ChatGPT Custom Instructions for Coding: 10 Setups to Copy

ChatGPT custom instructions for coding stop you from repeating your stack, versions, and style rules in every chat. Below are 10 setups to copy by role, plus how to fit them in your plan's limit. Swap the [variables] for your real tools and versions before you save.

Safety note: Custom instructions apply to every chat, so never put API keys, passwords, tokens, internal URLs, or client code in them. Describe your stack and rules instead, and review any code ChatGPT writes before you merge it.

Where they go and how long they can be

On web and desktop, open Settings, then Personalization, and make sure customization is on. On iOS and Android, open Settings, then Customize ChatGPT. Per OpenAI's help center, the limit is 1,500 characters on Free and Go plans and 5,000 on Plus, Pro, Business, Enterprise, and Edu. Changes apply to all your chats.

Setups by role

1. Backend developer

I'm a backend developer. Stack: [language and version], [framework], [database and version], [test tool]. Code first, then up to 3 short bullets. If a function might not exist in my versions, say so instead of guessing.

2. Frontend developer

I build UIs with [framework and version], [styling tool], and [state library]. Use function components and [language] types. Show only changed components. Point out accessibility issues.

3. Data engineer

I write pipelines in [language] with [tools and versions]. Prefer readable SQL with comments. Flag anything that could scan a full table or break on null values.

4. Mobile developer

I build [platform] apps with [language and version] and [UI framework]. Target [minimum OS version]. Mention when an API needs a newer OS version.

5. DevOps engineer

I work with [cloud], [infrastructure tool], and [CI tool]. Never suggest commands that delete resources without a dry-run step first. Explain each flag.

6. Learner

I'm learning [language] and know [what you know]. Explain new ideas in plain words, give a small example, and ask me to try the next step myself before showing the answer.

Setups by habit

7. Tests always

For any bug fix or new function, include a [test framework] test. For bugs, give the likely cause first.

8. No guessing

If you are not sure an API, flag, or package exists in [versions], say so and suggest how I can check.

9. Short answers

No intros, no restating my question, no summary at the end. Code, then at most three bullets.

10. Review mode

When I paste code without a question, review it: bugs first, then security, then readability, each with a line number.

Build yours from your stack

The flagship Coding Custom Instructions Prompt: ChatGPT Setup by Stack takes your role, stack and versions, code style, answer format, plan limit, and pet peeves. It returns a ready block with a character count, a fix line for each peeve, what to leave out, test prompts, and where to paste it. Its example sets up a Python 3.12 and FastAPI backend developer.

Tips

Keep global instructions about how you work, and put one repo's details in that chat instead. Update versions when you upgrade. If you call models through the API, custom instructions do not apply there; use a system message instead.

Related prompts

For non-coding setups, see the ChatGPT custom instructions examples, and clean up an old block with the custom instructions audit and rewrite prompt. If you use an agent that edits code, read the system prompt for a coding agent.

Find more in the AI Tools category, look through Browse, or post your own on Share.