I've been watching coaches panic about OpenAI's custom GPT retirement announcement for the past few weeks. If you built a custom GPT on your coaching method and your clients use it every week, you're probably wondering what happens to it and what you need to do before the deadline.

Let me walk you through exactly what's changing, what stops working, and what your actual options are. This isn't theoretical. I've watched 150+ experts build AI versions of their coaching on BuddyPro, and I've seen what happens when you build on someone else's platform versus owning the relationship yourself.

What Actually Happens on December 11, 2026?

Per OpenAI's retirement FAQ (checked November 20, 2026), here's what stops working.

Standard retirement date is December 11, 2026. On that date, custom GPTs and their GPT pages become inaccessible. There's a later date of February 28, 2027, but that only applies to FedRAMP workspaces and Enterprise workspaces with an approved deferral. If you're an individual coach or Plus user, that extended date doesn't help you.

Your GPT remains usable right now until the applicable retirement date. But here's the part that catches everyone off guard: once you migrate your GPT to the new "plugin" format, the original GPT becomes read-only. It still works until retirement, but you can't edit it anymore. And you can't delete it either.

The intended replacement is a ChatGPT "plugin." During migration, your GPT's instructions become a "skill" inside the new plugin, and any connected apps get added to the plugin as apps. Knowledge files copy over into the plugin's reference files.

Now here's what does not transfer automatically, and this is where the real problems start.

Custom actions (integrations to outside services) don't transfer. You have to rebuild those separately. The GPT's selected AI model doesn't carry over. Draft or unpublished edits aren't included, migration only uses the latest published version.

And here's the part that matters most for a coach with weekly clients: sharing and access settings do not transfer. A migrated plugin starts private. That means existing users of your GPT do not automatically get access to the replacement.

Read that again. If you have clients using your custom GPT every week, and you migrate it to a plugin, those clients don't automatically get the new version. You have to manually re-onboard every single one of them yourself.

OpenAI gives you nothing automatic for that. No notification tool, no redirect from the old link, no way to confirm people actually installed the replacement.

Your existing conversations with the GPT stay in your chat history and aren't deleted when the GPT retires. You don't need to do anything to preserve past chats. But they can't be moved into a ChatGPT "project" either.

OpenAI notes that dates and availability may change or vary by account, so check the notice inside your own account for the specific timeline that applies to you.

Should You Migrate to a Plugin or Rebuild Somewhere Else?

This is the real question, and the answer depends on what you're actually trying to accomplish.

If your custom GPT is a quick reference tool, something clients check occasionally for a specific framework or methodology, migrating to a plugin might work. The instructions transfer, the knowledge files copy over, and you can keep it inside the ChatGPT ecosystem.

But if your clients use this thing every week, if it's become part of their actual coaching experience, if they depend on it for ongoing guidance and accountability, you need to ask yourself a harder question.

Why are you rebuilding on a platform where OpenAI can retire the whole thing out from under you again?

Because that's what just happened. You built something valuable. Your clients use it regularly. And now you're scrambling to migrate before a deadline you had no control over.

This is the pattern I see over and over with experts who come to us after building a custom GPT: they started with a prototype that worked well enough, their clients liked it, usage grew, and then they hit a ceiling. Either the platform changed the rules on them, like this retirement, or the engagement dropped off after a few weeks, or the GPT started giving inconsistent answers and they couldn't fix it without rebuilding the whole thing.

Here's what I tell them: if you're going to rebuild anyway, rebuild in a way where you own the relationship and the distribution channel.

A custom GPT lives inside ChatGPT. OpenAI controls access, OpenAI sets the rules, OpenAI decides when to retire a feature your clients depend on. You're building on rented land.

An AI digital twin built on a platform like BuddyPro lives under your brand, with your own pricing, distributed how you want. You keep 100% of the profit. Your clients aren't ChatGPT users who happen to have access to your GPT, they're subscribers to your coaching product. Platform risk doesn't end with custom GPTs either, so before you commit to the next platform it's worth understanding what happens to your content if you ever need to leave it.

The experts I work with who've made this shift, from a free or low-cost custom GPT to a standalone AI coaching product, consistently move from zero recurring revenue to six-figure recurring revenue. That's not because the AI is smarter. It's because the business model is fundamentally different.

When you sell direct access to an AI version of your coaching for $1,000 to $2,000 a year, you're not competing with a general-purpose chatbot subscription. You're offering something much closer to the value of your own 1:1 coaching, available 24/7, at a fraction of what a high-ticket coaching program costs.

On the top-performing instances, 60% of subscribers are active daily and 80% weekly. That retention happens because the AI remembers what someone told it three months ago, follows up proactively, and builds a real mentoring relationship over time. It's not a Q&A tool. It's a coaching relationship.

What's Your Practical Checklist Before December 11?

If you're migrating to a plugin, here's what you need to do now.

Export and back up your GPT's instructions and knowledge files. Don't assume the migration will work perfectly. Have your own copies, outside of ChatGPT.

Decide where you're rebuilding before you migrate anything. If you're staying inside ChatGPT's ecosystem, understand upfront that you'll lose custom actions and sharing settings. If you're moving to a standalone AI coaching product, start that build now so you're not rushing at the deadline.

Tell your clients proactively, with a specific date. Don't wait until December 10 to announce the change. Give them real notice, explain what's happening in plain language, and walk them through the transition step by step.

Plan to re-onboard every user by hand. OpenAI won't do it for you. You need a list of everyone who uses the GPT, a clear communication plan, and a way to confirm they've actually installed the replacement or signed up for the new version.

If you're rebuilding as a standalone product, upload all of your knowledge, not just the one framework you put in the custom GPT. Books, courses, podcasts, coaching transcripts, your complete body of work. The AI trains itself on everything you upload, and the more context it has, the better the coaching gets.

Here's what creating an AI coach this way actually looks like: you upload your content, the AI analyzes it and maps your methodology into an AI-ready knowledge architecture, training itself within hours. You review and refine until it represents your coaching with integrity. You connect Stripe, set your pricing, configure message limits, and prepare your sales page. Then you launch, white-labeled under your own brand, with 24/7 availability, keeping 100% of the profit after covering the real AI usage your subscribers generate (typically 75-85% margins).

The build process takes days, not months. You're not hiring a dev team or maintaining infrastructure. The platform handles the technical side. You focus on making sure the AI coaches the way you would.

The Real Cost of Rebuilding on Rented Land

I've watched what happens when experts try to build their own AI coaching assistant from scratch, outside of any dedicated platform. A working prototype is easy. A product people actually pay for and keep using is not.

It forgets past conversations. It doesn't follow up. It gives generic answers. You end up firefighting wrong answers and complaints every week. Memory, proactive messages, payment edge cases, and app store requirements stretch weeks into months.

Retention drops. Users quit after a few weeks. You get refund requests from inconsistent answers. Cost pressure pushes you toward cheaper models, which lowers quality further. Models get deprecated and prompts need constant retuning. Payment failures, webhooks, and plan changes break in production.

It took my team more than 2 years and a team of 22 to solve this properly on BuddyPro: a 15+ page master prompt refined across 150+ instances, 50+ side-process prompts, and continuous model retuning. You can read the full breakdown of why building your own AI twin is a bad idea.

The reason I'm walking you through all this isn't to scare you away from rebuilding. It's to make sure you understand what you're actually choosing.

If you rebuild as a plugin inside ChatGPT, you're choosing convenience and familiarity. But you're also choosing to stay dependent on a platform that just proved it will retire features you and your clients depend on, without asking either of you first.

If you rebuild as a standalone AI coaching product, you're choosing to own the relationship, own the pricing, and own the long-term value of what you've built.

The experts who've made that shift don't look back. They're generating real recurring revenue, their clients get a better coaching experience, and they're not waiting on the next platform change OpenAI, or any other vendor, decides to announce.

The real question isn't what to do before December 11. The real question is whether you keep rebuilding on rented land every time a platform changes the rules, or finally own the thing you built.

Related Articles


If you want to talk more about custom GPT retirement and migrating a coaching assistant your clients actually depend on, feel free to catch me on LinkedIn or wherever I'm at in the world at the moment you're reading this, which is usually San Francisco, Prague or Bali.

David Riha · AI Digital Twin Builder · November 20, 2026

Share this article