Custom GPT Retirement Advice: Don't Just Convert to Plugins

If you are looking for Custom GPT retirement advice, the first thing I would say is this: do not leave the review until the last minute. I expected Custom GPTs to disappear eventually, but the speed of the change has surprised me.

OpenAI originally planned to stop the creation of new GPTs on 25 September, but that has now been pushed back to 26 October. OpenAI officially announced the planned retirement of Custom GPTs on 11 September 2026, and the change affects all ChatGPT plans. For Enterprise users, the clearest timetable currently shows migration beginning from 22 September, new GPT creation ending on 26 October and existing Custom GPTs being retired on 11 December 2026.

That is not very long when you consider how many businesses have spent months, and sometimes years, building GPTs into the way they work. For some people these are useful shortcuts. For others they have become genuine business tools containing detailed instructions, company knowledge, reference documents and connections into other systems.

So if you have a collection of Custom GPTs sitting in your account, I would work through them now and decide what should happen next.

The important Custom GPT retirement dates

• 11 September 2026: OpenAI announced the planned retirement of Custom GPTs.

• 22 September 2026: Target date for the GPT-to-Plugin migration experience to start appearing.

• 26 October 2026: Planned date when the creation of new Custom GPTs ends in affected Enterprise workspaces.

• 11 December 2026: Custom GPTs are currently scheduled to retire and stop running.

My Custom GPT retirement advice is not to wait until December. If you have only one or two simple GPTs, this may not be a huge job. If you have built a whole collection, especially ones used regularly by your team, there is quite a bit to think about.

Why are GPTs being moved to Plugins?

This is the part of the announcement that initially did not quite fit for me. I expected GPTs to disappear eventually, but I assumed the natural destination would be Projects, Skills or Agents rather than Plugins.

Having looked more closely at how OpenAI is structuring the new system, it starts to make more sense. A Plugin is not simply another version of a Custom GPT. It is better thought of as a container for capabilities, which can include Skills, connected Apps or a combination of the two.

The Skill can provide the instructions and workflow, while the App provides access to external information or systems. When OpenAI migrates a Custom GPT, the GPT's instructions are turned into a Skill inside the Plugin. Connected Apps can then sit alongside that Skill and knowledge files can also be brought across as reference material.

So we are not really choosing between a Plugin or a Skill. In many cases the Skill may sit inside the Plugin, which makes the migration strategy easier to understand.

Should every GPT become a Plugin?

No. Just because OpenAI gives you a Migrate to Plugin option does not mean every GPT you have created needs to continue existing in exactly the same form.

I think this is a good opportunity to tidy things up. Most of us have probably built GPTs that made complete sense at the time but would now be better built using one of the newer ChatGPT features. Rather than simply migrating everything, work through your GPTs one at a time and ask what each one is actually doing.

• Is this mainly giving ChatGPT access to a set of knowledge and files?

• Is it following a structured and repeatable process?

• Does it need to connect to other tools or systems?

• Is it trying to manage a bigger workflow involving actions, triggers or multiple stages?

  
  

The answer will normally give you a good idea of where that GPT should go next.

Knowledge-based GPT? Consider a Project

If you have built a GPT containing a collection of files and some instructions, perhaps so you can reference company information, research material, course content, policies or internal documents, there is a good chance this should actually be a Project.

Projects are designed to keep related chats, files and instructions together in a shared context, which makes them useful for ongoing work where the main requirement is access to a particular body of information.

Quite a few knowledge-based GPTs probably should have become Projects when Projects became a major feature. We often kept using the GPT because it was already there and it worked. The retirement of GPTs now forces that decision.

Repeatable process? Consider a Skill

The next question is whether the GPT exists because you want ChatGPT to follow a particular process.

Perhaps it takes a customer brief and turns it into a proposal, analyses a spreadsheet in a particular way, converts meeting notes into a structured report or follows the same sequence of checks every time you carry out a task. That is where Skills make much more sense.

If your requirement is essentially, 'Whenever I do this job, follow these steps and produce the result in this way', you are probably looking at a Skill rather than a standalone GPT.

Custom GPTs became a catch-all solution for lots of different things. Skills give us a clearer way of defining a repeatable process that ChatGPT can use when it is relevant.

Need a process plus connected tools? That is where Plugins fit

Plugins become more useful when you have a repeatable process but also need ChatGPT to connect with other systems.

You might have a Skill explaining how a particular task should be completed while connected Apps provide access to Google Drive, Slack, company data or another business system. The Plugin then packages those capabilities together.

Seen this way, Plugins are less a direct replacement for Custom GPTs and more of a framework that brings together the different components we previously tried to squeeze inside a GPT.

A more accurate way to think about it is: the instructions inside my GPT are becoming a Skill, and that Skill can sit inside a Plugin alongside any connections it needs.

Bigger process with actions and triggers? Consider an Agent

Then you have the GPTs that were probably trying to be Agents before Agents properly existed. These tend to be the more ambitious ones where you are asking ChatGPT to manage a bigger process, perhaps looking across several systems, carrying out multiple stages of work, taking actions or reacting when something happens.

That is where I would start looking at Agents. A Skill normally helps ChatGPT perform a particular task in a structured way, while an Agent can have responsibility for a wider workflow involving several tasks, tools and actions.

For example, a Skill might explain how to research and qualify a sales prospect. An Agent might monitor for new prospects, carry out the research, update your CRM, prepare an email and trigger the next stage of the process.

This distinction between Skills for repeatable tasks and Agents for larger workflows is going to become increasingly important.

Be careful if your GPT uses Custom Actions

This is one of the biggest things to watch during migration. If your GPT uses Custom Actions, do not assume the migration process will move everything across because OpenAI has confirmed that Custom Actions do not automatically transfer.

If your GPT connects to another service through an Action, you will need to look at whether an existing App can replace it. If it cannot, some rebuilding may be required using the newer integration options.

These are the GPTs I would review first because they are more likely to require work than a simple knowledge-based GPT containing a few files and instructions.

Migration does not copy everything

OpenAI says the migration process does not create an identical copy of your existing GPT, so you should expect to check and test the new version afterwards.

• Existing conversations do not move across.

• Your chosen model is not transferred.

• Sharing permissions are not automatically copied.

• The newly migrated Plugin starts off private.

• Draft changes to your GPT will not transfer unless they are published.

• Custom Actions need separate attention.

Once migrated, your original GPT can still be used until retirement but becomes read-only. Make sure important changes have been published before migrating, then test the new version using both familiar prompts and more difficult examples.

I was actually teaching GPTs when this was announced

There was some fairly good timing involved in this announcement for me because on the day it landed I was actually teaching Custom GPTs in a class. So my GPT section suddenly became part training session and part history lesson.

I have quite a few example GPTs that I normally show people, so rather than abandoning the session we started looking at each one and asking a different question: where would we build this today?

It turned into a useful exercise. Some clearly belonged in Projects, some made more sense as Skills, others would naturally become Plugins because they needed a combination of instructions and connected tools, while a few of the more complicated examples were really early versions of what we would now describe as Agents.

Do not just migrate everything. Review everything

I expected Custom GPTs to disappear eventually, but I did not expect the timescales to be quite this aggressive. I also do not think we should look at this purely as OpenAI taking something away.

Custom GPTs gave millions of people a simple introduction to the idea of customising AI. They taught us that we could give AI detailed instructions, provide our own knowledge, define processes and build useful tools around particular jobs.

The next generation of ChatGPT is becoming much more modular:

• Projects hold ongoing context, chats and knowledge.

• Skills define structured and repeatable processes.

• Apps connect ChatGPT with external systems and information.

• Plugins package Skills and connected Apps together.

• Agents handle larger workflows involving multiple stages, actions and triggers.

When you look at it like that, the structure is clearer than trying to make Custom GPTs do everything. It is just going to take people a little while to adjust.

So if you currently rely on Custom GPTs, my advice is simple: start reviewing them now rather than simply migrating everything at the last minute. Work through them one at a time, understand what each one actually does and decide where that capability belongs in the new ChatGPT setup.

Custom GPTs may be disappearing, but the useful workflows you created with them do not have to.