ChatGPTCustom instructionsLevel 1: Basic prompts

How Do You Set Up ChatGPT Custom Instructions?

Open Settings > Personalization, check that Enable customization is on, write a short list of preferences in the Custom Instructions field, and save it. On iOS and Android, go to Settings > Customize ChatGPT. OpenAI's docs say custom instructions are for preferences you want ChatGPT to follow across chats, such as your preferred response style.

Prompt to copy
Paste this into the Custom Instructions field, then replace the brackets:

I'm a [role] working in [field]. I mostly use ChatGPT for [two or three common tasks].
I know [topics you know well], so skip the basics there. I'm new to [topics you don't know], so define terms.

When you answer:
- Lead with the answer, then the reasoning.
- Keep it under [200] words unless I ask for more.
- Use [US English] and [metric units].
- Use a table when I ask you to compare options.
- If a fact may be out of date, search the web and link the source.
- If my request is missing something that would change the answer, ask one question first.
  • Free and Go1,500
  • Plus, Pro, Business, Enterprise, Education5,000
Custom Instructions field limit, in characters. Paid plans moved up from 1,500 on July 15, 2026. Source: OpenAI's Help Center, checked 6 November 2026.
  • Settings > PersonalizationThe Custom Instructions field
  • Settings > Customize ChatGPTOn iOS and Android
  • ~/.codex/AGENTS.mdYour global file for Codex
Where your instructions live on each surface, from OpenAI's Personalize page and Help Center.

Try it now

  1. 1On chatgpt.com, open Settings > Personalization. OpenAI's Academy also describes opening your profile and clicking Personalize.
  2. 2Paste the text above into the **Custom Instructions field, fill in the brackets, and save. If the field is off, turn on Enable customization** first.
  3. 3Select New chat and repeat a question you asked recently. It worked if the reply follows your format without being told. If it doesn't, make the rule more specific.

Use it when

When you've typed the same line into three different chats, such as "use US spelling" or "I'm not a developer, so explain the terms." Set it up in your first week, and do it before you judge ChatGPT's quality, since many weak answers come from it not knowing who you are.

Skip it when

Skip anything that applies to one client or one body of work, because it would follow you into unrelated chats. That belongs in project instructions. A one-off also stays in the prompt. OpenAI's Academy gives the example "Draft this in a casual, conversational tone" as guidance for a single chat.

Why it works

OpenAI's Prompting guide draws the line clearly: preferences that apply across chats belong in personalization settings, and details that matter only to the current chat stay in the prompt. Moving your repeated lines into settings shortens every prompt after that.

A separate control, Base style and tone, is on the same page. It sets how ChatGPT responds, and OpenAI says a personality changes style and tone but not capabilities. So a preset can handle "friendly" or "direct", while a rule like "put the table first" or "use metric units" needs written instructions.

Codex keeps these somewhere else. OpenAI's Personalize page says that in Codex, personal instructions are stored in your global AGENTS.md file. So if you code with Codex, add your coding rules to ~/.codex/AGENTS.md as well.

Example

A designer who talks to engineers all day writes: "I'm a product designer. I don't write code. When you mention APIs or databases, explain them in one plain sentence first. Keep answers under 150 words and end with one question I could ask my engineers." A question about rate limits now arrives in plain language with a question to take to the next meeting, and she doesn't need to set it up every time.

Going further

To confirm that Codex has picked up your global file, OpenAI's AGENTS.md page suggests running codex --ask-for-approval never "Summarize the current instructions." from a repository root. It should reference your guidance. If a recent edit seems to be ignored, restart Codex. Combined instruction files are capped at 32 KiB by default (project_doc_max_bytes), so keep the global one short.

OpenAI's Help Center puts the ChatGPT field's limit at 1,500 characters on Free and Go, and 5,000 on Plus, Pro, Enterprise, Business and Education. Paid plans moved up from 1,500 on July 15, 2026, so older advice about a 1,500-character cap may no longer apply to you.

Common mistakes

  • Writing rules nobody can check, like "be smart" or "be helpful". If you can't tell whether a reply followed the rule, it won't change much.
  • Adding one client's name and style guide, then getting that client's tone in a chat about your holiday.
  • Setting them on the web and expecting Codex to follow. OpenAI's docs say Codex keeps personal instructions in AGENTS.md.
  • Fighting your own settings, such as a warm Base style and tone alongside a rule demanding terse replies.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Beyond the course