Can you save prompts in ChatGPT?
Yes, but not as a prompt library: ChatGPT has no "saved prompts" list. What it has is four places to keep instructions you reuse: custom instructions (every chat), project instructions (every chat in one project), Skills (named workflows, on Business, Enterprise, Healthcare and Edu plans) and custom GPTs, which retire on December 11, 2026.
Those features store standing guidance, not a menu of separate prompts. If you want ten prompts you pick from on demand, such as a summary format, a review checklist and a reply template, you keep them outside ChatGPT and paste them in. This guide shows how each built-in option works, what it is good for, and how to keep the rest.
Where can ChatGPT keep instructions you reuse?
| Option | Where it applies | Plans (as of October 2026) | Best for | Not good for |
|---|---|---|---|---|
| Custom instructions | Every chat | All plans | Your role, tone and format preferences | Task prompts you only want sometimes |
| Project instructions | Chats inside one project | Depends on plan and workspace settings | Context for one client, product or long task | Prompts you use across projects |
| Skills | Loaded when a request matches, or invoked directly | Business, Enterprise, Healthcare, Edu | Repeatable multi-step workflows | Short one-off prompts |
| Custom GPTs | Chats with that GPT | Retiring December 11, 2026 | Existing GPTs until you migrate | Anything new |
| Memory | Context carried from earlier chats | Where available | Facts about you and your work | Exact prompt text |
| Outside library | Wherever you paste | Any | Many separate prompts with blanks | Instructions that should always apply |
The plan column comes from OpenAI's help articles on custom instructions, Skills in ChatGPT and custom GPT retirement. Plan details change, so check the article for your account.
How do you save a prompt as a custom instruction?
Custom instructions are the right home for a prompt you want in every chat. OpenAI's personalization docs describe them as "preferences you want ChatGPT to follow across chats."
- Open Settings, then Personalization on web and desktop, or Settings, then Customize ChatGPT on mobile (labels can vary by app version).
- Make sure the customization toggle is on.
- Paste the part of your prompt that should always apply, such as "Answer in short paragraphs, lead with the answer, and flag anything you are unsure of."
- Start a new chat to check the effect.
Keep this short and general. A custom instruction that says "always write a code review in this format" will leak into chats where you only wanted a recipe. Tone, format and facts about you belong here; task prompts do not. For ready-made wording, see these ChatGPT custom instructions examples.
How do you save a prompt as project instructions?
A project groups related chats with their own files and instructions. OpenAI's projects docs say "Project instructions apply across its chats," and suggest you "Create a project when work will continue over time."
- Create a project for one body of work, such as a client, a product launch or a thesis.
- Paste your standing prompt into the project's instructions: the role, the audience, the output format and the rules.
- Upload the files the prompt depends on, such as a style guide or a product brief.
- Start every chat for that work inside the project.
This is the closest ChatGPT gets to a saved prompt with context attached. The limit is scope: the instructions live in that one project, so a good set gets copied by hand into the next one. The ChatGPT Projects guide covers files, sharing and memory in detail.
When should a saved prompt become a skill?
When the prompt is really a procedure. OpenAI's developer docs describe skills as "folders of instructions and resources that teach ChatGPT and Codex how to complete repeatable workflows." Each has a SKILL.md file with a name, "a description that tells the model when to consider the skill," the steps, and optional references, scripts or templates. The model sees the name and description first and "loads the complete instructions when the user's request matches the skill," or when you invoke it directly.
Per OpenAI's help center, you find skills by opening Plugins in the sidebar and choosing the Skills tab, and eligible accounts start with a built-in skill-creator skill that helps you build or fix others. The same article says ChatGPT skills follow the Agent Skills open standard, so a skill can move between products, and that availability, installation and syncing can differ between products and surfaces such as desktop and web.
A prompt is a good skill candidate when:
- You paste it often, and it has several steps or a fixed output format.
- It depends on reference material, such as a template or a checklist.
- You want ChatGPT to pick it up on its own when a request matches.
Keep it a prompt when it is short, used now and then, or you want to tweak it each time. The trade-offs are covered in skills vs prompts, and how to write a SKILL.md covers the format.
What about custom GPTs?
Do not start new prompt collections in GPTs. OpenAI's retirement FAQ schedules custom GPTs to stop running on December 11, 2026, and encourages plugins as replacements. OpenAI's migration guide says "Migration turns your GPT's instructions into a skill and brings over its knowledge files and connected apps," while "Custom actions do not transfer."
If a GPT holds your best prompt, copy its instructions into a text file now. You keep a copy whatever happens to the migration, and you can reuse the text as project instructions, a skill or a prompt in your library. The custom GPTs guide covers the timeline and migration options.
Does ChatGPT memory save prompts?
Not reliably. OpenAI's docs say memories "let ChatGPT carry useful context from earlier chats into future work," and that they are "separate from required project guidance." Memory is good for facts such as your role or the stack you use. It does not store a prompt word for word for you to call up later. See ChatGPT memory for how to view and manage it.
How do you keep prompt templates with blanks outside ChatGPT?
For the prompts you pick from on demand, an outside library works better than any built-in option. The simplest durable format is one Markdown file per prompt, with blanks for the parts that change:
Summarize {{source:the pasted article}} for {{audience:a busy manager}}.
Lead with the one decision it asks for, then three bullet points.
Keep it under {{length:150}} words and flag any claim without a source.
Fill in the blanks, paste it into ChatGPT, and the same prompt works in Claude or Gemini too. To make the library worth keeping:
- One prompt per file, with a clear title and a few tags by job (summary, review, email) and by tool.
- Defaults in every blank, so most uses need no typing.
- Search that covers the prompt text, not only titles.
- Version history, so you can see what changed when results got worse. Git does this well, and prompt version control explains how.
- Move what always applies into ChatGPT: tone into custom instructions, project context into project instructions.
The AI prompt library guide compares notes apps, text expanders, Git and dedicated tools. Kipt is one option: it keeps prompts with {{name:default}} blanks as Markdown in a private GitHub repo you own, with tags and Cmd+K search, and it can store Agent Skills next to them. To turn a long prompt into a SKILL.md, the free SKILL.md generator runs in your browser with no sign-up.
How do you organize saved prompts so you find them again?
- Name prompts by job, not by date. "Meeting notes to action items" beats "prompt 14".
- Split always-on from on-demand. Always-on guidance goes into custom or project instructions. On-demand prompts go into your library.
- Prune monthly. Delete prompts you have not used, and merge near-duplicates.
- Store the output format in the prompt. A saved prompt that names the format, length and audience gives consistent results; one that only names the task does not.
- Keep a copy outside the app. Features change: GPTs are retiring and Skills plans may widen. A plain-text copy keeps your best prompts safe through every change.
If you also use Claude or Gemini, the same split applies there. See how to save prompts in Claude and how to save prompts in Gemini.