Custom GPT Instructions 2026: Build a GPT That Writes in Your Voice

You open ChatGPT to draft an email, paste your request, and get back something that technically answers the question but sounds nothing like you.

So you try custom instructions. You paste in a description of your tone. Ask it to "be concise and professional." Maybe add a few writing examples.

The output improves, but only marginally. It's still missing something—the rhythm of how you actually write, the phrases you'd use, the way you adjust for different audiences.

The problem isn't that custom GPT instructions don't work. It's that most people don't know how to use them effectively.

ChatGPT's custom instructions feature is powerful. But it requires more than describing your tone. It requires structured documentation of your writing patterns—the rules that drive how you communicate.

In this guide, you'll learn how to build custom instructions that actually analyze your writing patterns, avoid the common pitfalls that kill instruction performance, and deploy advanced techniques that work across contexts and audiences.

---

What Are Custom GPT Instructions (And Why They Matter)

Custom instructions are system-level directives you set once in ChatGPT, and the AI applies them to every conversation going forward.

Unlike individual prompts where you explain your needs each time, custom instructions are persistent. They shape how ChatGPT approaches all your requests without you having to repeat yourself.

ChatGPT has two custom instruction fields:

1. "What would you like ChatGPT to know about you to provide better responses?"

   • This is where you document yourself—your role, context, preferences, constraints
   • Think of it as your "about me"

2. "How would you like ChatGPT to respond?"

   • This is where you document your style preferences
   • Your tone, format, length, communication rules

Custom instructions interact with other ChatGPT features: Memory stores facts about you, Projects add workspace-specific context, and Custom GPTs let you build shareable assistants. Custom instructions sit at the base layer—they set your default voice, which Projects and GPTs can then override or extend. For the full setup on Projects, see our ChatGPT Projects setup guide. For a deep comparison of Memory vs. Style Profiles, see ChatGPT Memory vs. Style Profiles.

The magic happens when you stop treating these as vague personality descriptions and start treating them as explicit rules for how to generate text.

Most people write custom instructions like:

"I'm a marketing manager. I'm direct and concise. I like short paragraphs."

This is a starting point. But it's missing the specificity that makes AI actually match your voice. (For the full picture of every personalization method — including Custom Instructions, Memory, Custom GPTs, and Style Profiles — see our guide on how to make ChatGPT sound like you.)

---

The Gap Between Good and Great Custom Instructions

Before we build your instructions, let's understand why most versions fall short.

The Generic Tone Problem

When you say "professional but friendly," you've described maybe 50,000 different writing styles.

ChatGPT tries to split the difference—professional enough to sound competent, friendly enough to seem approachable. The result is the default ChatGPT voice: correct, forgettable, interchangeable with everyone else's AI output. This is The Median User Problem in action—AI trained on millions of writers converges on a generic average that sounds like no one in particular.

Your actual voice isn't a middle ground. It has specific patterns and specific rules.

The Context Blindness Problem

You write differently to different people. An email to your CEO is not the same as a Slack to your team.

Generic custom instructions don't capture audience-specific rules. So ChatGPT writes in a default tone that works "okay" for all contexts—but matches none of them perfectly.

The Anti-Pattern Blindness Problem

You not only have patterns you use. You have patterns you actively avoid.

You might never use exclamation points. You might always lead with the bottom line, never bury it. You might refuse certain clichés or corporate phrases.

Standard custom instructions miss these negative rules. ChatGPT fills in the gaps with its defaults, which sound generic.

The Multilingual Problem

If you communicate in multiple languages, generic instructions compound. Your English might be direct. Your Japanese might use formal keigo. Your French might have a different sentence rhythm.

One set of generic instructions produces generic output in every language. Download our free multilingual writing checklist to audit your cross-language consistency before you start building instructions.

---

The Custom Instructions Template That Works

Here's the framework that actually captures voice:

FIELD 1: About Me (Your Context)

This field should answer:

• What's your role and primary communication contexts?
• Who do you typically write to?
• What does success look like for your writing?

Template:

I'm a [role] at [organization/context]. I communicate primarily via:
- [Email type]: To [audience]
- [Slack/Chat]: To [audience]
- [Documents]: For [purpose]

My communication contexts:
1. [Context 1] — [Key characteristics]
2. [Context 2] — [Key characteristics]
3. [Context 3] — [Key characteristics]

Success metric: [What makes my writing effective?]

Languages: [If multilingual: List languages and default contexts for each]

Real example:

I'm a Product Manager at a B2B SaaS company. I communicate primarily via:
- Email: To executives, customers, and team members
- Slack: Internal team updates and quick questions
- Specs & Docs: Feature requirements and strategy

My communication contexts:
1. Upward (to executives) — Need clear bottom-line first, support with data, never more than 3 main points
2. Peer (to other departments) — Collaborative tone, acknowledge dependencies, propose solutions
3. Direct reports — Detailed, open to questions, encourage pushback

Success metric: People understand my intent on first read; minimal back-and-forth clarification

Languages: English (primary), Japanese (formal emails and internal communication)

FIELD 2: How to Respond (Your Voice Rules)

This is the bigger field. It documents your actual writing patterns.

Template:

BASELINE VOICE SETTINGS:
- Formality: [Your default formality level]
- Sentence length: [Short / Medium / Long / Mixed]
- Active vs. passive voice: [Your preference]
- Punctuation style: [Em-dashes / Semicolons / Minimal punctuation / Other]
- Directness: [State plainly / Hedge / Context-dependent]

CRITICAL PATTERNS YOU ALWAYS USE:
- Pattern 1: [Describe it with context]
- Pattern 2: [Describe it with context]
- Pattern 3: [Describe it with context]

PATTERNS TO NEVER USE:
- Never: [Describe what to avoid with reason]
- Never: [Describe what to avoid with reason]
- Never: [Describe what to avoid with reason]

CONTEXT-SPECIFIC ADJUSTMENTS:
When writing to [Audience 1]:
- Adjust to: [Specific style changes]
- Include: [Required elements]
- Avoid: [Specific words/phrases]

When writing to [Audience 2]:
- Adjust to: [Specific style changes]
- Include: [Required elements]
- Avoid: [Specific words/phrases]

OUTPUT RULES:
- Length: [Typical target length for different outputs]
- Format: [Preferred structure—bullets, paragraphs, numbered lists]
- Tone indicators: [How I signal emphasis or importance]

[IF MULTILINGUAL]
LANGUAGE-SPECIFIC RULES:
Language: [Language 1]
- Default formality: [Level]
- Context shifts: [How you adjust by audience in this language]
- Never use: [Language-specific patterns to avoid]

Language: [Language 2]
- Default formality: [Level]
- Context shifts: [How you adjust by audience in this language]
- Never use: [Language-specific patterns to avoid]

Real example (condensed for space):

BASELINE VOICE SETTINGS:
- Formality: Neutral-to-professional, minimal slang, no corporate clichés
- Sentence length: Mostly short (8-12 words average), occasional longer sentences for context
- Active vs. passive: Strongly prefer active voice
- Punctuation: Em-dashes for emphasis, occasional semicolons, minimal exclamation points
- Directness: Lead with the main point, always

CRITICAL PATTERNS YOU ALWAYS USE:
- Start with the action/decision/bottom line, then explain why
- Use "we" for collaborative statements, "I" for personal commitments
- Summarize trade-offs explicitly rather than hiding them

PATTERNS TO NEVER USE:
- Never: "As per my previous email" — Use "Following up on..."
- Never: "Synergize" or "leverage" — Use "improve" or "increase"
- Never: Apologize excessively — Only apologize if I caused the problem

CONTEXT-SPECIFIC ADJUSTMENTS:
When writing to executives:
- Adjust to: More formal, numbers-first
- Include: Supporting metrics and precedent
- Avoid: Jargon not in their vocabulary

When writing to direct reports:
- Adjust to: More personal, open feedback loops, ask for pushback
- Include: Context behind decisions
- Avoid: Anything that feels like a decree

OUTPUT RULES:
- Emails: 3-5 sentences maximum
- Slack: 1-2 messages max, use threads for follow-up
- Docs: Structured with bold headers, bullet points for action items

---

How to Set Up Custom Instructions: Step-by-Step

Step 1: Access Custom Instructions

1. Open ChatGPT
2. Click your profile icon (bottom left)
3. Select "Custom instructions"
4. Click "Create new"

You now have two text fields.

Step 2: Fill Field 1 ("About Me")

Start with your context. What's your role? Who do you communicate with? What languages do you use?

This field should be 150-300 words. Specific is better than vague.

Don't say: "I'm a marketer and I value clear communication." Do say: "I'm a VP of Marketing managing a team of 5. I write daily emails to the CMO and weekly Slack updates to the marketing team. I also write quarterly strategy documents for the board."

Step 3: Fill Field 2 ("How to Respond")

This is where you document your actual voice rules.

Copy the template above and fill in your patterns. Be specific about:

• What your baseline sounds like
• What you always do
• What you never do
• How you shift for different audiences
• If multilingual: how each language differs

This field can be longer—up to 8,000 characters. Use it.

Step 4: Test and Iterate

Don't expect perfection on the first draft.

Run these tests:

Test 1: Same Prompt, Different Audiences Write a prompt: "Draft a brief update on [project X] status."

Run it three times with context modifications:

1. "...to the CEO"
2. "...to my team"
3. "...to a client"

If the three outputs sound basically identical, your audience-specific rules need clarification.

If they sound different but not like your different, you're still missing specificity.

Test 2: The "Match My Writing" Test Take a real email you wrote. Ask ChatGPT to "continue this email about [topic]."

Does the continuation feel like it came from you? If not, which part feels off—the tone, the structure, the word choices?

Use that feedback to refine your custom instructions.

Test 3: The Sensitive Content Test Ask ChatGPT to draft something you find tricky to write: declining a request, giving critical feedback, delivering bad news.

Does the output match how you actually deliver these? If it's too apologetic, too harsh, or too neutral, adjust your "directness" rules.

Step 5: Refine Based on Results

After testing, go back to custom instructions and edit.

For each area where output didn't match:

• Add a specific rule
• Or clarify an existing rule
• Or add an example

Iterate until the outputs consistently feel like they're coming from you.

---

Common Mistakes (And How to Fix Them)

Mistake 1: Vague Tone Descriptions

Wrong:

"Write in a professional, friendly tone that's approachable but authoritative."

Why it fails: "Professional, friendly, approachable, and authoritative" describes thousands of voices.

Right:

"Use straightforward language. Lead with the main point. Address the reader as 'you' and myself as 'I.' Use active voice. Keep sentences short except when explaining context. Never use buzzwords like 'ecosystem' or 'synergy.'"

The fix: Replace adjectives with rules. Describe what you do and don't do, not how you're "supposed" to sound.

Mistake 2: Ignoring Context-Specific Voice

Wrong:

"Write clearly and concisely for all audiences."

Why it fails: You don't write the same way to everyone. Pretending you do makes custom instructions useless for context shifts.

Right:

"When writing to executives: Lead with numbers and outcomes. When writing to engineers: Lead with technical details and trade-offs. When writing to customers: Lead with benefits and simplicity."

The fix: List your major communication contexts. For each one, specify how your voice shifts.

Mistake 3: Missing Anti-Patterns

Wrong:

"Use clear, professional language."

Why it fails: This tells the AI what to do, not what to avoid. ChatGPT fills in gaps with defaults.

Right:

"Never: Use exclamation points. Never: Start sentences with 'Obviously' or 'Clearly.' Never: Use corporate phrases like 'reaching out' or 'touching base.'"

The fix: For every positive pattern, add the inverse. What do you explicitly avoid?

Mistake 4: Forgetting Format Preferences

Wrong:

"Write in your style."

Why it fails: This ignores that you might prefer bullets in some contexts, paragraphs in others, numbered lists for processes.

Right:

"For emails: Use short paragraphs (2-3 sentences max). For updates: Use bullet points. For feedback: Paragraph format with specific examples."

The fix: Document your format preference for each output type. ChatGPT will match it.

Mistake 5: Skipping Multilingual Rules

Wrong:

[Same instructions for all languages]

Why it fails: Your English voice doesn't translate directly to other languages. You shift formality, directness, and structure.

Right:

"English: Direct, short sentences, active voice. Japanese: Formal (敬語) for external, casual for internal, passive voice acceptable. French: Vous default unless specifically told otherwise."

The fix: If you communicate in multiple languages, add a language-specific section. Document how your voice changes by language.

---

Advanced Custom Instructions Techniques

Technique 1: The "Tone Slider" System

Instead of one formality level, define a scale.

Formality levels (1 = casual, 5 = formal):

Level 1 (Casual): "Hey, just wanted to flag that the deadline moved."
Level 3 (Neutral): "I wanted to let you know the deadline has been moved."
Level 5 (Formal): "Please be advised that the deadline has been adjusted."

For [Audience A]: Default to Level 3, shift to Level 5 if discussing budget/risk
For [Audience B]: Default to Level 2, never go above Level 3
For [Audience C]: Default to Level 1, shift to Level 3 for official communication

This gives ChatGPT a clear dial to turn based on context.

Technique 2: The "Signature Phrases" Inventory

List phrases that are distinctly you:

SIGNATURE PHRASES (use liberally):
- "Bottom line:" [You use this to lead conclusions]
- "The tricky part is..." [You use this for nuance]
- "Worth noting:" [You use this for asides]

FORBIDDEN PHRASES (never use):
- "Per my last email"
- "Circle back"
- "Low-hanging fruit"
- "At the end of the day"

This prevents ChatGPT from writing in default phrases while still using your actual patterns.

Technique 3: The "Example Bridge" System

Instead of just describing your voice, include structured examples:

EXAMPLE: Email to executive about resource request

YOUR VERSION: "We need 2 more engineers for the Q2 roadmap. Without them, we slip the mobile launch. Budget is $180K annually."

WHY: Lead with the need, then consequence, then cost. No hedging. No setup.

PATTERNS VISIBLE:
- Short sentences
- Numbers come last
- No "I believe" or "we recommend" — just state it
- Specific number of people, not "additional headcount"

When you show the example plus the analysis, ChatGPT understands not just what you wrote, but why you wrote it that way.

Technique 4: The "Audience Personality Profiles" Technique

Create mini-profiles of your major audiences:

EXECUTIVE PERSONA:
- Needs: Bottom line first, supporting data, decision clarity
- Style: Formal, numbers-driven, no surprises
- Tone from me: Confident, data-backed, brief
- Red flags: Vagueness, emotion, over-explanation

ENGINEER PERSONA:
- Needs: Technical specifics, trade-offs, implementation clarity
- Style: Detailed, trade-off aware, assumption-explicit
- Tone from me: Collaborative, specific, humble about unknowns
- Red flags: Non-technical explanation, glossing over complexity

Then reference these profiles when asking for drafts: "Draft this for the Executive Persona" gives ChatGPT much clearer targets than "write this for my boss."

Technique 5: The "Multilingual Voice Matrix" Technique

If you use multiple languages, create a comparison matrix:

| Dimension | English | Japanese | French |
|-----------|---------|----------|---------|
| Formality default | Neutral | 敬語 | Vous |
| Sentence length | Short | Medium | Long |
| Directness | High | Medium | Medium |
| Punctuation | Em-dashes | Minimal | Commas |
| Key pattern | Lead with bottom line | Provide context first | Build narrative |

This shows ChatGPT not just what to do in each language, but how your voice differs across languages.

---

Testing Your Instructions: The Verification Checklist

Before you consider your custom instructions "done," run this checklist:

Voice Consistency Test

• Ask ChatGPT to draft 3 emails on the same topic to 3 different audiences
• Do they shift appropriately?
• Does each sound like you writing to that audience?

Anti-Pattern Test

• Ask ChatGPT to write something and check for forbidden phrases
• Do your "never use" patterns appear? If yes, strengthen that rule.

Format Compliance Test

• Ask for an email. Does it match your email format preference?
• Ask for a Slack message. Does it match your Slack style?
• Ask for a document. Does it follow your document structure?

Multilingual Test (if applicable)

• Ask for the same message in multiple languages
• Does each language version feel like you in that language?
• Are formality levels appropriate?

Baseline Test

• Share a draft ChatGPT produced with someone who knows your writing well
• Ask: "Does this sound like [your name]?"
• If not, which part feels off?

Use their feedback to refine further.

---

A Practical Copy-Paste Template

Here's a complete, fill-in-the-blanks template you can use right now:

FIELD 1: ABOUT YOU

I'm a [your role] at [organization]. I primarily write:

- Emails to [audience types]
- [Communication type] to [audience]
- [Communication type] to [audience]

My main writing contexts:
1. [Context 1]: [What's at stake? How formal?]
2. [Context 2]: [What's at stake? How formal?]
3. [Context 3]: [What's at stake? How formal?]

Success = [What makes your writing work?]

[IF MULTILINGUAL: I use [languages]. Each has different conventions I follow.]

FIELD 2: HOW TO RESPOND

VOICE BASELINE:
- Default formality: [Level from 1-5]
- Sentence length: [Typical length]
- Emphasis method: [How do you emphasize?]
- Punctuation signature: [What's distinctive?]

WHAT YOU ALWAYS DO:
1. [Pattern with one example]
2. [Pattern with one example]
3. [Pattern with one example]

WHAT YOU NEVER DO:
- Never: [Avoid this, because...]
- Never: [Avoid this, because...]
- Never: [Avoid this, because...]

CONTEXT SHIFTS:
For [Audience/Context A]:
→ More [formal/casual], lead with [what], include [what], avoid [what]

For [Audience/Context B]:
→ More [formal/casual], lead with [what], include [what], avoid [what]

FORMAT RULES:
- Emails: [Preferred structure]
- [Other format]: [Preferred structure]
- [Other format]: [Preferred structure]

[IF MULTILINGUAL]
LANGUAGE RULES:
[Language 1]: [Default formality, context shifts, never patterns]
[Language 2]: [Default formality, context shifts, never patterns]

---

How to Generate Comprehensive Custom Instructions

The template above is powerful. But filling it out requires something difficult: objectively analyzing your own writing.

Most people guess at their patterns. "I think I'm direct." "I probably use short sentences." But self-analysis is inherently limited—you can't see the habits you don't notice.

A Style Profile extracts what you'd write in custom instructions if you could analyze your own writing objectively. It's not an alternative to custom instructions—it's the systematic way to build the comprehensive content above.

The difference:

• Manual approach: You guess at surface patterns—sentence length, punctuation, common phrases
• Style Profile approach: Your actual writing is analyzed forensically to extract intent-level patterns you'd never spot yourself

For example: You might use short sentences when emphasizing and longer ones when building context. Manually, you'd write "prefer short sentences." A Style Profile captures when you use short vs. long—and generates rules that reflect that strategic variation.

This is especially powerful for multilingual communicators or anyone who shifts voice significantly across contexts. If you want to understand the full picture of why AI writing doesn't sound like you, that's worth reading too. The more complex your communication patterns, the harder they are to self-document—and the more value systematic extraction provides.

The output is the same: detailed content you paste into ChatGPT's custom instructions fields. The methodology is what changes. Your Style Profile becomes part of your operational intelligence layer — the accumulated knowledge of how you communicate, encoded in a form AI can execute consistently.

---

Your Next Step

Custom GPT instructions are free and available right now in ChatGPT. You can implement everything in this guide in about 30-45 minutes.

Start with the template above. Fill it in honestly. Test it with the checklist. Refine based on what doesn't match. For ready-to-use prompt templates you can apply immediately, grab our free AI prompt collection.

The goal isn't perfection on day one. It's moving from "AI drafts I heavily edit" to "AI drafts I mostly approve."

Once you have that working, you've built the foundation for deeper voice documentation that can work across ChatGPT, Claude, Gemini, and any other AI platform you use.

---

Frequently Asked Questions

How long should ChatGPT custom instructions be?

Use as much space as you need. Field 1 ("About You") supports about 1,500 characters — enough for role context and communication patterns. Field 2 ("How to Respond") supports up to 8,000 characters. The more specific your rules, the better the output matches your voice.

Can I have multiple sets of custom instructions?

Yes. ChatGPT now lets you create multiple instruction sets and switch between them. Use one set for your default voice, another for specific contexts (e.g., client-facing work vs. internal communication).

Do custom instructions work with GPT-5?

Yes. Custom instructions apply to whatever model ChatGPT uses in your conversation. They work across GPT-5, GPT-5 Mini, and any future models. The instructions shape how the model responds, regardless of which model powers the conversation.

What's the difference between custom instructions and ChatGPT Memory?

Memory stores facts ChatGPT learns about you passively (your name, role, preferences). Custom instructions are rules you set deliberately about how ChatGPT should write. Memory handles what it knows about you; custom instructions handle how it writes for you. Use both.

Do custom instructions affect Custom GPTs?

No. Custom GPTs have their own instruction fields that override your global custom instructions. If you want the same voice rules in a Custom GPT, paste them into the GPT's instructions separately.

---

Want AI That Actually Sounds Like You?

You just read 3,000+ words on how to write custom instructions. You learned:

• How to analyze writing samples for patterns
• How to document voice baselines and anti-patterns
• How to create context-specific rules for different audiences
• How to handle multilingual variations
• How to test, iterate, and refine

That's a lot of work. And here's the thing: even if you do it perfectly, you'll capture surface patterns—sentence length, punctuation habits, common phrases. What you won't capture is the intent behind your voice. When you use short sentences to emphasize vs. long ones to build context. How your tone shifts subtly between a skeptical executive and a friendly collaborator.

My Writing Twin does what manual instructions can't.

Instead of you guessing at your own patterns, it analyzes your actual writing forensically—across dozens of samples, multiple contexts, different audiences. It detects patterns you don't even notice. The subtle rhythm shifts. The punctuation signatures. The phrases you reach for under pressure vs. when you're relaxed.

Then it generates Style Profiles that are 10x more detailed than anything you'd write manually. Not "be concise and professional"—but hundreds of specific rules, weighted by context, that tell AI exactly how you write.

The result: AI drafts you approve instead of rewrite.

What you get:

• Automatic pattern extraction — No more analyzing your own writing samples
• Context-aware profiles — Rules that shift based on audience, format, and stakes
• Multilingual support — Separate Style Profiles for each language you use
• Cross-platform compatibility — Works with ChatGPT, Claude, Gemini, and any AI tool
• Continuous refinement — Your profile improves as you use it

You can spend the next hour filling out templates and testing iterations. Or you can upload a few writing samples and get a complete Style Profile in minutes.

---

Get Your Free Writing DNA Snapshot

Curious about your unique writing style? Try our free Writing DNA Snapshot — it's free and no credit card is required. See how AI can learn to write exactly like you — for ChatGPT, Claude, Gemini, or any AI tool. Get started with My Writing Twin.

If you use Claude, check out our guide on how to make Claude sound like you for platform-specific tips. Or explore how to build custom instructions that analyze your writing patterns for a deeper framework. For the complete picture of all ChatGPT voice settings, see our ChatGPT Voice Settings guide.