On this page
TL;DR:
- Yes, custom GPTs are being retired. OpenAI's documentation confirms it for Enterprise, and reports say all plans are affected. Those reports, citing OpenAI's retirement FAQ, give the date as 11 December 2026, which was 33 days away when we checked on 8 November 2026. Timing can vary by account.
- OpenAI's Enterprise guide says a retired GPT stops running and leaves the GPT directory. The replacement is the plugin, which can hold skills and apps.
- In that guide, instructions, knowledge files and connected apps can move across. Custom actions, your chosen model and your sharing settings do not.
- Back up first. The guide says migrating makes the original GPT read-only, so save its instructions and files, and test it, before you press the button.
- Sort your GPTs before you move them. Some deserve a plugin, some suit a Project, and some can simply end. Keep one plain-text master copy of each that survives, so the next platform change costs you an afternoon.
Who this guide is for
This guide is for anyone who built a custom GPT or depends on one: teachers, trainers, researchers, small business owners and team leaders. You do not need to code. One short section covers custom actions, and a developer may need to help with that part.
We think the first question is which of your GPTs should move at all, and in what form.
We checked every product detail below on 8 November 2026. This is moving quickly, so check your own account for any notice from OpenAI.
What is actually happening to custom GPTs?
OpenAI is ending custom GPTs and moving people to plugins. Its Enterprise migration guide refers to the period "before custom GPTs are retired" and states what happens next: "At retirement, custom GPTs stop running and leave the GPT directory." A separate OpenAI page on building plugins says custom GPTs in ChatGPT Enterprise are being retired.
Here are the dates as reported, as of 8 November 2026.
| Date | What happens | Who it applies to | Where the date comes from |
|---|---|---|---|
| 11 September 2026 | OpenAI announces the retirement | All | Virtualization Review; SVA Consulting |
| 26 October 2026 | Creating new GPTs due to end | Affected Enterprise workspaces | Virtualization Review; EdTech Innovation Hub |
| 11 December 2026 | Custom GPTs stop running | All plans, with timing that may vary by account | Virtualization Review; EdTech Innovation Hub; SVA Consulting |
| 11 February 2027 | Later retirement | Enterprise workspaces with an approved deferral | Virtualization Review; EdTech Innovation Hub |
What we confirmed, and what we could not
Much of the writing on this topic comes from companies selling a replacement, so here is exactly how far the evidence goes.
| Claim | How sure are we? |
|---|---|
| Custom GPTs are being retired and replaced by plugins | Confirmed for Enterprise on OpenAI's documentation site, which we read directly |
| At retirement they stop running and leave the directory | Confirmed for Enterprise on OpenAI's documentation site |
| What carries over and what does not | Confirmed for Enterprise in OpenAI's guide |
| The date of 11 December 2026 | Reported. Several write-ups give this date, and Virtualization Review and SVA Consulting attribute it to OpenAI's Help Centre. The OpenAI pages we could open give no date |
| It applies to Free, Plus and Pro accounts | Reported. OpenAI's guide that we read is written for Enterprise. Two reports say the retirement covers all plans |
| When personal accounts can migrate | Unclear. One vendor's review of OpenAI's pages says no date is published for personal plans. In November an OpenAI support reply on its forum said availability can vary by account and device |
OpenAI's Help Centre refused our automated reading tools, so we did not read it ourselves. Please read the Custom GPT retirement and migration FAQ in your own browser. We found no report of the reported 11 December date being moved for everyone, and Virtualization Review notes that timing may differ by account and workspace.
What changes when a GPT becomes a plugin?
A plugin is a package you install in ChatGPT. It can contain skills, which are reusable instructions for a type of work, and connections to outside tools. OpenAI says the same plugin directory serves ChatGPT and Codex.
OpenAI's Enterprise guide maps a GPT like this. On a forum thread about Plus accounts, OpenAI support gave the same answer for instructions, knowledge files and custom actions. Treat the other rows as a guide only on personal plans.
| In your custom GPT | After migration |
|---|---|
| Instructions | Become a skill inside the plugin |
| Knowledge files | Copied into the plugin as reference files |
| Connected apps | Added to the plugin as apps |
| Custom actions | Do not transfer. Rebuild them with a supported app or a custom MCP server |
| Selected model | Does not carry over. In Enterprise, workspace defaults apply |
| Sharing | Does not carry over. The plugin starts private, and each person must be given access |
| Conversation starters and past chats | May not copy, so OpenAI advises saving useful prompts |
| The original GPT | Becomes read-only, and stays usable until retirement |
The biggest change is how you start it. A GPT had its own chat, and its instructions applied to everything you typed there. OpenAI says an installed plugin will not run on every request. You pick a plugin or skill by typing @, or ChatGPT may choose a relevant installed skill for you. A GPT that felt dependable can therefore seem to ignore its instructions, simply because the skill was never selected.
OpenAI also warns that a migrated plugin may respond differently, and recommends retesting.
On cost, OpenAI's Enterprise guide says converting a GPT does not consume credits. Editing or updating a plugin does, under credit-based pricing.
Should every GPT be migrated?
No. Migrating is one of six sensible outcomes. Sort each GPT using this table.
| Your GPT | Best next step | Why |
|---|---|---|
| Used weekly, with clear instructions and a few files | Migrate to a plugin | Least rebuilding |
| Really a shared workspace for one ongoing area, where you go back and forth | Rebuild as a ChatGPT Project | Project instructions apply across its chats, and Sources hold your files |
| A procedure your team might run in another AI tool too | Keep a portable skill file, then migrate or upload it | One master copy works in several tools |
| Calls your own system through a custom action | Get a developer involved early | This part never migrates by itself |
| Public, with a link on your website or in course notes | Migrate, then plan new links | Old links are reported to fail, and sharing resets |
| No real use in the last 30 days | Back it up and let it end | Nothing is lost if the instructions are saved |
The 30-day test comes from OpenAI's Enterprise guide, where the admin report is ordered by recent use and a sample prompt looks at the last 30 days. Anyone can apply it by asking colleagues.
If a GPT you use was built by someone else, you cannot move it yourself. OpenAI's Enterprise guide says having access to another person's GPT is not enough: you must be its creator, or a workspace owner or admin. Write to the builder now, or copy down how you use it and rebuild something similar.
How to run a custom GPTs migration, step by step
Work in this order. Backing up and testing come before migrating, because the original stops being editable afterwards.
Step 1: List every GPT you rely on
Spend 30 minutes on a simple table. This one is invented.
| GPT | Owner | Who uses it | Used in last 30 days? | Files | Custom action? | Decision |
|---|---|---|---|---|---|---|
| Quote Writer | Sales lead | 4 staff | Yes, daily | 2 | No | Plugin |
| Stock Checker | Ops lead | 6 staff | Yes, weekly | 1 | Yes | Developer |
| Old Event Helper | Nobody | Unknown | No | 3 | No | Back up, let end |
Step 2: Back up before you touch anything
Open each GPT's editor and save these into one folder per GPT.
- Name, description and instructions. Copy them into a plain-text file with today's date.
- Knowledge files. Collect your original copies. The OpenAI documentation we could read describes no one-click export of a GPT, so do not rely on one. If you have lost an original, a migration copies the file into the plugin's reference files, which may be your way to keep it.
- Conversation starters. Copy them word for word.
- Custom action details. Note what each action does and save its schema. Keep passwords and API keys out of your notes.
- Your best results. Copy two or three chats you would hate to lose into a document.
On that last point, sources differ. One report says past chats stay with the original GPT until retirement, and another quotes OpenAI as saying they stay accessible afterwards. We could not open OpenAI's help article on data export, so we cannot say whether an export includes a GPT's setup.
Finally, publish any draft you need, if you still can. OpenAI's Enterprise guide says a GPT must be published to migrate, and one report says only the latest published version moves. One vendor says publishing is no longer available on personal accounts, so copy out any draft you cannot publish.
Step 3: Write a portable master copy
Turn the instructions into a clean text file that belongs to you and does not depend on any tool. The free prompt further down does this. The format to use is a skill: a folder holding a SKILL.md file, plus any reference files. Agent Skills is an open format, first developed by Anthropic, and OpenAI's documentation says its skills build on that standard.
Step 4: Record how the GPT behaves today
Choose five real tasks and one hard case, and save the answers. This is your baseline, and you can only create it while the GPT still runs.
One builder on OpenAI's forum scored a GPT against 100 written scenarios first. Six tests will do for most of us.
Step 5: Migrate
If your account has the option, go to My GPTs, find the GPT under "Created by me" and select "Migrate to plugin".
If you cannot see the option, finish Steps 1 to 4 and watch for a notice in your account. OpenAI also offers Plugin Creator: type @ in a chat, select it, describe the workflow and attach your files. OpenAI says this needs no code, but its steps assume plugin creation is enabled in your workspace. We could not confirm which personal plans have it, though one Plus user reported on the forum in September seeing an option to create skills and receiving the migration notice.
Step 6: Compare the plugin with your baseline
Run the same six tasks. OpenAI suggests checking that the right skill is selected, that reference material is used, that the hard case is handled, and that the tools you need are available. Check format and speed too. If something essential is missing, keep using the original until you have a fix. It works until retirement.
Then ask a colleague to try it cold, because the builder already knows which words make it work.
Step 7: Share it and fix every old link
Share the plugin with each person or group. In Enterprise, bulk migration does not copy sharing settings, and recipients must install the plugin unless an admin does it for them. Have a teammate confirm access.
Then hunt for old links. GPT pages are reported to become inaccessible at retirement, so check your website, course notes, printed handouts, QR codes, onboarding documents and pinned WhatsApp messages. Name one person who owns the plugin from now on.
A free prompt to make your GPT portable
Paste this into any AI tool. Remove client names, personal details, passwords and API keys first.
I am moving a custom GPT out of the GPT builder. Turn it into a
portable skill that does not depend on any one AI tool.
Work in three stages, and wait for my reply after Stage 1.
STAGE 1: AUDIT. Do not rewrite anything yet. Tell me:
a) What this GPT does, for whom, in three lines.
b) Which instructions only make sense inside the GPT builder
(for example "you are a GPT", or references to custom actions).
c) Which instructions contradict each other or are too vague to
test.
d) What depends on a connected app, custom action, web search,
image generation or a specific model.
e) Anything that looks like a secret or personal data that should
be removed.
f) All your questions, together in one numbered list.
STAGE 2: REWRITE. After I answer, write one file called SKILL.md:
- Start with YAML frontmatter holding two fields:
name: lowercase words joined by hyphens, under 64 characters.
description: under 200 characters, saying what the skill does
and exactly when to use it. Put trigger words first.
- Then these headings: Purpose, When to use, When not to use,
Inputs needed, Steps, Rules (each with its reason), Output
format, Reference files, Stop and ask a person when.
- Use only what is in my GPT and my answers. Invent no steps,
policies or facts. Mark anything uncertain [CHECK].
- List any feature you could not carry over in a separate section
called "Needs rebuilding".
STAGE 3: TEST. Write five realistic test requests and one hard
case that should make the skill stop and ask. For each, say what
a correct answer must include.
MY GPT'S NAME AND DESCRIPTION:
[paste]
MY GPT'S INSTRUCTIONS:
[paste]
LIST OF ITS KNOWLEDGE FILES AND WHAT EACH IS FOR:
[paste]
CAPABILITIES, APPS AND ACTIONS IT USES:
[paste]
Why those limits on the name and description? Tools differ. Claude's help pages cap names at 64 characters and descriptions at 200. Gemini asks for lowercase names with hyphens. Writing to the strictest rule gives you a file that uploads in more places.
A worked example (invented)
Example Logistics Ltd does not exist. Here is the kind of instruction block many GPTs start with.
You are QuoteBot, a super helpful GPT for our sales team. Always be
professional. Use the price list. Write great quotes. Never make
mistakes.
And here is the same GPT as a portable skill.
---
name: delivery-quote-writer
description: Drafts a delivery quote email. Use when asked to quote, price or cost a delivery for a customer.
---
# Purpose
Draft a delivery quote that sales can review and send the same day.
# When not to use
International freight. Pass those to the operations lead.
# Inputs needed
Pickup town, delivery town, weight in kg, delivery date.
If any is missing, ask before drafting.
# Steps
1. Find the route in references/price-list.pdf.
2. Calculate the price using only that list.
3. Draft a short email with the price, what it includes and how
long it is valid.
# Rules
- Take prices only from the price list, because a guessed price
becomes a promise to the customer.
# Stop and ask a person when
- The route is not in the price list.
- The customer asks for a discount.
"Never make mistakes" has gone, because nobody can test it. "Stop and ask" is new, so the skill knows when to hand back to a person. The description matters most: it helps decide when the skill gets chosen, which the GPT's dedicated chat used to settle for you.
Where else can the same instructions live?
A clean master copy works in other tools, and this is the moment to check whether you want it to. These details are correct as of 8 November 2026. Which tool suits you depends on your work, and we do not recommend one for everybody.
| Tool | Where your GPT's instructions could go | Worth knowing |
|---|---|---|
| ChatGPT plugin | Skill plus reference files | OpenAI's direct route. A plugin's skills work in Chat and Work on web, desktop and mobile. Check what your plan allows |
| ChatGPT Project | Project instructions plus Sources | Stays in ChatGPT with no plugin needed. A ChatGPT project has no direct access to folders on your computer, so upload or connect files |
| Claude | Custom skill, uploaded as a ZIP | Anthropic lists skills on Free, Pro, Max, Ultra, Team and Enterprise. Code execution must be on |
| Gemini | Skill, uploaded as a SKILL.md, folder or ZIP |
100 MB limit, and no old .doc, .xls or .ppt files. You must be 18 or over, signed in, with Keep Activity on |
| Microsoft Copilot | Copilot Notebook instructions plus reference files | Answers only from what you add. Microsoft lists .docx and .pdf among supported files, so save your master in one of those. Copilot and Copilot Chat licences qualify, with a SharePoint or OneDrive plan needed to create a notebook, and some personal Microsoft 365 plans too. See our Copilot training guide |
Moving to Gemini Gems is no escape. Google is replacing Gems with skills too. For personal accounts aged 18 and over, Gems become read-only on 17 November 2026 and are removed on 17 December 2026. Workspace business accounts follow in March 2027 and education accounts in June 2027.
So OpenAI and Google are both moving people from custom bots towards skills within a few months of each other. Our view is that the plain file you control is the part worth investing in.
Before you move a file to any new tool, check that your organisation has approved that tool for that kind of information.
If your GPT uses custom actions
Custom actions need the most planning, so begin with these this week. OpenAI says they do not transfer, and that you should replace them with a supported app's actions or a custom MCP server.
MCP (Model Context Protocol) is an open standard for connecting AI applications to outside systems. Building a server is developer work.
For each action, write down three things: what it reads, what it changes, and who would notice if it stopped. Before paying anyone to rebuild it, check whether a supported app already does the job.
Be strict about actions that write, send or delete. Have a person approve them until the new version has earned trust.
What goes wrong, and how to avoid it
| Problem | What you see | Fix |
|---|---|---|
| Migrated before backing up | The original is locked and you wanted to tidy it first | Save and improve your master copy first, then edit the plugin |
| Skill never fires | Answers ignore your instructions | Type @ and choose it, or rewrite the description with the words people really use |
| Model changed | Same prompt, different quality | Compare with your baseline, then adjust the instructions |
| Colleagues lose access | "The link doesn't work" | Share and install individually, then confirm with one teammate |
| The builder has left | Nobody can migrate it | Ask a workspace admin. On a personal account, you may only have what you wrote down |
One safety net appears only in OpenAI's Enterprise guide: a GPT can still be migrated after retirement, though it will no longer run. Do not assume this exists on other accounts.
What this means in Kenya
Three points deserve a look if you work in Kenya.
GPTs on personal accounts. If staff built useful GPTs on their own ChatGPT accounts, the business has no admin who can step in. Each person has to do their own backup. Ask this week, because the person may be on leave in December. If a freelancer built yours, ask them for the instructions and files now.
Personal data in knowledge files. A migration is a good time to look at what you uploaded years ago. Learner lists, payroll sheets, M-PESA statements or client contracts should not simply travel into another tool. The Data Protection Act, 2019 still matters when a file moves. Our guide to data protection and AI is written for public bodies, and its principles are useful for anyone. Remove what is not needed, and set the ground rules with an AI usage policy.
Links in learning materials. If you are a teacher, lecturer or trainer who shared a GPT by link in printed notes or a WhatsApp group, tell learners before December what replaces it. If the GPT supports students or researchers, keep your academic integrity rules in the new skill.
Your custom GPTs migration checklist
Copy this list. The dates assume the reported retirement of Friday 11 December 2026, and we have set them early on purpose.
By Friday 13 November
- [ ] Read the OpenAI FAQ and any notice in your account
- [ ] List every GPT, its owner and its users
- [ ] Mark each one: plugin, Project, skill file, developer, or let it end
- [ ] Contact anyone else who built a GPT you depend on
By Friday 20 November
- [ ] Save instructions, starters and action details as text
- [ ] Gather original knowledge files, with personal data removed
- [ ] Publish any draft you need, if you still can
- [ ] Run the free prompt and save each
SKILL.md - [ ] Record your six baseline tests
- [ ] Start any custom action rebuild
By Friday 27 November
- [ ] Migrate your most-used GPTs
- [ ] Repeat the six tests and compare results
- [ ] Have a colleague test each one cold
By Friday 4 December
- [ ] Share each plugin and confirm access
- [ ] Update links, handouts, QR codes and onboarding notes
- [ ] Name an owner for each plugin
- [ ] Check the OpenAI FAQ again for date changes
After retirement
- [ ] Confirm each replacement still works
- [ ] Store your master copies where the team can find them
- [ ] Add a version line each time a master copy changes
Questions people ask
Are custom GPTs really going away?
Yes. OpenAI's Enterprise documentation says custom GPTs are being retired and will stop running, and reports say all plans are affected. Reports citing OpenAI's Help Centre give 11 December 2026, with a later date for Enterprise workspaces that have an approved deferral. Check OpenAI's FAQ for any change.
Will my custom GPT convert automatically?
Do not assume so. The route OpenAI documents is a step you take in My GPTs, or an admin bulk process in Enterprise. If you do nothing, plan on the GPT stopping at retirement.
I am on Free or Plus. What should I do?
Do the backup now, whatever your plan. Reports say the retirement covers all plans, while migration access may arrive at different times for different accounts. Look for "Migrate to plugin" in My GPTs and for a notice in your account.
What happens to my old chats with a GPT?
Reports say they do not move into the new plugin. Reports differ on what remains accessible and for how long. Copy anything you need into a document before December.
Can I still create a new custom GPT?
Probably not. In affected Enterprise workspaces, creation was reported to end on 26 October 2026. One report describes ChatGPT offering a plugin when you try to make a GPT, and one vendor says creation is off on personal accounts. Unmigrated Enterprise GPTs were reported to remain editable after that cut-off.
Is a plugin the same as a Project?
No. A plugin packages a reusable workflow you install and can share. A Project is a workspace where chats share instructions and files. Use a plugin for a procedure, and a Project for an ongoing area of work.
Should I switch to Claude, Gemini or Copilot instead?
Only if the new tool fits your work and your organisation has approved it. A portable skill file means you can try one without starting over. Test the same six tasks in each and decide from the results.
Sources
OpenAI's own documentation and forum
- OpenAI, "Moving your custom GPT workflows to plugins", last checked 8 November 2026
- OpenAI, "Build plugins", last checked 8 November 2026
- OpenAI, "Plugins", last checked 8 November 2026
- OpenAI, plugin concepts, last checked 8 November 2026
- OpenAI, "Build skills", last checked 8 November 2026
- OpenAI, ChatGPT projects documentation, last checked 8 November 2026
- OpenAI Developer Community, migration path for personal Plus accounts, a user forum with one OpenAI support reply, last checked 8 November 2026
- OpenAI Developer Community, 100-scenario baseline, one member's own test, last checked 8 November 2026
Reporting on the retirement dates
- Virtualization Review, "OpenAI to Retire Custom GPTs, Replace Them with Plugins", 28 September 2026, last checked 8 November 2026
- EdTech Innovation Hub, "OpenAI to retire custom GPTs and move creators to plugins", last checked 8 November 2026
- SVA Consulting, "OpenAI Is Retiring Custom GPTs: What You Need to Know", 6 October 2026, last checked 8 November 2026
- GoMimic, "Custom GPTs Are Retiring", 5 November 2026, from a company that sells an AI content tool, last checked 8 November 2026
Portability and other tools
- Agent Skills, overview, last checked 8 November 2026
- Model Context Protocol, introduction, last checked 8 November 2026
- Claude Help Center, "What are skills?", last checked 8 November 2026
- Claude Help Center, "How to create custom skills", last checked 8 November 2026
- Google, "About the transition from Gems to skills", last checked 8 November 2026
- Google, "Create & manage skills for Gemini Apps", last checked 8 November 2026
- Microsoft Support, "Provide custom instructions for your Microsoft Copilot Notebook", last checked 8 November 2026
- Microsoft Support, "How Microsoft Copilot Notebooks works", last checked 8 November 2026
OpenAI's Help Centre refused our automated tools, so statements about its contents come from the reports above.
Where FuKazee fits in
FuKazee trains people in practical AI for the work in front of them. We have trained more than 3,000 people. We assume nobody in the room knows AI yet, avoid jargon, show the free version of each feature and do not push any single tool. Microsoft 365 and Copilot are part of the training, because so much office work happens there. Nobody becomes an expert in a few days, so we focus on habits people can keep using.
Training fades when it stays in the room, a problem we cover in why corporate AI training doesn't stick. We design sessions around a team's real tasks, and the GPTs your staff already rely on are a natural place to begin. See how this works on our page for AI training in Kenya.
To talk about moving your team's custom GPTs, or building AI habits that survive the next product change, email us at lead@fukazee.com.
Want guided support instead?
Not sure which programme fits? Tell us who you are and we will point you to the right path: corporate AI training, the ScholarShift Accelerator for postgraduates, or KCSE study plans and tuition.
Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.
