For a brief moment back in late 2023, it felt like we were all about to become app developers, and the best part was that nobody needed to know how to code. When OpenAI introduced Custom GPTs that November, the idea sounded almost too good to be true. Anyone could build their own specialized version of ChatGPT, publish it in a new GPT Store, attract users, and eventually get paid. OpenAI said so pretty plainly: builders would be able to earn money based on how many people used their GPTs.
The internet did what the internet always does. YouTube filled up with videos on how to make money selling GPTs, courses appeared promising to teach you how to build profitable ones, and consultants started telling every business they met that it needed a Custom GPT. People built marketing GPTs, fitness GPTs, resume GPTs, travel GPTs, real estate GPTs, and I'm fairly sure somebody built a GPT whose whole job was to help you build more GPTs. It had the same feel as the early days of the iPhone App Store: get there early, build something useful, get discovered, and make money while you sleep.
There was only one problem. The money never really showed up.
The Store Opened. The Gold Rush Didn't.
The GPT Store did launch in January 2024, and OpenAI said more than three million GPTs had already been created by then. You could search for them, share them, and use them. What never arrived was the broad revenue-sharing marketplace everyone was waiting for. OpenAI ran a limited payment experiment with a small group of U.S. builders, but it never turned into an open market where an ordinary person could build a popular GPT and watch the monthly checks roll in.
Now the story has taken an even stranger turn. OpenAI has announced that Custom GPTs will be retired on December 11, 2026, and it is asking creators to migrate them into something called Plugins. The thing people were once told might become a new source of income is going away entirely. I don't think the lesson is simply "OpenAI changed its mind," though. I think a couple of bigger things happened, and they matter to anybody trying to build a business with AI.
Easy to Build Means Easy to Copy
The biggest strength of a Custom GPT was also its biggest weakness: it was incredibly easy to create. You gave ChatGPT some instructions, uploaded a few documents, added a name and an icon, and suddenly you had your own AI. That was wonderful for experimenting. It wasn't so wonderful for building a business.
Say I spend several weeks writing instructions, adding examples, and uploading my marketing material to create The Ultimate Small Business Marketing GPT. The next week someone publishes The Ultimate Small Business Growth GPT, then someone else releases Marketing Mastermind GPT. Before long there are five hundred of them, all running on the same underlying OpenAI models and many built with nearly identical prompts. So what makes mine worth paying for? When anyone can create a competing product in an afternoon, you don't have a moat (the business term for whatever keeps competitors from simply copying you). You have a crowded parking lot.
Then there was a problem that I suspect was even bigger: I don't think people really wanted thousands of chatbots. Think about your phone for a minute. If you had to open Travel Siri to get directions, Calendar Siri to schedule a meeting, and Spreadsheet Siri to analyze some numbers, you'd be sick of it by lunch. Yet that was essentially the Custom GPT model. Every GPT was its own destination that you had to find, open, and use. What people actually want is one AI that knows how to do hundreds of things, not hundreds of AIs that each know how to do one thing. That sounds like a small difference, but it changes the entire business model.
The AI Is Becoming the Store
The GPT Store worked like an app store. You searched for a specialized GPT, picked one, and used it. The direction OpenAI is heading now is different. You tell ChatGPT what you want done, ChatGPT figures out which capability it needs and uses it, and you may never know or care what happened behind the scenes. OpenAI's own help documentation now says ChatGPT may automatically use an installed skill when its description matches your request.
There's some irony in the name OpenAI picked. It actually had something called Plugins back in 2023, before Custom GPTs existed. Those plugins let ChatGPT connect to outside services, and they were shut down in the spring of 2024, replaced largely by Custom GPTs with "Actions." Now we're back to Plugins again, but this time it's a far more ambitious version of the idea.
A modern Plugin can combine several pieces. One is a Skill, which is essentially reusable expertise: a set of instructions telling ChatGPT, "When someone asks you to do this kind of job, here's the process to follow." A Skill could hold your method for writing a proposal, evaluating a sales call, or reviewing a business plan. When you migrate a Custom GPT, its instructions become a Skill inside the new Plugin, and its knowledge files become reference files.
The more interesting piece is MCP, which stands for Model Context Protocol. The name sounds technical, but the idea isn't. Before USB, every device seemed to need its own special cable, and then USB came along and gave everything a common way to connect. MCP does the same thing for AI. It gives AI a standard way to plug into outside tools and information, instead of requiring a custom connection for every piece of software. Through MCP, a Plugin could look up records in a CRM (Customer Relationship Management system), check inventory, pull numbers from your analytics, work with a scheduling system, or talk to your own software.
That's where the leap happens. A Custom GPT was mostly "a smart chatbot that knows my stuff." A Plugin built on Skills and MCP can become "a capability ChatGPT uses to actually get work done." Instead of "go talk to Greg's marketing chatbot," the request becomes "help me figure out why my sales are down and what to do about it." ChatGPT then pulls the data, applies the right methodology, and helps finish the job.
I want to be careful here, because it's tempting to say this new architecture explains why the GPT Store money never came. It doesn't. Revenue sharing stalled back in 2024, long before ChatGPT could pick capabilities on its own, and the reasons were the ordinary ones I described above. What the new architecture does tell you is that the storefront model isn't coming back. If ChatGPT is choosing the capability for you, nobody is browsing a store looking for yours.
So Who Gets Paid Now?
This is where the story gets a lot less exciting for entrepreneurs. The original pitch was easy to understand: build a GPT, publish it, get users, get paid. The new model is built around capabilities, integrations, and workflows, which will probably make ChatGPT far more useful. What it doesn't answer is how the person who created the expertise gets paid. If I build a Skill based on thirty years of experience and thousands of people benefit from it, who owns the customer? Who sets the price? Who gets the check?
So far, OpenAI's answer hasn't been encouraging. When it opened its app directory to outside developers in December 2025, apps could send users to outside websites to buy physical goods, but selling digital goods, subscriptions, or in-app services wasn't permitted, and OpenAI said only that it was looking at other ways to monetize over time. A month later, developers on OpenAI's own forum were still being told the answer was no for any digital product or service. I haven't seen anything since that changes the picture for independent creators.
And if you did build your business on a Custom GPT, OpenAI's migration FAQ has some fine print worth reading. Your custom Actions, the integrations that actually connected your GPT to other services, don't transfer, so you'll have to rebuild them. Your original GPT becomes read-only as soon as you migrate. Your sharing settings don't come along either, because the new Plugin starts out private and making it public requires a separate submission. In other words, the people who were using your GPT don't automatically follow you to its replacement. And if your GPT was public, you probably never knew who most of them were in the first place.
Building on a Platform Isn't Owning One
The danger was never really that Custom GPTs failed. The danger was building a business around something OpenAI controlled completely. OpenAI controlled the store, discovery, the interface, the models, whether revenue sharing ever happened, and ultimately whether Custom GPTs continued to exist at all. Now it has decided they won't.
That's not a criticism of OpenAI. Facebook changes its algorithm, Google changes search, Amazon changes the rules for sellers, and Apple changes its App Store policies. That's what platforms do. The mistake is assuming that something you built on a platform is the same as owning the platform yourself.
The smarter question isn't "What GPT can I build and hope OpenAI sends me customers?" It's "What valuable product can I build that uses AI underneath it?" That way you own the website, the customer relationship, the pricing, and the recurring revenue, and OpenAI becomes one piece of your business instead of the whole thing. That's the approach I took with ReplicAItor Coach. It captures a coach's methodology and delivers AI-powered coaching between live sessions, but it lives on a WordPress platform we control, and the subscriptions and client relationships belong to us and to the coaches, not to someone else's store. Every time OpenAI renames something, I'm glad I built it that way.
There's one piece of this story I find genuinely encouraging. MCP isn't an OpenAI invention. It's an open standard developed by Anthropic, one of OpenAI's rivals, and OpenAI built its app system on top of it. That means if you put your expertise behind your own MCP server, it isn't locked into any one company's store. Your product can connect to ChatGPT, Claude, or whatever comes along next, while the customer stays yours. That's exactly the opposite of the deal Custom GPT builders got.
The Gold Rush Wasn't Completely Wrong
The ironic part is that the people predicting AI would create a new generation of millionaires may still turn out to be right. They were probably just wrong about how. The opportunity was never to build a chatbot and put it in the GPT Store. It's to take something you know how to do extremely well (your expertise, your processes, your data, your way of solving a problem) and turn it into a product that AI can deliver and scale. That's a lot harder than writing a clever prompt, and it creates a lot more value.
So yes, I remember when Custom GPTs were going to make everyone rich. They didn't, but the underlying idea wasn't crazy. The opportunity just moved, and like most gold rushes, the people who make the real money probably won't be the ones staking claims inside somebody else's store.
They'll be the ones who build something people actually want to buy.