Tired of telling ChatGPT the same things in every chat, such as your job, how long answers should be, or “please stop using so many emojis”? Custom instructions fix that. You write your preferences once in Personalization, and ChatGPT follows them in every new chat.
In 2026 Personalization grew beyond the single custom instructions box. It now also has a Base style and tone preset and Characteristics sliders. Memory and Projects sit alongside them too, so it can be confusing to know which setting does what.
This guide explains each one and shows how to set them on the web and in the app. It also gives copy-ready custom instruction examples for students, developers, writers and business users, and covers how to fix instructions that ChatGPT ignores. Details are checked against OpenAI’s help center and Academy pages in November 2026.
- Where: profile icon > Personalization (web, desktop and mobile).
- Custom instructions tell ChatGPT about you and how to answer. They apply to new chats until you change or turn them off.
- Base style and tone: Default, Professional, Friendly, Candid, Quirky, Efficient or Cynical. It changes how replies sound, not what ChatGPT can do.
- Characteristics: nudge Warm, Enthusiastic, Headers & Lists and Emojis up (+) or down (-).
- Rule of thumb: stable preferences go in custom instructions; facts that change go in memory; topic-specific rules go in a Project.
- Keep it short and concrete. A few clear lines beat a wall of text.
What Are ChatGPT Custom Instructions?
Custom instructions are standing preferences that ChatGPT reads before your messages. OpenAI describes them as settings that tell ChatGPT about you and how you want it to respond. They stay in effect for new conversations until you change, disable or remove them.
Typical things to include:
- Who you are: “I lead customer onboarding at a SaaS company” or “I am a second-year nursing student in India”.
- Tone: concise, formal, friendly, plain English.
- Format: bullet points, tables, copy-ready drafts, code only.
- Guard rails: “Ask clarifying questions if requirements are unclear”, “Say when you are not sure”.
They do not give ChatGPT new abilities and they do not override OpenAI’s safety rules. They shape the style and focus of answers.
How to Set Custom Instructions
On the web or desktop app
- Click your profile icon (bottom left on the web).
- Choose Personalization.
- Find the Custom instructions area and type how you want ChatGPT to respond.
- Fill in the optional fields about you (such as nickname, occupation and more about you) if your account shows them.
- Click Save, then start a new chat to see the effect.
On iPhone and Android
- Open the sidebar and tap your profile icon to open Settings.
- Tap Personalization.
- Edit your custom instructions and the about-you fields, then save.
Your settings sync across web, desktop and mobile on the same account. Menus can vary slightly by plan, region and app version as OpenAI rolls out changes gradually.
Base Style and Tone (Personalities)
Under Personalization you will see a Base style and tone dropdown. It sets the overall personality of replies across all chats, including existing ones.
| Option | What it feels like | Good for |
|---|---|---|
| Default | Neutral, standard ChatGPT style | Most people |
| Professional | Polished and businesslike | Work emails, reports |
| Friendly | Warm and conversational | Learning, casual help |
| Candid | Direct and honest, less sugar-coating | Feedback on your work |
| Quirky | Playful and creative | Brainstorming, fun chats |
| Efficient | Short and to the point | Quick answers, busy users |
| Cynical | Dry, sarcastic humour | Entertainment |
OpenAI notes that the personality only changes how responses sound. For specific content like code or an email draft, ChatGPT generally follows the tone your request needs rather than the preset. A saved memory such as “keep replies serious” can also tone down a playful personality.
Characteristics: Warm, Enthusiastic, Lists, Emojis
Characteristics fine-tune the feel of answers on top of your base style. Each can be nudged more (+) or less (-):
| Characteristic | More (+) | Less (-) |
|---|---|---|
| Warm | Sincere, friendly tone | More emotionally neutral |
| Enthusiastic | More excitement and interest | Calmer and more objective |
| Headers & Lists | More headings, bullet lists and tables | More paragraph-style writing |
| Emojis | Slightly more emoji | Fewer, and none in informational answers |
Set them under Personalization > Characteristics on the web or in the app. OpenAI says the feature is rolling out gradually, so you may not see it yet. If you only ever wanted “fewer emojis” or “no bullet points”, Characteristics are quicker than writing an instruction.
Custom Instructions vs Memory vs Projects vs GPTs
| Feature | What it is | Scope | Use it for |
|---|---|---|---|
| Custom instructions | Rules you type yourself | All new chats | Role, tone, format, guard rails |
| Base style and Characteristics | Presets and sliders | All chats | Overall feel of replies |
| Memory | Facts ChatGPT saves or you ask it to remember | All chats (if on) | Changing details: current projects, preferences learned over time |
| Projects | A folder with its own instructions and files | Chats in that project | One client, one course, one book |
| Custom GPTs | A shareable assistant with its own instructions | When you open that GPT | Repeated workflows, sharing with others |
OpenAI’s own tip: use custom instructions for stable preferences like role, tone and format, and use the chat prompt for task-specific requests. For deeper dives see our guides to ChatGPT memory, ChatGPT Projects and creating a custom GPT.
How to Write Good Custom Instructions
- Be specific. “Keep answers under 150 words unless I ask for detail” works; “be concise” is vague.
- Say who you are and what you know. “I am a beginner in Python” stops over-technical answers.
- State the format you want. “Start with a one-line answer, then details” or “Use tables for comparisons”.
- Add guard rails. “If you are unsure, say so”, “Ask one clarifying question before long tasks”.
- Use positive instructions. “Write in plain British English” is clearer than a list of don’ts.
- Keep it short. Five to ten clear lines are usually enough. Long lists get followed less reliably.
- Do not paste private data. No passwords, bank details or ID numbers.
- Test and refine. Start a new chat, ask a normal question, and adjust the wording if needed.
Copy-Ready Examples
Student
I'm a first-year engineering student. Explain concepts step by step with
simple examples, then give a 3-question quiz at the end. Show formulas
clearly. If I ask for homework answers, guide me with hints first.
Developer
I'm a senior TypeScript and Python developer. Give code first, explanation
after, and keep explanations short. Use modern syntax and type hints.
Mention security issues and edge cases. Don't explain basics unless asked.
Writer or marketer
I write blog posts for a general audience. Use short sentences, active
voice and a friendly tone. Avoid buzzwords and clichés. When I ask for
headlines, give 5 options under 60 characters.
Business user
I manage a small team. Start every answer with a one-line summary. Use
bullet points and tables for comparisons. Flag risks and costs. Ask a
clarifying question if a request is ambiguous.
Non-native English speaker
English is my second language. Use simple words and short sentences.
After complex answers, add a 2-line summary in Hindi.
Fact-checker mode
When you state facts, say how confident you are. Separate facts from
opinions. If something may have changed recently, tell me to check an
official source and suggest which one.
Turn Off or Bypass Custom Instructions
- Turn off or clear: go to Personalization and switch custom instructions off or delete the text, then start a new chat.
- Just this once: use a Temporary Chat and choose Unpersonalized before your first message. It ignores memory and custom instructions.
- In one message: say “Ignore my usual format for this answer and write a long essay.” Instructions in the chat can override the default style.
- For one topic: create a Project with its own instructions instead of changing your global settings.
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| ChatGPT ignores my instructions | Old chat, or instructions too long or vague | Start a new chat; shorten and make rules concrete |
| Personality has no visible effect | A memory or your request conflicts with it | Check memory for tone preferences; personality matters less for code and drafts |
| Characteristics option missing | Gradual rollout | Update the app and check again later |
| Instructions apply in the wrong context | Global setting used for a specific job | Move topic rules into a Project or a custom GPT |
| Different behaviour on phone and web | Different account signed in | Check the account in Settings on both devices |
| Work account will not let me edit | Workspace admin controls | Ask your admin |
| Answers suddenly changed style | New model or personality update | Re-check Base style and tone and your instructions |
Still unsure if paying helps? Our Is ChatGPT Plus worth it guide compares what each plan adds.
FAQ
Are custom instructions free?
Yes. Custom instructions are available on Free and paid ChatGPT plans.
Do custom instructions apply to old chats?
They apply to new conversations. Start a new chat after you change them. (The Base style and tone preset is different: OpenAI says it applies across all chats, including existing ones.)
What is the difference between custom instructions and memory?
Custom instructions are rules you write yourself. Memory is information ChatGPT saves about you over time. Use instructions for how to answer and memory for facts about you.
Do custom instructions work in custom GPTs and Projects?
Projects and GPTs have their own instructions, which focus on that context. Keep your global instructions general so they do not clash.
Can I have different custom instructions for work and personal use?
Not in one account’s global settings. Use Projects with their own instructions, or separate accounts (the mobile app supports switching between two signed-in accounts).
Will custom instructions make ChatGPT break its rules?
No. They change style and focus only. OpenAI’s safety rules always apply.



