By Clara Bennett
How to Tell AI What You Do Not Want
A beginner guide to adding limits, exclusions, and preferences so AI answers feel more useful and less off-target.
Quick answer
To tell AI what you do not want, add a clear limit to your prompt. Say what to avoid, what to keep out, what tone not to use, or what format not to choose. This helps the AI narrow the answer before it starts and saves you from fixing the same problem later.
Key takeaways
- You can tell AI what to avoid before it answers.
- Negative instructions work best when they are specific.
- Useful limits include tone, length, format, topics, and assumptions.
- A simple avoid-this line can make beginner prompts much easier.
What does it mean to tell AI what you do not want?
Telling AI what you do not want means adding a boundary to the prompt. A boundary is a simple instruction that rules out an answer style, topic, format, tone, or assumption.
This is helpful because beginners often know what feels wrong before they know the perfect way to ask for what feels right.
- Do not make it too formal.
- Do not use technical words.
- Do not make it longer than five bullets.
- Do not give medical advice.
- Do not assume I already understand the topic.
Start with the part you want to avoid
If AI keeps giving answers that miss the mark, name the part you dislike. You do not need perfect wording.
The clearest pattern is: do the task, but avoid this specific thing.
- Explain this in plain language, but do not use tech jargon.
- Rewrite this email politely, but do not make it sound fake or overly cheerful.
- Make a checklist, but do not include more than seven items.
- Help me plan my week, but do not schedule every minute.
Use this avoid-this prompt
This prompt is useful when you are asking for an email, explanation, plan, checklist, or message.
Replace the bracketed parts with your real situation.
- Help me with [task]. My goal is [goal]. Please avoid [what I do not want]. Keep the answer [format or length]. Use a [tone] tone.
- Rewrite this message so it is polite. Avoid sounding stiff, fake, or too apologetic. Keep it under 120 words.
- Explain this topic for a beginner. Avoid technical terms unless you define them first.
Set tone limits when writing messages
Tone is one of the best places to use a do-not-want instruction. Many AI messages sound too formal, too excited, too long, or too polished.
You can steer that before the draft appears.
- Make this professional but not corporate.
- Make this friendly but not bubbly.
- Make this direct but not rude.
- Make this calm but not cold.
- Make this confident but not aggressive.
Set length and format limits
If AI gives you too much, add a length rule. If AI gives you a paragraph when you wanted steps, add a format rule.
These limits are especially useful for beginners because they make the answer easier to scan.
- Do not write more than one short paragraph.
- Do not include an introduction.
- Do not give me ten options. Give me three.
- Do not use a table. Use bullets.
- Do not explain the theory first. Start with the steps.
Set topic limits for safer answers
Sometimes the most important limit is about what the AI should not pretend to know. This matters for health, legal, financial, safety, and personal information topics.
For important decisions, ask for general guidance and verification steps instead of a final answer.
- Do not diagnose anything. Give me questions to ask a professional.
- Do not make a legal conclusion. Explain what I should verify.
- Do not guess current prices. Tell me what information I need to look up.
- Do not ask for private personal details.
Example: weak prompt to stronger prompt
Weak prompt: Write a reply to this email.
Stronger prompt: Write a reply to this email. Keep it polite and brief. Do not sound defensive, do not over-apologize, and do not make it longer than six sentences.
- The task is clear.
- The desired tone is clear.
- The unwanted tone is named.
- The length is limited.
Common mistake: using a vague no
A vague instruction like do not make it bad does not help much. AI needs to know what bad means in this situation.
Translate the vague feeling into a specific limit.
- Instead of not bad, say not too formal.
- Instead of not weird, say not overly cheerful.
- Instead of not long, say under 100 words.
- Instead of not confusing, say use simple words and one example.
Your five-minute action step
Take one prompt you already use and add one avoid-this line.
Then compare the answer with and without the line. You will usually see the difference quickly.
- Choose one real task.
- Name one thing you do not want.
- Add a length, tone, or format limit.
- Ask for the answer again.
- Save the version that worked best.
Related reading
Beginner FAQ
Can I tell ChatGPT what not to do?
Yes. You can add instructions like do not use jargon, do not make it too formal, do not write more than five bullets, or do not assume I know the topic.
What should I do if AI ignores what I said not to do?
Repeat the limit more specifically in a follow-up. For example, say revise this and remove the technical terms, or make this shorter and keep it under 100 words.
Is it better to say what I want or what I do not want?
Use both when you can. Say what you want first, then add one clear limit for what to avoid.
What is an example of a good negative instruction?
A good negative instruction is specific: do not sound defensive, do not use technical language, or do not include more than three options.
Next step
Want a guided path instead of random tips?
AI Basics Bootcamp turns these beginner ideas into a short, practical course with examples, practice prompts, and progress you can follow at your own pace.