Best AI Prompts for ChatGPT Instructions (2026)
Prompts for ChatGPT instructions#
Five starting points#
These prompts ask ChatGPT to draft the instructions for you. You then paste the result into Settings > Personalization (custom instructions) or into a project's instructions, and edit it by hand.
Act as a prompt designer. Write custom instructions for my ChatGPT account from the notes below. Keep the whole text under [1,500 OR 5,000] characters. Use three short sections: who I am and what I use ChatGPT for, how I want answers formatted, and what to never do. Write each rule as a direct instruction and add a one-clause reason where the reason changes behaviour. Do not include anything that changes weekly. My notes: [PASTE YOUR ROLE, TOOLS, PREFERRED TONE AND FORMAT, AND PET PEEVES].Act as a prompt editor. Rewrite [PASTE YOUR LONG CONVERSATIONAL PROMPT] as a reusable instruction block with labelled sections for identity, rules, examples and context, using XML tags to wrap any pasted reference material. Replace vague words such as "good" or "professional" with checkable requirements, and add three examples of input and ideal output that differ from each other. List anything you had to assume.Act as a ChatGPT project architect. Write project instructions for a project called [NAME OF YOUR PROJECT] whose purpose is [DESCRIBE THE RECURRING WORK]. State the role ChatGPT plays, the standing rules for every chat in this project, the output format, and when to ask a clarifying question before answering. Then list which of my files should be uploaded, given a limit of [NUMBER] files, and which should be merged or left out.Act as a configuration auditor. I have preferences, facts and rules that must persist in ChatGPT. Sort each item below into one of four places: global custom instructions, project instructions, saved memory, or a pasted-in-each-time template. Use this rule: custom instructions for stable style and rules, project instructions for rules tied to one body of work, memory for high-level facts, and templates for exact wording that must not vary. Explain each placement in one line. Items: [PASTE YOUR LIST OF PREFERENCES, FACTS AND RULES].Act as a test designer. Here are my current ChatGPT instructions: [PASTE YOUR INSTRUCTIONS]. Write eight test prompts I can run in a fresh chat to check whether each rule holds: two ordinary requests, two that tempt the model to break a rule, two edge cases, and two where rules could conflict. For each test, state the behaviour that counts as a pass and the behaviour that counts as a failure.What to add to any ChatGPT instructions prompt#
- Your character budget. OpenAI's custom instructions page lists 1,500 characters for Free and Go and 5,000 for Plus, Pro, Business, Enterprise and Education, so tell the drafting prompt which cap applies.
- The reason behind each rule. Anthropic's guide says explaining why a behaviour matters gives the model more information about your goals, and "never use ellipses" works better as "this will be read aloud, so avoid ellipses".
- Precise, checkable wording. OpenAI's guide says GPT models perform best with explicit, precise instructions describing the output you want.
- Three diverse examples at most, not near-duplicates. Anthropic suggests 3 to 5 examples that are relevant and diverse; OpenAI says to show a diverse range of inputs with desired outputs. Google recommends keeping their structure identical.
- A structure the model can parse. OpenAI's guide orders a durable message as identity, instructions, examples, then context, with Markdown headers for sections and XML tags for pasted material.
- The scope. Say whether the text is for your global settings or one project, because OpenAI states project instructions override global custom instructions inside that project.
A worked example#
Here is the first prompt filled in for a hypothetical freelance analyst on a free plan.
Act as a prompt designer. Write custom instructions for my ChatGPT account from the notes below. Keep the whole text under 1,500 characters. Use three short sections: who I am and what I use ChatGPT for, how I want answers formatted, and what to never do. Write each rule as a direct instruction and add a one-clause reason where the reason changes behaviour. Do not include anything that changes weekly. My notes: I am a freelance market analyst writing client briefs in UK English. I mostly ask for summaries of reports, spreadsheet formulas in Excel, and edits to my drafts. I want the answer first, then reasons, in short paragraphs. I dislike emojis, filler openers and made-up statistics. When I paste a draft, change as little as possible and show what changed. If a request is ambiguous, ask one question before answering.
A good result for this task:
- Fits the 1,500-character cap. Count it yourself in the settings box, because the model's own count may be off.
- Contains rules, not a life story, and nothing tied to this week's client or deadline.
- Turns "no made-up statistics" into a behaviour, such as saying when a figure is unknown, with a reason.
- Keeps one rule per line so you can test and delete rules individually.
- Does not contain passwords, client names or other sensitive data.
Privacy matters as much as wording. OpenAI says information from using custom instructions may be used to improve models unless you opt out, and that if Memory is on you should avoid entering information you would not want remembered. Do not put client data or secrets in instructions. In a project, check the memory setting at creation: OpenAI says project-only memory can be chosen only when you start a new project, and sharing a project forces it to project-only.
The full prompt library is on our prompt library page.
Limits of this guide#
OpenAI's pages tell us the character caps, where the settings live, how project instructions override global ones, and what is stored. They do not tell us which wording makes ChatGPT follow instructions best, and we did not test the prompts above. The memory and projects pages were read from earlier archive copies (January and December), because later copies were blocked, so file limits and plan rules may have changed. We could not confirm a character limit for project instructions.
Sources & Further Reading#
All sources fetched on 2026-10-22. OpenAI Help Center pages are bot-blocked and were read through Internet Archive copies; the snapshot date is given in each row.
| Claim | Source | Section or date |
|---|---|---|
| Custom instructions apply to all chats; 1,500 characters on Free and Go, 5,000 on Plus, Pro, Enterprise, Business and Education; not shared with shared-link viewers; usage may improve models, with opt-out; edits apply to future conversations | ChatGPT Custom Instructions (OpenAI Help Center) | Archive snapshot 2026-10-05 |
| Saved memories work like custom instructions; memory is for high-level details, not exact templates; Temporary Chat; "Improve the model for everyone" may include saved memories; avoid entering what you would not want remembered | Memory FAQ (OpenAI Help Center) | Archive snapshot 2026-01-08 |
| Project instructions override global custom instructions; chats moved into a project inherit instructions and files; project-only memory chosen at creation; sharing forces project-only memory; file limits by plan | Projects in ChatGPT (OpenAI Help Center) | Archive snapshot 2025-12-10 |
| GPT models need explicit, precise instructions; identity, instructions, examples, context order; Markdown and XML delimiters; diverse examples; evals to catch regressions | Prompt engineering (OpenAI) | Undated; read 2026-10-22 |
| Explain the reason behind instructions; 3 to 5 relevant, diverse examples; XML tags to separate content types | Prompting best practices (Anthropic) | Undated; read 2026-10-22 |
| Put constraints and role in system instructions; include few-shot examples with identical structure; state the response format | Prompt design strategies (Google) | Undated; read 2026-10-22 |
Need a different task?
Browse all 36 tasks and 180 prompts, with search and level filters.
Frequently asked questions
What are the best AI prompts for ChatGPT custom instructions?
Put standing preferences in custom instructions and task rules in project instructions, and keep both short. OpenAI caps custom instructions at 1,500 characters on Free and Go and 5,000 on paid plans, so spend them on rules, not on biography.
Do project instructions replace my custom instructions?
Inside a project, yes. OpenAI says project instructions only apply inside that project and override your global custom instructions. Chats outside the project still use your custom instructions. A chat moved into a project inherits the project instructions and files.
Are my custom instructions and memories private?
Partly. OpenAI says custom instructions are not shared with shared-link viewers, but information from using them may improve models unless you opt out. Saved memories can be used for training if "Improve the model for everyone" is on. Business, Enterprise and Edu are not trained on by default.
By TheBestSuperintelligence Editorial Team. Last reviewed .
How we sourced this. Model names, features and limits change often. The prompts below are our own untested starting points, and we did not measure which wording performs best. Facts are as printed on each source page on the date shown. Methodology
