it
- Add disclaimers about seeing a financial advisor
- Give me more than 3 steps
- End with a motivational statement
Just give me a clear, direct answer in plain English.
Under 200 words total.
Notice how the constraints aren't there to be restrictive — they're there to strip away everything that would make the answer less useful, and focus the AI entirely on what you actually need.
Key Takeaway: Negative constraints are a form of editing your prompt before you hit send. They pre-trim the output, removing the AI defaults that would otherwise waste your time.
Template: The Constrained Prompt
[ROLE if applicable]
[TASK + CONTEXT using TCFT framework]
Do not:
- [BEHAVIOR 1 you want to avoid]
- [BEHAVIOR 2 you want to avoid]
- [BEHAVIOR 3 you want to avoid]
Instead:
- [WHAT YOU WANT INSTEAD]
Building Your Constraint Library
As you use AI tools more, you'll notice the same defaults coming up again and again. Keep a running note of the constraints that work well for you. Build a personal style constraint block you can paste into any writing prompt:
Style constraints: No sycophantic opener. No summary conclusion.
Write in direct, clear prose. No bullet points unless I specifically ask.
Under [X] words. Do not use the word "utilize."
That one block, pasted into the right prompts, will save you more editing time than almost anything else in this module.