For two and a half years, custom GPTs have been the easiest way to package know-how inside ChatGPT. Instructions, knowledge files, a few connected Actions, and in an afternoon you could have an assistant that wrote copy in your tone of voice, answered customer questions, or followed an internal procedure. Many consultants and agencies turned them into products, course bonuses, or components of a service. Now OpenAI has decided to close that chapter.
The news is official and comes from the OpenAI Help Center: the company plans to retire custom GPTs and recommends replacing them with plugins, which combine reusable instructions with connected apps. This isn’t a rumor, and it isn’t exactly coming out of nowhere either. On April 22, 2026, OpenAI had already launched Workspace Agents, Codex-powered agents that work inside ChatGPT, connect to tools such as Slack, Salesforce, and Gmail, and execute multi-step workflows. From that point on, it was clear where investment was heading, while GPTs had been left in a sort of routine maintenance mode.
The Custom GPT Retirement Dates
The published timeline currently applies to Enterprise workspaces. On September 11, administrators were notified, September 17 is the target date for the migration and user-facing banner, September 25 is when the creation of new GPTs is expected to close, and December 11, 2026, is the scheduled retirement date, meaning the day GPTs will stop working. OpenAI notes that these dates may change and that the September 17 migration is a target, not a guarantee for every workspace.
What about personal accounts? Here’s the detail many articles are overlooking. For personal accounts, meaning Free, Go, Plus, and Pro, creating and publishing new GPTs is already no longer available, while existing ones remain usable. As for the shutdown date, the transition applies to all plans, and the others may follow the same Enterprise timeline, with announcements appearing inside the product. In other words, December 11 is not officially the date for everyone, but it would be naive to plan as though it weren’t.
What Happens When You Migrate to Plugins
OpenAI promises a tool for moving GPTs to plugins, so in theory you won’t have to start from scratch. There is, however, one constraint you should know before pressing the button: once migrated, the original GPT remains usable until retirement but becomes read-only, so any essential changes need to be made beforehand. After that, only the plugin is maintained. Anyone who still has drafts in progress should publish them before GPT creation closes, otherwise they risk being unable to take them along.
From a technical standpoint, the change makes sense. A GPT was a separate chatbot, a place you went to. A plugin is designed to work inside conversations and tools, bringing instructions and connections with it. For people who used GPTs as operational procedures, this is probably a step forward. But the more interesting question isn’t technical.
Was the GPT an Asset or Someone Else’s Feature?
If you work in marketing and have sold “custom AI assistants” to clients, now is the time to take an honest look at what you actually delivered. A GPT lives inside ChatGPT, can only be used with a ChatGPT account, and follows OpenAI’s rules and timelines. The client paid you, but the infrastructure supporting that value belonged to someone else, and that someone else has now decided to change it with three months’ notice. It’s the same dynamic we’ve seen countless times with platforms: a Facebook page losing organic reach, an algorithm changing, a service shutting down. The technology changes, but the lesson doesn’t.
So what is actually yours? The instructions you wrote, the knowledge you structured, the process you designed. That’s the capital. The container is replaceable, and if you designed the content well, you can move it into a plugin, an agent, another model, or your own automation without rebuilding the reasoning from scratch. If, on the other hand, only the GPT knew the instructions behind your GPT, the problem isn’t OpenAI.
What to Do Now
The first step is to take inventory: which GPTs do you actually use, which ones have you delivered to clients, which ones are public, and which ones are tied to your name. For each one, save the instructions, knowledge files, and Actions configuration in an external document today, not tomorrow when the banner appears. Then, once migration becomes available on your account, make any final changes before migrating, move the GPT over, and test the plugin using the same use cases as the original GPT, because an architectural change rarely produces identical results on the first try.
If those GPTs are part of a commercial offering, let your clients know before they discover the change themselves. It’s also an opportunity to ask whether it makes sense to keep building value inside a single ecosystem or whether it would be better to keep your know-how in a portable format that you can run wherever it’s needed.
Custom GPTs are being retired, but personalization isn’t. It is simply moving, as always happens when a platform grows. The difference comes down to whether you understood from the beginning that the value wasn’t in the GPT itself, but in what you put inside it.
