Most people use ChatGPT like they are meeting it for the first time, every single time. Then they act surprised when it answers like a polite stranger in a shop queue.
Custom Instructions fix that. They let you tell ChatGPT who you are, what matters to you, and how you want answers delivered, so you stop repeating the same setup over and over. Think of them as your standing preferences, not your one-shot prompt of the day.
What ChatGPT Custom Instructions Are
Custom Instructions are account-level directions you save inside ChatGPT. They tell it what to keep in mind when replying, such as your role, goals, writing preferences, formatting rules, or the kind of tone you like. OpenAI says they apply across chats right away, and they are available on Web, Desktop, iPhone, and Android on all plans.
That means you can stop typing things like “keep it concise,” “use bullet points,” “assume I am a beginner,” or “write in plain English” in every fresh chat. Your future self will thank you, even if it does so in a well-formatted paragraph.
Where To Find Them
On web and desktop, open your profile, go to Personalization, turn Enable customization on, and enter your instructions. On iPhone and Android, open Settings, choose Customize ChatGPT, switch customization on, and add your instructions there.
The key thing most older articles miss is that ChatGPT’s personalization area is now a bit more layered than it used to be. Depending on your account, you may also see personality choices and newer style controls in the same general area, which can overlap with how your Custom Instructions feel in practice.
What Custom Instructions Actually Change
They mostly change how ChatGPT talks to you and what defaults it carries into a chat. That can include:
- Your background or role
- Your main goals
- Your preferred tone
- Your preferred output format
- Your default language or reading level
- Your standing rules, like “ask clarifying questions first” or “give the direct answer before the explanation”
Used well, they make ChatGPT feel less generic. A writer can make it keep things sharp and readable. A marketer can make it think in terms of audience, offer, and conversion. A beginner can make it explain things without jargon. A developer can make it show code first and commentary second.
What They Do Not Do
Custom Instructions are useful, but they are not magic. They do not turn ChatGPT into a perfect clone of your brain. They do not override safety rules. They do not guarantee perfect obedience on every answer. And they do not replace giving a clear prompt when a task is specific or high-stakes.
They also do not rewrite the past. If you change your settings, the new version takes effect right away, even in ongoing chats, but earlier answers do not get retrofitted like a kitchen remodel on last year’s house.
And one more easy-to-miss detail: custom GPTs are separate. OpenAI says GPTs do not use your saved memory, your Custom Instructions, or your previous conversations. Each GPT conversation starts fresh unless that GPT has been configured with its own instructions.
Custom Instructions Vs Memory Vs Personality Vs GPTs
| Feature | What It Does | Best Use |
|---|---|---|
| Custom Instructions | Manual rules you set for how ChatGPT should respond and what it should know | Stable preferences you want across chats |
| Memory | Useful details ChatGPT remembers from conversations or when you ask it to remember something | Personal preferences, goals, and recurring context |
| Personality | Changes the style and tone of replies, not the core capabilities | Making ChatGPT feel more professional, friendly, candid, and so on |
| GPTs | Separate custom versions of ChatGPT with their own instructions, tools, and files | Task-specific assistants for a workflow or project |
Here is the plain-English version: use Custom Instructions for rules you want to set on purpose, use Memory for details you want ChatGPT to remember over time, use Personality for the vibe, and use a GPT when you want a dedicated assistant built for one job.
If you work inside Projects, that is another clue. OpenAI says Projects can also carry their own instructions, which is often better than stuffing every rule into your global settings. Account-wide instructions are for broad defaults. Project instructions are for one lane of work.
How To Write Custom Instructions That Actually Help
The best Custom Instructions are usually boring in a good way. Clear. Short. Useful. They are not ten pages of control freak energy trying to micromanage every sentence.
- Start with durable facts. Add things that stay true across many chats, such as your job, skill level, target audience, or main use cases.
- Add response rules that save time. Ask for the format you usually want: concise first, then detail; tables when comparing; examples after definitions; no fluff.
- Keep it broad enough to travel. Good instructions work across many chats, not just one article, one email, or one spreadsheet.
- Avoid contradictions. “Be concise” and “cover every detail” are not best friends.
- Leave room for context. ChatGPT still needs to adapt to the task in front of it.
A simple formula:
I am [role]. I usually use ChatGPT for [tasks]. Assume I am [skill level]. Respond in [tone]. Format answers with [structure]. Default to [preferences]. Avoid [things you dislike].
That is usually enough. You do not need to write a constitution for the machine.
Good Examples You Can Steal
Here are a few starter versions that are actually useful.
For Writers
I write for a general audience. Use plain English, short paragraphs, and varied sentence length. Start with a strong hook. Avoid filler, jargon, and generic intros. Give the direct answer first, then build out the details.
For Marketers
I work in marketing. Frame ideas around audience, offer, problem, and conversion. Keep suggestions practical. Prefer clear headlines, punchy summaries, and examples that could work in real campaigns.
For Learning
Assume I am a beginner unless I say otherwise. Explain terms simply. Use one real example for each key point. If something is likely to confuse a beginner, flag it before moving on.
For Coding Help
When giving code, show the working code block first. Then explain it in simple steps. Point out likely errors and how to test the solution. Keep extra theory brief unless I ask for it.
These work because they set defaults without strangling the answer. That is the balance you want.
Common Mistakes That Make Them Worse
- Writing a giant wall of text. OpenAI’s long-form Custom Instructions fields have a 1,500-character limit, and stuffing them full does not automatically make the results better.
- Using one-time task details. A standing instruction should not be “write my blog post about sea turtles.” That belongs in the prompt, not in the permanent settings.
- Trying to force every answer into one shape. A table is great until you ask for a heartfelt note. Then it gets weird.
- Forgetting the overlap. Your personality setting, memory, chat-level instructions, and newer style controls can all affect the final result too.
This is why some people think Custom Instructions “stopped working” when the real problem is that they piled on too many competing controls. They did not build a helpful setup. They built a committee.
When To Turn Them Off
Turn them off when you need a clean slate. That includes cases where you are testing prompts, doing a task far outside your normal workflow, comparing outputs fairly, or troubleshooting strange behavior. OpenAI lets you disable customization without deleting your saved text, which is handy when you want to step out of your usual lane for a while.
Temporary chats can also help when you do not want memory involved. And if you are working with one very specific long-running job, a Project or a custom GPT may be a cleaner home for those instructions than your global settings.
Are They Safe To Use?
Useful, yes. Private in the absolute sense, no tool should be treated that way. OpenAI says you can edit or delete Custom Instructions at any time, they are included in data export, and if you use third-party plug-ins, relevant information from your instructions may be shared with those developers. So the sensible rule is simple: do not put anything there that would make you sweat if it left the room.
The Best Way To Think About Them
Custom Instructions are not there to do the whole job for you. They are there to remove the boring repetition.
Used badly, they become a cluttered drawer full of half-dead preferences. Used well, they turn ChatGPT from “generic chatbot on standby” into “assistant that already knows how you like to work.” That is a small shift on paper. In practice, it is the difference between starting every conversation from zero and starting three steps ahead.
And in a tool built on words, starting three steps ahead is not a minor upgrade. It is the whole game.
