Distinguish announcement, migration and retirement
The 11 September announcement does not mean existing GPTs have already disappeared. They remain usable until their applicable retirement date, within existing permissions. In Enterprise guidance, 22 September was a target for migration and user notices, not a promise that every account would receive the feature that day. The account or workspace where a GPT was created also matters. A publicly used GPT can therefore follow a different transition schedule from its users’ accounts. Record the origin, owner, visible notice and proposed transition date for each important GPT before communicating a single deadline to the whole team.
Inventory the GPTs that people actually use
Look beyond formally approved assistants. In a small business, useful writing instructions often live in a personally created GPT shared with colleagues through a link. Record its purpose, recurring tasks, reference files and required connections. Assign a business owner and a deputy. For each GPT, save three typical requests and one difficult case with the expected outcome. This collection is more valuable for migration than a list of names. Prioritise GPTs whose disappearance would interrupt proposals, support or internal reporting. Rarely used experiments may not need rebuilding if another established workflow already serves their purpose.
What transfers and what needs separate review
The provider describes migration from the GPT’s latest published version: instructions become a skill, knowledge files become reference files and connected apps are included. Unpublished edits, the selected model configuration and existing conversations do not move across. Custom actions need a separate replacement, such as a suitable app or technical integration. A migrated personal plugin starts private, so existing users do not automatically receive access. The original GPT remains usable until retirement but becomes read-only after migration. Finish essential changes beforehand and record which version is the approved handover baseline for comparing the replacement.
Example: a proposal assistant connected to a CRM
Suppose a sales team uses a GPT to prepare proposals. Its writing rules and service descriptions can support the replacement, while a customer-number lookup implemented as a custom action needs separate attention. Split testing accordingly. First ask the plugin to create the required draft from fixed sample data. Then test the rebuilt CRM connection using authorised and unauthorised test identities. Pay particular attention to similar company names and to whether a missing record triggers a question. Keep the draft internal until customer matching and scope are correct. This is a proposed implementation example, not a claim about an existing customer result.
Organise the switch as a controlled handover
Have the owner and deputy test first, followed by an ordinary user without extra administrator permissions. If output format, references and permitted actions pass review, move a small group across. Share more than a new link: provide a short instruction covering the plugin, its purpose, required input and the contact for problems. Record any old connection that is no longer needed. Remove it only after acceptance and a dependency check. During the first week, track missing access, incorrect tool selection and additional rework. This produces a practical issue list instead of declaring the project complete merely because a plugin has been created.
Common GPT migration questions
Can any user migrate a GPT? Permission to use it is not ownership; creators or eligible administrators need the appropriate migration feature. Do previous conversations appear in the replacement plugin? No, arrange the handover of relevant business information explicitly. Will a custom action continue automatically? No, its business purpose and technical connection require a replacement and testing. Does the plugin need to become public? Public distribution is not the objective for an internal workflow; check the sharing actually needed. What if migration is not yet available? Prepare the inventory, examples and test specification now. Keep a documented manual process so changes in the provider’s timeline do not interrupt operations.
Keep it verifiable
Primary sources
- OpenAI: Custom GPT retirement and migration FAQSource checked:
- OpenAI: ChatGPT release notes, 11 SeptemberSource checked:




