OpenAI Is Retiring Custom GPTs. Marketing Teams Need an Inventory.

OpenAI says it plans to retire custom GPTs and migrate eligible workflows into plugins. For affected Enterprise workspaces, the scheduled retirement date is December 11. That gives marketing teams less than three months to find the small assistants that have quietly become part of their work.

The deadline is what matters. A marketing workflow that’s invisible today becomes urgent when it stops running.

OpenAI’s Enterprise timeline calls September 25 the planned cutoff for creating new custom GPTs. Existing GPTs are scheduled to keep working until retirement. The migration experience was targeted for September 17, but OpenAI says availability can vary by plan and workspace, and the dates remain subject to change.

Custom GPTs tend to disappear into marketing teams precisely because they are easy to make. Someone builds a brand-voice reviewer, a product-launch brief helper, a sales-enablement assistant, a campaign QA bot, or a way to turn a founder’s notes into a first draft. The person who made it may not be the person still using it. The instructions, reference files, and app connections may be documented nowhere else.

That makes this an operations story, not a feature-announcement story.

A migration is not a copy and paste

OpenAI’s planned migration moves a GPT’s instructions into a skill inside a new plugin and brings connected apps across as apps. That’s a useful starting point. It doesn’t promise that the replacement will behave the same way.

The FAQ says reference files, templates, examples, and tools need review before a team relies on the new plugin. Conversation starters and previous chats may not copy. The selected model does not carry over either. In an Enterprise workspace, the replacement follows the workspace defaults.

The sharpest edge is custom actions. OpenAI says they do not transfer through the migration workflow. A GPT that calls an external API or triggers a workflow needs that integration assessed and rebuilt through a supported connector or custom MCP server. The new version also needs its own permissions review and testing.

This calls for an orderly review, not panic. Treating a migration button as a finished project would leave important work unchecked.

OpenAI’s existing GPT documentation makes the practical distinction clear. Instructions are the rules for how a GPT behaves. Knowledge files are the source material it can draw from. Actions connect the GPT to external APIs. Those are three different things to inventory, test, and assign an owner to.

Find the hidden marketing workflows first

Start with a simple list of every custom GPT the marketing, content, growth, product marketing, and sales-enablement teams use. Include the owner, users, business job, link, instructions, source files, connected apps, custom actions, and a few prompts that show what a good answer looks like.

That last field matters. A migration can preserve the text of an instruction and still change the result because the selected model, available tools, or account context changed. A small test set turns a vague question, “Does it still work?”, into a check a marketer can run: does the brand reviewer catch the same unsupported claim, does the campaign helper use the right source file, does the sales assistant produce the required format?

The access model needs the same attention. OpenAI says a migrated plugin starts private, and people who used the original GPT don’t automatically get access to its replacement. For a shared marketing workflow, the team needs to test both the output and whether the right colleagues can install and use the replacement.

This is a chance to retire the bad ones too

Some teams will find custom GPTs that shouldn’t be migrated. A helper with stale files, vague instructions, no owner, or an unclear business job creates a maintenance obligation.

That pruning is useful. The best internal marketing assistants are usually narrow enough to test. They have clear instructions, current source material, a named owner, a known place in a workflow, and limits on what they are allowed to claim or do. The rest are often prompts with a title.

OpenAI listed the planned retirement in its September 11 release notes, then pointed teams to test migrated plugins and review sharing before switching. The migration will be a forcing function for that distinction. Teams that start with an inventory can keep the workflows that save time and remove the rest deliberately. Teams that wait for a broken link will be rebuilding under pressure.

The immediate move is simple: make the list, preserve the source material, record the test prompts, and give every workflow an owner. The migration path may make the technical move easier. The marketing work is deciding which assistants deserve to survive it.

The same discipline applies to the marketing-team knowledge problem: an AI assistant only stays useful when its source material, rules, and owner stay current. And when a workflow needs to take action outside ChatGPT, the WhatsApp Business MCP rollout shows the adjacent task: permissions and the technical connection deserve their own review.