Workbranch › Guides
GuideHow to create a custom GPT
Custom GPTs are changing fast: personal accounts can no longer make new ones, and OpenAI plans to retire them in favor of plugins. Here's who can still build a GPT, how to do it, and what to do if you can't.
Quick answer: as of October 2026, only people in a ChatGPT Business, Enterprise or Edu workspace whose settings allow it can create a new GPT (in affected Enterprise workspaces, only until the planned 26 October 2026 cutoff). On the web, open Explore GPTs (or chatgpt.com/gpts), select Create, fill in the builder, test it in Preview and select Create. Free, Go, Plus and Pro accounts can't make new GPTs, and OpenAI plans to retire custom GPTs and move them to plugins.
These steps were checked against OpenAI's help pages on 18 October 2026. This area is changing quickly, so check the notices in your own account too.
First: can you create one?
| Account | New GPTs | Existing GPTs |
|---|---|---|
| Free, Go, Plus, Pro | Can't create or publish | Can use them, and edit ones you made if your plan and permissions still allow it |
| Business, Enterprise, Edu | Can create, edit and publish if workspace settings and your role allow it | Same, within workspace rules |
OpenAI says plainly that this is an eligibility rule, not a bug, so no setting on a personal account will bring back the Create button. Building also only works on the web: the mobile apps let you use GPTs but not make them.
The retirement timeline
OpenAI plans to retire custom GPTs and recommends moving reusable workflows to plugins, which can bundle instructions, reference files and connected apps. The help center lists these dates, all marked as subject to change:
- 26 October 2026 (planned): creation of new custom GPTs ends in affected Enterprise workspaces. Editing existing, unmigrated GPTs is still allowed.
- 11 December 2026: the scheduled retirement date. After it, custom GPTs and their pages become inaccessible. Your past conversations with GPTs stay available.
- 11 February 2027: only for Enterprise workspaces with an approved deferral.
Other plans are expected to follow the same timeline, so watch for the notice on your account or workspace. If you're starting something new and long-lived, a plugin or a project is the safer home.
Step by step: build a GPT in a workspace
- On chatgpt.com, open Explore GPTs in the sidebar, or go straight to chatgpt.com/gpts.
- Select Create to open the GPT builder.
- Pick a way to build. In the conversational builder you describe what you want and ChatGPT drafts it. In the configuration view you edit each field yourself. You can draft by chatting, then tidy up the fields directly.
- Fill in the Name, a short Description (what it's for and who it helps) and a few Conversation starters that show people what to ask.
- Write the Instructions (tips below).
- Upload reference files under Knowledge.
- Optionally set a Recommended model. Users can still switch to another model they have.
- Under Capabilities, turn on what the GPT needs: Web search, Image generation, Canvas, Code Interpreter & Data Analysis or Apps. Developers can add Actions that call external APIs, but a GPT can use apps or actions, not both.
- Try it in Preview with real questions. Changes save to a draft as you go.
- Select Create when you're happy with it.
Writing instructions that work
Instructions apply to every conversation and control tone, structure and what to avoid. OpenAI's own advice boils down to a few habits:
- Spell out the steps. For a multi-step job, write it as "when this happens, do that", and separate sections clearly with headings or lists.
- Say what to do. Concrete positive rules work better than a long list of don'ts.
- Show examples. If the GPT must classify or follow definitions, add a short example of a good and a bad output.
- Keep rules out of files. Behavior belongs in instructions; knowledge files are reference material.
For example: "You help our support team answer refund questions. First check the refund policy in your knowledge files. Answer in three short paragraphs, quote the policy section you used, and suggest escalation if the order is over 60 days old."
Knowledge files
Knowledge works best for documentation, handbooks and guides. Each file can be up to 512 MB. OpenAI's pages currently give two different file counts, 20 files per GPT in the builder article and 10 knowledge files over a GPT's lifetime in the uploads article, so plan for the lower number until the builder tells you otherwise. Some file types only work when Code Interpreter & Data Analysis is on.
Prefer clean, text-heavy files. Complex layouts and images inside PDFs are harder for the GPT to use, so describe important visuals in text. If you want citations or quotes from the files, ask for them in the instructions.
Sharing settings
Open the GPT in the editor, select Share, choose who can access it and what they can do, then select Save.
- Who: Invite-only (specific people or groups), Anyone at your workspace with the link, Your workspace, Anyone with the link (where allowed) or the GPT Store.
- What they can do: Can chat, Can view settings (use, duplicate and see the configuration) or Can edit. Public links and the GPT Store only allow Can chat.
- Direct sharing reaches up to 100 recipients, where a group counts as one.
To edit later, go to My GPTs, pick the GPT, select Edit GPT, then Update. The ••• menu in the editor has version history and delete.
Good to know
- Training: Business, Enterprise and Edu data isn't used for training by default. On personal plans it depends on your data controls.
- Apps and actions send data out. Parts of a conversation can go to the connected service, and OpenAI doesn't control how that service uses it.
- Migrating to a plugin: when it's offered, use Migrate to plugin in My GPTs. Only the latest published version moves, custom actions and sharing settings don't carry over, and the original GPT becomes read-only.
If it doesn't work
- No Create button. You're on a personal account, or your workspace doesn't let your role create GPTs. Ask your admin, or use a ChatGPT project with instructions instead.
- The GPT ignores its instructions. Shorten them, remove conflicting rules, add a clear example and retest small changes in Preview.
- It doesn't use the files well. Check the answer is really in them, simplify complex PDFs into plain text and ask narrower questions.
- Sharing options are grayed out. Your workspace may block that level, or the GPT uses apps, or public actions lack a Privacy Policy URL.
Sources
- Creating and editing GPTs (OpenAI Help Center, checked 18 October 2026)
- GPTs in ChatGPT (OpenAI Help Center, checked 18 October 2026)
- Sharing and publishing GPTs (OpenAI Help Center, checked 18 October 2026)
- Custom GPT retirement and migration FAQ (OpenAI Help Center, checked 18 October 2026)
- Troubleshooting GPTs (OpenAI Help Center, checked 18 October 2026)
- Uploading files and audio to ChatGPT (OpenAI Help Center, checked 18 October 2026)
Frequently asked questions
Can I still use GPTs other people made?
Yes. Any signed-in ChatGPT user can chat with GPTs they have access to, whether public, shared by link or shared in a workspace, until the planned retirement date.
Do GPTs remember my custom instructions or memories?
No. A GPT ignores your saved memory, your custom instructions and your earlier chats, so every conversation with it starts from scratch.
Can the person who built a GPT read my chats with it?
No. OpenAI says GPT builders can't see individual conversations people have with their GPTs.
Can I put my GPT on my own website?
No. GPTs only run inside ChatGPT. To build an assistant into your own site or product, OpenAI points you to its API instead.