Custom GPTs are retiring. Here's what that means for you, and what to do about it.

If you've spent the last couple of years building Custom GPTs, there's a change coming that's worth understanding properly, both for your own tools and for anything your team has quietly built into the way they work.

Existing Custom GPTs are currently scheduled to keep working until 11 December 2026, though some Enterprise organisations have an approved deferral that extends that date. New Custom GPTs can be created until 26 October 2026. So there's time to act, but if you have GPTs you genuinely rely on, I wouldn't leave it until December to think about this.

The more interesting question isn't the deadline. It's what's replacing them, and what that tells us about where AI is heading.

What's actually changing

Custom GPTs gave us a wonderfully simple way to create a specialised version of ChatGPT. You could give it detailed instructions, provide documents for it to reference, and tell it how you wanted it to behave, which meant you didn't have to start every conversation by explaining everything again.

Imagine you created a Custom GPT called Frank the Football Pundit. You told Frank he was an expert on English football, that he should explain matches in the style of an experienced pundit, that he should understand formations and tactics, and that he should always use plain English rather than drowning you in statistics. You gave him some reference material and detailed instructions about how you wanted him to analyse a match. Whenever you wanted that particular expertise, you opened Frank.

Plugins change that model. Rather than having to go somewhere else and start a conversation with a separate AI personality, the useful capability becomes something ChatGPT has available while you're already doing other things. So instead of thinking "I need to go and talk to Frank," you start thinking "I need football analysis for this task." It sounds like a small distinction, but I think it's actually quite significant.

So what is a plugin?

The terminology can get unnecessarily confusing, particularly because we now have plugins, skills and apps appearing inside ChatGPT. The easiest way I've found to think about it is this: a skill teaches ChatGPT how you want a particular job done, an app gives ChatGPT access to another service or source of information, and a plugin can bring those things together, along with reusable instructions and reference material, into a capability that ChatGPT can draw on when it's relevant.

When OpenAI migrates a Custom GPT, the GPT's instructions become a skill within the plugin, and its knowledge files become reference files. That means plugins aren't simply Custom GPTs with a new badge on them. They're a different way of thinking about how we give AI specialist capabilities, and in some situations ChatGPT may even select an installed skill automatically because it recognises it's relevant to what you're working on.

What happens to the GPTs you've already created

OpenAI is providing a migration process. When it becomes available for your account, you'll be able to go to your GPTs and select Migrate to plugin. But I wouldn't treat this as an exercise in blindly migrating everything you've ever created.

If you're anything like me, you've probably created GPTs because you had an idea, wanted to experiment, or wondered whether you could make ChatGPT do a particular job. Some became genuinely useful. Others were interesting for about 20 minutes and haven't been opened since. There's very little point carefully migrating those.

Instead, use this as an opportunity for some AI housekeeping. Look through the GPTs you've created and identify the ones you genuinely use. Then ask yourself a more interesting question: what is it about this GPT that actually makes it valuable? It might be a carefully written set of instructions, documents or reference material you've built up over time, a methodology you've embedded within it, or a connection to another system that performs a useful business process. That's the stuff worth protecting. The GPT itself isn't really the intellectual property. The thinking you put into it is.

Before you migrate anything, do this

Keep a handful of prompts that represent how you normally use each important GPT, perhaps five or ten examples. Then, when you've migrated it, ask the new plugin exactly the same questions and compare what comes back. This matters because OpenAI specifically warns that a migrated plugin may behave differently from the original GPT. You need to check that it follows the instructions properly, uses the right reference information, and still produces the output you expect. Include at least one deliberately difficult prompt in your testing, not just the straightforward cases, but something ambiguous or complicated, to find out how it handles the edges.

A couple of other things worth knowing: your existing conversations with a GPT don't migrate across to the new plugin, and the particular model you selected for your GPT doesn't transfer either. Most importantly, if you've built a GPT that uses Custom Actions to connect to another system, those actions don't automatically migrate. You may need to recreate that integration using an available app or, for more specialised integrations, a custom MCP server. If you don't know what that means, you probably don't need to worry about it. If you did build a GPT using Custom Actions, it's something worth investigating sooner rather than later.

What if you use somebody else's GPT?

There's a slightly different issue if you've become reliant on a GPT that somebody else created. You can't migrate it yourself simply because you use it. The creator will need to migrate it and decide how the replacement plugin is made available, and access to the original GPT doesn't automatically give you access to its replacement. So take a look through any third-party GPTs that have become part of your normal workflow. If there's something you genuinely depend on, find out what its creator is planning to do with it. You don't want to discover on 11 December that something you've quietly built into your working life has disappeared.

There's a bigger issue for businesses

This change creates a useful moment for organisations to look at something that has largely happened under the radar. Over the last few years, employees have created large numbers of small AI tools. Some are experiments, some are essentially sophisticated prompts, some contain useful company information, and some have quietly become part of the way people actually do their jobs. The problem is that many organisations probably don't know they exist.

If a member of your team has created a GPT that saves them three hours every week, that's interesting. If five people now depend on it, that's more interesting. If it contains company knowledge, connects to another system, and nobody except its creator really understands how it works, you've got a governance issue. So use this migration as an excuse to ask a few straightforward questions: what have your people built, which GPTs are genuinely being used, which contain valuable company knowledge or methodologies, who owns them, who maintains them, and who should have access to them? That might sound rather grand for something somebody originally knocked together one afternoon, but that's exactly how technology adoption often happens. Experiments gradually become processes.

What I'd do now

I wouldn't spend the next week frantically migrating everything. I would start with a simple audit during October: look at the GPTs you actually care about, make sure their instructions are up to date, check the reference material, and identify any that use Custom Actions. If you have unpublished changes that matter, deal with them before migration, because OpenAI says the migration uses the latest published version. Save those test prompts I mentioned. Once migration is available for your account, migrate the important GPTs and test them properly rather than assuming that because the migration completed successfully the new plugin behaves exactly as the GPT did.

If other people use something you've created, check the sharing arrangements too. A migrated plugin starts private, and the people who could access the GPT don't automatically get access to its replacement. Do all of that during October and November and, hopefully, 11 December becomes uneventful. Which is exactly what you want from any technology migration.

I think this is actually a good change

Custom GPTs gave ordinary users a simple way to move beyond generic AI. Instead of everyone having exactly the same ChatGPT, we could give it specialist knowledge, instructions and ways of working, and start capturing some of our own expertise inside AI. Plugins take that idea further. Rather than an ever-growing collection of separate AI personalities that we have to remember to visit, we're moving towards giving our AI a collection of capabilities that can be drawn on when they're needed.

For the last few years, the question has often been "which AI should I use for this?" I suspect we're going to ask that less and less. The more interesting question will become "what does my AI need to know, and what does it need to be able to do?" That's quite a different way of thinking, and if you've already invested time creating Custom GPTs, that work doesn't need to be wasted. The important thing was never the GPT itself. It was the knowledge, instructions, processes and thinking that you put into it. Custom GPTs may be disappearing. The useful thinking inside them doesn't have to.

---

Frequently asked questions

What is the deadline for the phase-out of Custom GPTs?

New Custom GPTs can be created until 26 October 2026, while existing Custom GPTs are currently scheduled to keep working until 11 December 2026. However, some Enterprise organisations have an approved deferral that extends this date. It is best to audit and migrate your essential tools well before the final December deadline.

What are Custom GPTs being replaced with in ChatGPT?

Custom GPTs are being replaced by plugins. Instead of opening a separate AI personality, plugins add capabilities directly to ChatGPT. When migrated, a GPT's detailed instructions become a skill within the plugin, and its uploaded knowledge files become reference files. ChatGPT can then draw on these capabilities automatically when relevant.

Will Custom Actions and existing conversations automatically migrate to plugins?

No, existing conversations with a Custom GPT do not transfer over to the new plugin, and chosen AI models do not migrate either. Furthermore, Custom Actions connecting to external systems will not automatically migrate. You may need to recreate those integrations using an available app or a custom MCP server.

Can I migrate a Custom GPT created by someone else?

No, you cannot migrate a third-party Custom GPT yourself. The original creator must perform the migration and decide how to share the replacement plugin. Because access to the original GPT does not transfer automatically, and migrated plugins start as private, you should contact the creator if you rely on their tool.

How should I prepare my Custom GPTs before migrating them?

Audit your GPTs to keep only the valuable ones, ensure all instructions and files are up to date, and publish any unpublished changes. You should also record five to ten test prompts, including complex edge cases, to compare responses from the original GPT against the new plugin to ensure it behaves correctly.