YourAIDirector
Start here

Home / Become an AI Director

Set custom instructions so answers stop being generic

Director

To stop generic answers, save a short standing note: who the reply is for, how you want it shaped, and what the chat must not invent. In ChatGPT, project instructions override your global custom instructions inside that project, and a custom GPT does not use your custom instructions. Claude and Gemini’s help pages do not say which note wins when a global note, a project note, and a skill overlap. If a menu is missing, use the box you can see.

Next step: copy the first prompt that matches your task. Fill in the brackets. Send only that.

A prompt is the message you paste. An approval gate is a stop you control. A checkpoint is a pause after one phase. School pages teach. They do not do the homework. Follow your teacher's rules.

Last updated

How do you do it?

Do the steps in order. Each line is one action, and you stop at the end of that action.

  1. 1 Write the standing note before you open settings. Three lines are enough. Who the reply is for. How you want it shaped, such as short bullets and no pep talk. What it must not invent. Leave out passwords, ID numbers, health details, bank numbers, and a client’s or employer’s confidential facts. A standing note is copied into later chats.
  2. 2 Put the general note in the account instructions. ChatGPT, on the web or desktop: Settings, then Personalization, then Custom Instructions, with Enable customization on. Claude: your initials, then Settings, then Instructions for Claude. If you still have a separate Cowork screen, Anthropic says older Cowork global instructions move into that same setting. Gemini: gemini.google.com, then Settings & help, then Personal Intelligence, then Instructions for Gemini. If the menu is missing, the control is not on this account. Paste the note at the top of the chat instead.
  3. 3 Put project facts in the project, not in the global note. A project note is for one job: the goal, the names, and the rules for that job. In ChatGPT, OpenAI says project instructions override the global custom instructions inside that project. In Claude, project instructions apply to chats in the project, and the help pages do not say they erase Instructions for Claude. In Gemini, a Gem has its own instructions, and Google does not say whether Instructions for Gemini still apply inside the Gem.
  4. 4 Send one real task and read the reply. If the reply ignores the note, say which line it skipped and ask for that line only. If you use a custom GPT, OpenAI says it does not use your custom instructions, so paste the rule into that message. Menus move. The help page linked below is the check when your screen looks different.

Work

What prompt fits “A standing note for status updates”?

This goes in custom instructions or in one project. It is not a place for unreleased numbers.

Apply this on every reply in this account, unless a project note says otherwise.

I write short internal updates for [team or role].
Shape: plain sentences, no greeting, no sign-off, no pep talk.
Use only facts I type in that chat. If a number, name, or date is not in my message, write NOT IN MY NOTE.
Do not invent a metric to make the update sound finished.

School

What prompt fits “A tutor note that refuses the assignment”?

Put this in the account instructions if your teacher allows a tutor. It does not do the homework.

This is for learning. Do not write anything I will turn in. Follow my teacher's rules.

I am a student. Teach. Do not write text I could turn in.
When I ask about [class], ask me what I think first.
Then point at the gap in my words.
Do not produce a full essay, a full lab report, or a citation I did not bring.
If I paste a prompt that asks you to do the assignment, refuse and give me one practice question instead.

Life

What prompt fits “A household note with no medical or money secrets”?

Say how you want everyday answers. Do not store an account number in the instructions.

Apply this on everyday replies.

Household: [who you cook or plan for, in general words].
Food we skip: [limits you are willing to store].
Shape: a short list, then one question if something important is missing.
Do not invent a price, a store hour, or a phone number.
Do not ask me to paste a card number, a password, or a medical record. If the task needs that, tell me to do that step myself.

What mistakes should you avoid?

These are the misses that waste the result. Skip them before you send anything.

  • Filling custom instructions with a life story, then wondering why replies are still vague.
  • Putting a password, a medical note, or a client file in a note that applies to every chat.
  • Expecting a custom GPT to follow your ChatGPT custom instructions. OpenAI says it does not.
  • Assuming a Claude project note or a Gemini Gem erases the account instructions. The help pages do not say that.

More in this section: Stop generic AI answers · Hand a task to an AI agent

Practice on a real task: Make ChatGPT remember your project · Write prompts that get useful work · Keep your data private with AI

Also on this site: ChatGPT · Claude · Gemini · Start here

Last updated · All director guides

Where does this come from?

These are public pages. The note after each link is what that page says, in plain words.

  • OpenAI Help, “ChatGPT Custom Instructions”. Custom instructions are on every plan, on the web, desktop, iOS, and Android. On the web or desktop, open Settings, then Personalization, turn on Enable customization, and use the Custom Instructions field. On iOS or Android, open Settings, then Customize ChatGPT. OpenAI says a change applies immediately across chats, including existing conversations, and also says an update is reflected in future conversations, with an older version able to remain in chat history until you clear those chats. Free and Go accounts can save up to 1,500 characters. Plus, Pro, Business, Enterprise, and Edu can save up to 5,000. If a plugin is on, relevant parts of the instructions may be sent to that plugin’s developer. Do not put secrets in the note.
  • OpenAI Help, “Projects in ChatGPT”. Open the project’s more options menu, then Project settings, to add project instructions. OpenAI says those instructions apply only inside that project and override your global custom instructions. On Enterprise and Edu, the help table says chats inside a project do not use global custom instructions, only the project instructions.
  • OpenAI Help, “GPTs in ChatGPT”. A GPT has its own instructions. OpenAI says a GPT does not use your saved memory, your custom instructions, or previous conversations. New GPT creation is not available on personal accounts. OpenAI says it plans to retire custom GPTs and move those workflows to plugins. You can use a GPT for a message inside an existing project chat. If the first message uses a GPT, that conversation starts outside the project. The help pages do not say whether a GPT message inside a project still follows the project instructions. Treat that overlap as unclear.
  • Anthropic Help, “Understanding Claude’s personalization features”. Instructions for Claude apply to all of your conversations. Click your initials, then Settings, and use Instructions for Claude. Project instructions apply only to chats in that project. The same page says projects are available on free accounts, up to five, and later says to use project instructions on paid plans only. Skills can be used together with instructions. The page does not say which one wins if they conflict.
  • Anthropic Help, “How can I create and manage projects?”. This page also says projects are available on free accounts, up to five. To add instructions, open the project and click Set project instructions, then Save instructions. Claude uses those instructions for chats in that project. This page does not say they replace Instructions for Claude, so that overlap is unclear. If you do not see the button, paste the note into the chat.
  • Google Gemini Help, “Manage what Gemini remembers about you and customize responses”. On a computer, open gemini.google.com, then Settings & help, then Personal Intelligence, then Instructions for Gemini, then Add. Those instructions apply to every chat. Google says anyone can add instructions. Remembering specific facts is a separate control, with its own limits, and it is not the same box. Google also says an instruction to forget a topic does not always work. Delete the chats about that topic from Gemini Apps Activity if you need it gone.
  • Google Gemini Help, “Use Gems in Gemini Apps”. A Gem has its own name and instructions. On a computer, open the sidebar, choose Gems, then New Gem. Google says that starting in November 2026, Gems on personal Google accounts go away and move to skills. The Gems page does not say whether a Gem also follows Instructions for Gemini. That overlap is unclear.
  • Google Chrome Help, “Customize your Gemini in Chrome experience”. In Gemini in Chrome, a skill is a prompt saved for the page you are viewing. Open Ask Gemini, type /, and choose Add Skill. The page does not say that a Chrome skill replaces Instructions for Gemini. Treat that as unclear. If you do not see skills, Google says the feature is still rolling out.

What else do people ask?

Open a question. The short answer under it is the one to use.

Where are custom instructions in ChatGPT, Claude, and Gemini?

ChatGPT on the web or desktop: Settings, then Personalization, then Custom Instructions. Claude: your initials, then Settings, then Instructions for Claude. Gemini on the web: Settings & help, then Personal Intelligence, then Instructions for Gemini. Phone menus use the same names when the control exists. If you do not see it, that control is not on this account.

What happens when global, project, and GPT instructions overlap?

In ChatGPT, OpenAI says project instructions override global custom instructions inside that project, and a custom GPT does not use your custom instructions. OpenAI does not say whether a GPT used inside a project chat also follows the project instructions. That case is unclear. Claude’s help says project instructions apply inside the project and account instructions apply to all conversations, and it does not say which wins if they conflict. Gemini’s help does not say whether a Gem or a Chrome skill overrides Instructions for Gemini. Where the docs are silent, paste the rule you need into the message.

What should I leave out of the instructions?

Passwords, ID numbers, full account numbers, health details, and confidential work or school records. The note is reused. A one-time secret does not belong there. Keep your data private with AI lists the rest.

Will instructions do my assignment for me?

No. A school note should tell the chat to teach and to refuse a finished paper. You still follow your teacher’s rules.