OpenAI says custom GPTs will be retired across all plans, with a migration path to plugins. No date yet. The reps who wrote the instructions need to find them first.
A lot of the sales-team AI that actually gets used lives inside a custom GPT someone built a long time ago: the objection-handling one, the "write it in our voice" one, the one that knows the pricing sheet. The reusable instructions inside those GPTs are the asset, and they are about to need a new home. The seller who copies out and owns those instructions now gets to choose where they run next. The seller who waits inherits whatever the migration tool produces, on a date set by a plan tier.
A custom GPT is a saved version of ChatGPT with standing instructions, optional uploaded files, and sometimes a connection to an outside service, built once and reused by whoever has the link. A plugin, in OpenAI's current usage, is the newer container for the same idea: reusable instructions plus connected apps, installed from a directory and called by name inside a conversation.
"Retire" means the GPT format stops working on a date OpenAI has not yet announced. "Migration path" means OpenAI intends to convert existing GPTs into plugins rather than leaving you to rebuild them, but that tooling does not exist yet. "Subject to your permissions" is the quiet clause: if the GPT you use was built by someone else, whether it survives migration depends on their access and their sharing settings, not yours.
The short version: the GPTs your team relies on will stop working on a date OpenAI has not set, and the instructions inside them are worth pulling out before that date arrives.
OpenAI announced on September 11, 2026, in the ChatGPT release notes, that it is "planning to retire custom GPTs across ChatGPT plans and provide a migration path to plugins, which can combine reusable instructions with connected apps." Existing GPTs keep working "until the retirement date that applies to you, subject to your permissions."
Announced: September 11, 2026, ChatGPT release notes.
Availability: No retirement date given. OpenAI states that "timing and availability may vary by plan and workspace" and points users to applicable notices.
Price: Not stated.
Regions: Not stated.
OpenAI's own preparation advice is short: "review the GPTs you rely on and identify their creators and anyone who needs access," and once migration is available, "test migrated plugins and review sharing."
What it doesn't do: The entry gives no date, no per-plan schedule, and no migration tool yet. Migration is something that "becomes available" later. It says nothing about GPTs shared publicly, about what happens to conversation history inside a GPT, or about which connected apps a migrated plugin can reach. A rep who reads this as "nothing changes today" is correct. A rep who reads it as "nothing needs doing" is not.
SDR — You are the heaviest GPT user on the team and probably the least likely to own one. List every GPT in your sidebar this week, open each, and copy the instructions and any uploaded files into a document you control. If a GPT was built by a rep who has since left, that is the one to rescue first, because "subject to your permissions" is aimed at you.
AE — Your GPTs tend to carry account-specific context: pricing rules, competitive positioning, the deal-desk policy. Treat this as a version-control moment. The migration tool will convert whatever is in the GPT on the day, including the stale instruction you meant to fix months ago. Clean the instructions before they are migrated, not after.
Manager — The team's GPT estate is undocumented, because nobody was asked to document it. Run the inventory OpenAI describes, which GPTs, who built them, who needs access, and turn it into a short list of the five that matter. Those five become plugins on purpose; the rest can lapse. This is also the moment to decide whether the team's standard prompts should live inside a vendor container at all.
RevOps — Two governance items. First, GPTs with connected apps have been reaching your systems under individual users' credentials, and migration to plugins will re-ask those permission questions; decide the answer before reps do. Second, the retirement date "may vary by plan and workspace," so a team spread across different ChatGPT plans or workspaces will lose GPTs on different days. Consolidate the seat mix or plan for a staggered outage.
The instructions inside a good GPT are a well-written prompt with a home. Keep the prompt as the source of truth and let the container change around it. Start from prompts built for AI tool operations and custom assistant configuration.
Two minutes, once a week. What changed in AI, and what to run because of it.