How to Humanize AI Responses, Step by Step

How to Humanize AI Responses - Step-by-Step Guide

To humanize AI responses, save a standing instruction in your assistant's custom instructions settings that tells it how to write: use contractions, vary sentence length, skip the opening pleasantries, and drop the numbered list unless the answer is genuinely a sequence. Every major assistant — ChatGPT, Claude, Copilot, Gemini, Perplexity and Grok — has somewhere to store this, and it then applies to every new conversation without you repeating yourself.

The rest of this guide covers what to put in that instruction, where each assistant keeps the setting, and how to tell whether it actually worked.

What Makes AI Sound Robotic?

Before you can fix it, you need to recognize the problem. Robotic AI responses typically have:

• Excessive politeness: "I'd be absolutely delighted to assist you!"
• Formulaic structure: Every response follows the same pattern
• Meta-commentary: "As an AI language model, I should note..."
• Perfect uniformity: Same tone, same structure, same enthusiasm for everything
• Numbered lists for everything: Defaulting to 1, 2, 3... format

Which Use Case Are You Writing For?

Different use cases need different approaches:

• Work/Professional: Professional but conversational
• Personal/Learning: Casual and friendly
• Creative: Expressive and vivid
• Technical: Clear and precise but not robotic

Understanding your use case helps you choose the right humanizing approach.

What Should Your Humanizing Prompt Say?

Use this template as a starting point:

You are a helpful assistant with a natural, conversational communication style.
Communication guidelines:
Use contractions (I'll, you're, it's) naturally
Vary your enthusiasm—not everything is "amazing" or "incredible"
Skip meta-commentary about being an AI
Use analogies and examples to explain complex ideas
Acknowledge when something's tricky or nuanced
Mix paragraphs with occasional lists (not everything numbered)
Show personality while staying helpful
Avoid:
Formulaic openings like "Certainly!" or "I'd be delighted"
Numbering everything (save lists for when they help)
Announcing you're an AI
Overly corporate grammar
Generic sign-offs after every response

How Do You Adapt It to Your Use Case?

Where Does Each AI Keep Its Custom Instructions?

Every assistant has somewhere to store a standing instruction. Every one of them puts it somewhere different and calls it something different. This is the table to bookmark.

Two things are worth knowing before you start. Claude scopes its instructions to a project rather than to your whole account, so a prompt saved in one project does nothing in another. And all of them apply to new conversations only — the chat you already have open will carry on exactly as it was.

How Do You Add the Prompt to Your Assistant?

ChatGPT

1. Go to chat.openai.com
2. Click your profile icon → Settings → Personalization → Custom Instructions
3. Paste your prompt in the second field ("How would you like ChatGPT to respond?")
4. Toggle "Enable for new chats" to ON
5. Click Save
6. Start a new conversation to test

Claude

1. Go to claude.ai
2. Click Projects → Create or select a project
3. Click Edit project → Custom Instructions
4. Paste your prompt
5. Click Save
6. Start a new conversation in the project

Perplexity

1. Go to perplexity.ai
2. Click your profile icon → Settings → Custom Instructions
3. Paste your prompt
4. Click Save
5. Start a new conversation

Gemini

1. Go to gemini.google.com
2. Click Settings → System Instructions (or "Modify how Gemini responds")
3. Paste your prompt
4. Click Save
5. Start a new conversation

Microsoft Copilot

1. Go to copilot.microsoft.com
2. Click your profile icon → Settings → Personalization
3. Paste your prompt into Custom Instructions
4. Save, then start a new chat

Copilot is the one most people forget has this setting at all, which is why it so often stays in its stiff default register. Note that this changes how Copilot writes. It has separate voice settings that control how it speaks, and those are unrelated — changing one does nothing to the other.

How Do You Know If It Worked?

Ask your AI to explain something complex:

What Should a Good Response Look Like?

• Natural opening ("So machine learning is basically...")
• Uses analogies or examples
• Acknowledges complexity
• Varies sentence structure
• Ends naturally (not with formulaic sign-off)

• Starts with "Certainly!" or "I'd be happy to..."
• Defaults to numbered list
• Includes meta-commentary
• Uses uniform structure
• Ends with "I hope this helps!"

What If the Replies Still Feel Wrong?

If Responses Are Too Formal

Add to your prompt:

• "Use more casual language"
• "Write like you're explaining to a friend"
• "Drop the corporate speak"

If Responses Are Too Casual

Add to your prompt:

• "Keep it professional but conversational"
• "Professional tone, not overly casual"
• "Like a knowledgeable colleague, not a friend"

If Structure Is Still Robotic

Add to your prompt:

• "Vary your response structure more"
• "Don't default to numbered lists"
• "Mix paragraphs with occasional bullets"

If Tone Is Inconsistent

Add to your prompt:

• "Maintain this conversational tone consistently"
• "Keep the same energy level and style throughout"

Can You Use Different Prompts for Different Tasks?

Once you've got the basics working, create specialized prompts for different contexts:

Work Email Prompt

You're helping draft professional emails. Be clear, direct, and friendly—like a 
knowledgeable colleague. Use professional language without being stiff. Get to 
the point quickly.

Learning Prompt

You're a patient teacher. Break down complex concepts step by step. Use analogies 
and examples. Acknowledge when something's difficult. Check for understanding 
naturally.

Creative Prompt

You're a creative collaborator. Use vivid language, metaphors, and analogies. 
Show enthusiasm for interesting ideas. Think outside the box. Be expressive.

What Usually Goes Wrong?

Pitfall #1: Being Too Vague

Pitfall #2: Overcorrecting

Pitfall #3: One-Size-Fits-All

Pitfall #4: Not Testing

Pitfall #5: Ignoring Context

Quick Reference: Humanizing Checklist

• Unchecked:☐Created humanizing prompt with specific guidelines
• Unchecked:☐Customized for your use case (professional/casual/creative)
• Unchecked:☐Added prompt to your AI platform's settings
• Unchecked:☐Enabled the prompt (if required)
• Unchecked:☐Started a new conversation to test
• Unchecked:☐Verified responses sound more natural
• Unchecked:☐Refined prompt based on results
• Unchecked:☐Created context-specific prompts (if needed)

Before and After Comparison

Before (Robotic)

1. Quantum computing uses quantum bits (qubits)
2. Qubits can exist in superposition
3. Quantum entanglement enables parallel processing
4. Potential applications include cryptography and drug discovery

I hope this explanation helps! Feel free to ask if you have questions."

After (Humanized)

Think of it like this: a regular computer bit is either 0 or 1. But a quantum bit (qubit) can be both 0 and 1 at the same time—that's called superposition. It's like a coin spinning in the air—it's not heads or tails, it's both until it lands.

This lets quantum computers explore many possibilities simultaneously. Where a regular computer checks options one by one, a quantum computer can check them all at once (in a sense).

The tricky part? It's really hard to build and keep stable. But if we can figure it out, it could revolutionize things like cryptography, drug discovery, and optimization problems."

Key Takeaways

1. Start with a template - Don't reinvent the wheel
2. Customize for your use case - Professional vs. casual vs. creative
3. Test and refine - Iteration is key
4. Be specific - Concrete guidelines work better than vague ones
5. Create context-specific prompts - Different scenarios need different approaches

Ready to Humanize Your AI?

The fastest way to get started is with our interactive wizard. We'll guide you through creating a custom humanizing prompt tailored to your preferences, then help you set it up on your preferred AI platform step-by-step.