Do not index
Do not index
OpenAI announced on September 11, 2026 that it will retire custom GPTs and migrate users to a new "Plugins" system. Custom GPTs, the ChatGPT feature that let users build a custom assistant with instructions and uploaded knowledge files, will stop running entirely for Enterprise workspaces on December 11, 2026.
Enterprise users have already received admin notices and are seeing active migration tools, with the ability to create new custom GPTs ending around September 25-26. Personal accounts (Free, Go, Plus, Pro) have already lost the ability to create new custom GPTs, though existing ones remain functional for now. OpenAI states these accounts will follow "a similar transition timeline" but has not published a firm retirement date for personal users as of this writing.
During migration, a GPT's instructions become a "skill" within a plugin, and connected apps transfer over. Custom actions, meaning any custom integrations built into the GPT, do not transfer automatically and must be rebuilt manually. After a GPT is migrated, the original remains usable until final retirement but becomes read-only.
This directly affects coaches and experts who built their "AI coach" or "AI clone" using a custom GPT, according to OpenAI's own migration FAQ, as summarized in reporting checked September 21, 2026. OpenAI's help center is the source of record for current dates, since this is an active rollout.
What Coaches Using Custom GPTs Must Do Now
If you have been running an AI coaching product through a custom GPT, you are facing a forced migration with a hard deadline for at least one segment of users, and a coming one for everyone else. The migration process preserves your GPT's core instructions by converting them into a plugin "skill," but this is not a seamless handoff.
Any custom integrations you built, whether for client management, payment processing, or specialized workflows, will not transfer automatically. You will need to rebuild these connections manually in the new Plugins system. OpenAI's public documentation does not clearly specify what happens to uploaded knowledge files during the migration. That is exactly why coaches should export and back up their own source material now rather than assume it carries over automatically.
You will also need to communicate the transition to your clients. The way they access your AI coach is changing. They will need to learn a new system, which can create friction in what is otherwise a personal, trust-based relationship.
For Enterprise users, the timeline is fixed: December 11, 2026 is when the GPT stops working. Personal account users face the same eventual outcome, just without a confirmed date yet, which means planning around a deadline that is certain but not fully specified.
What To Check This Week
- Save your system instructions. Copy the full instructions that define your coach's tone, expertise, and boundaries, and store them outside the platform.
- Export your original source files. Do not assume uploaded knowledge files transfer cleanly. Pull the books, transcripts, and frameworks from your own drive.
- List every custom integration. Document what each one connects to and what breaks if it disappears.
- Plan client communication. Decide how and when you will tell clients about the change, and what continuity you can promise.
- Test an alternative now. If you are evaluating a different setup, test it with your real content before the deadline forces the decision.
Why Building on Consumer Features Was Always Risky
The retirement highlights something many coaches overlooked when they first heard "build an AI version of yourself" and reached for ChatGPT because it was free and required no setup. A custom GPT was never designed for what a coaching business actually needs.
There was no way to charge clients directly for ongoing access. There were no proactive notifications to bring people back between sessions. The memory available inside a single chat did not carry the kind of long-term, cross-session context a real coaching relationship depends on.
The GPT Store itself is a telling example of the same gap. OpenAI's own creator payout program pays US-based builders based on user engagement, not a subscription price the creator sets, and most builders earn nothing because they never reach the required weekly conversation volume. There was never a mechanism for a coach to set a $1-2K annual price and collect it directly through the store. This retirement is really the same gap showing up a second time: a general-purpose AI feature was never built around a creator charging for access to their own expertise.
Most critically, coaches building a monetizable product on a free consumer feature were depending on decisions made for OpenAI's broader roadmap, not for their business. There was no revenue-sharing model tied to coaching outcomes, no dedicated onboarding for this use case, and no say in the timeline when the feature changed direction.
Dedicated AI coaching platforms are built differently from the ground up. Delphi and Coachvox, along with platforms like BuddyPro, exist specifically to give an expert billing, retention mechanics, and knowledge management as core product features, not an add-on to a general chat interface. Their business model depends on the coach's product working, not on how many people use a free consumer app.
Should You Rebuild DIY or Move to a Dedicated Platform?
There are two real paths forward: rebuild with API-based tools, or move to a platform built for this specific business.
The DIY route is technically possible. A capable development team can build a custom retrieval pipeline on top of OpenAI's or another provider's API, add subscription billing, build a client interface, and maintain all of it as the underlying models and APIs keep changing. That gives full control.
It also has a real cost. Designing the knowledge ingestion, the conversation logic, the payment flow, and ongoing support, then keeping all of it working as AI infrastructure shifts again, is months of work and budget for most coaches, not a weekend project.
The platform route is faster to get back to actually coaching. On a platform built for subscription billing from day one, you upload the content you already have, books, courses, recordings, frameworks, and the system does the technical work of turning it into a working coach.
The retention side matters more than it first appears. An AI twin that can start a conversation instead of only responding to one, remember months of a person's history, and hold a real mentoring relationship is what supports a premium, recurring price. That is a product design question, not just a knowledge base with a chat window on top of it.
If you are already charging, or planning to charge, $1,000 or more a year for access, the infrastructure needs to match that positioning: reliable, private, and built for the relationship, not retrofitted from a free chat feature.
What "Live Within Days" Actually Involves
It is worth being concrete about what moving to a dedicated platform looks like in practice, since "live in days" can otherwise sound like marketing shorthand.
You start by uploading the same material you would have put into a custom GPT: books, courses, transcripts, frameworks, whatever constitutes your methodology. The platform's AI analyzes that material and maps it into a working coach, organizing it into the knowledge areas your audience actually asks about.
From there you customize tone and test real conversations, adjusting how the AI responds until it sounds like you rather than a generic assistant. Connecting billing, usually through Stripe, comes next, along with setting your price and deciding how the product launches under your own brand.
Most of the time that a migration actually takes goes into these decisions, tone, pricing, how to introduce it to your audience, not into technical setup, since the platform handles the underlying engineering. That is the practical difference between this path and a from-scratch API build: the work left for the coach is the part only the coach can do.
The custom GPT retirement is a reminder that a coaching business built on a free consumer AI feature is only ever as stable as that company's product roadmap. Coaches sorting through this transition can see how a platform built specifically for AI coaching handles knowledge, billing, and retention at buddypro.ai.
Related reading
Written by