Preparation begins. Confirmed notice date.
IntelligenceMove the workflow, not just the bot.
Inventory one custom GPT, map what will and will not transfer, run an old-versus-new acceptance pack, and leave with a cutover decision.
Your inputs stay in this browser unless you export them or request a review.
New custom GPT creation is planned to end. Date may change.
Custom GPTs are scheduled to stop running. Confirm your workspace notice.
Name the workflow people will miss.
Start with business dependence, not the GPT’s title. A rarely used test bot and a daily operating tool should not share a migration queue.
Separate what transfers from what must be rebuilt.
OpenAI says instructions become a skill and connected apps are added as apps. Custom actions do not transfer. Prior chats, conversation starters, and the chosen model may not carry over as expected.
Make the replacement earn the switch.
Run every checked case in both the current GPT and replacement plugin. Save the input, output, result, model, tool use, tester, and date.
Check a case only after the replacement meets the written acceptance rule.
The migration decision.
This decision combines business dependence, dependency capture, access readiness, and side-by-side tests. It is an operating recommendation, not an OpenAI migration status.
Sources, method, and limits
- OpenAI Custom GPT retirement and migration FAQ, updated September 11, 2026. OpenAI says it plans to retire custom GPTs. For affected Enterprise workspaces, it lists a September 17 migration target, a planned September 25 end to new custom GPT creation, and a scheduled December 11 retirement. Dates may change and plan or workspace timelines can vary.
- The same FAQ says instructions become a skill and connected apps are added as apps. Custom actions do not transfer. Conversation starters and prior chats may not copy, the selected model does not carry over, a migrated plugin may respond differently, and the replacement starts private.
Method: Readiness gives 40% weight to eight dependency checks, 40% to twelve acceptance cases, 10% to named owners and fallback, and 10% to a dated cutover and structured evidence. Any custom action without the rebuild check, missing fallback, missing owner, or fewer than eight passed tests prevents a Ready to switch recommendation. These are Aule operating gates, not OpenAI requirements.