Can custom GPTs still make money now?

Why has making money with custom GPTs suddenly become much harder?

Making money with custom GPTs is still possible today, but the window for building a new business around the GPT Store itself is closing fast.

The reason is unusually concrete. OpenAI is preparing to retire custom GPTs and move their workflows into plugins. Existing GPTs can still run for now, but OpenAI has already stopped new GPT creation and publishing on personal Free, Go, Plus and Pro accounts. Business, Enterprise and Edu workspaces still have more flexibility depending on administrator settings.

That is a huge change from the original pitch. When OpenAI announced GPTs in 2023, it said builders would eventually earn money according to usage. By the time the GPT Store launched in early 2024, users had already created more than 3 million GPTs. OpenAI again said a builder revenue program was coming.

The Store grew much faster than its monetization system. OpenAI eventually tested usage-based payments with a small group of popular US builders, but it never opened the program broadly. There was no standard payout rate, no generally available revenue-share system and no equivalent of setting a $10 monthly price for your GPT.

Today, the direction is much clearer. OpenAI's current Help Center says custom GPTs are being retired and recommends moving useful workflows to plugins. For affected Enterprise customers, existing GPTs are scheduled to stop running on December 11, 2026, and other plans are expected to follow a similar transition.

That changes what “making money with custom GPTs” means. The interesting opportunities now are extracting value from GPTs that already have users, using them to bring customers into another business, and selling the underlying AI workflows to companies.

Stage	What happened	What it meant for builders
GPT launch	OpenAI announced customizable GPTs	Easy entry and promised future earnings
GPT Store launch	More than 3 million GPTs had already been created	Huge supply appeared almost immediately
Monetization pilot	A small group of US builders received usage-based payments	Direct monetization stayed restricted
Current transition	New personal GPT creation is closed and GPT retirement is planned	Long-term value is moving toward workflows, plugins and apps

Is the GPT Store still a real way to build a business?

The GPT Store can still send users to existing GPTs today, but we would no longer build a new business that depends on Store traffic or GPT payouts.

The original opportunity looked enormous because ChatGPT itself became one of the largest consumer software products in the world. OpenAI now says ChatGPT has more than 1 billion weekly users. Even a tiny share of that audience sounds attractive.

But ChatGPT traffic and GPT Store traffic are very different things. A builder has to compete with ordinary ChatGPT, OpenAI's own tools and an enormous catalogue of other GPTs. The more than 3 million GPTs already created by the Store's launch showed how quickly supply could explode when creating a product required almost no code.

Consensus is one of the better examples of what worked. Its research GPT accumulated more than 5 million conversations. According to co-founder Eric Olson in an interview with WIRED, around 10% to 15% of new Consensus subscribers were coming through the GPT.

That is commercially meaningful, but the money came from Consensus subscriptions rather than from OpenAI paying for millions of GPT conversations.

Another builder profiled by WIRED shows the other end of the market. Medical student John Villocido created more than 250 GPTs and even had his Books GPT featured at launch. He was not included in OpenAI's monetization pilot. After adding third-party advertising to his GPTs, he said a good month brought in roughly $200.

Those two cases tell us much more than the raw number of GPT Store users. One GPT became a meaningful customer-acquisition channel for an established subscription product. Hundreds of GPTs plus official Store exposure produced modest advertising income for another builder.

The difference was what happened after people used the GPT.

Can you still create and publish a new custom GPT today?

For an individual creator using a personal ChatGPT account, no: new custom GPT creation and publishing are currently closed.

OpenAI's current documentation explicitly lists Free, Go, Plus and Pro accounts. Existing GPTs remain usable for now, and eligible creators may still be able to edit GPTs they already own.

Managed workspaces are different. Business, Enterprise and Edu customers may still have GPT creation, editing and publishing depending on workspace permissions, although OpenAI is preparing those customers for the same broader retirement.

For affected Enterprise workspaces, OpenAI's current migration plan includes a September 25, 2026 planned cutoff for creating new GPTs before the December 11 retirement. The company is telling customers to identify valuable GPTs, preserve their instructions and reference material, and move the workflows they want to keep into plugins.

So someone searching today for a side hustle based on publishing dozens of new GPT Store products is already working from an outdated playbook. That strategy is unavailable on normal personal accounts and has a limited remaining life elsewhere.

Is OpenAI actually paying custom GPT creators now?

For an ordinary custom GPT creator, we would currently assume $0 in direct payments from OpenAI.

OpenAI did launch a real monetization test. Its published FAQ said it was working with a small group of US-based builders who had created popular and engaging GPTs, with earnings based on usage.

The important detail was the admission policy: OpenAI said it was not accepting additional builders.

That pilot never became the broad creator program many people expected after the GPT Store announcement. We found no current OpenAI program where a new builder can enroll, see a published rate per conversation, qualify after hitting a usage threshold or receive a standard percentage of revenue.

OpenAI's current GPT documentation has now moved on to retirement and migration. A widespread GPT creator payout system emerging at this stage looks very unlikely.

So if someone is estimating how much a new custom GPT could earn directly from OpenAI, using zero as the baseline is much more realistic than trying to guess a payout per thousand conversations.

Did people actually make money from custom GPTs?

Yes, people have made real money from custom GPTs, but the best documented examples usually made the money around the GPT rather than from OpenAI paying for its usage.

John Villocido's experience gives us one lower-end benchmark. WIRED reported that he had created more than 250 GPTs, including a GPT featured by OpenAI, before trying third-party advertising. A good month brought in around $200.

That is revenue, but it is hardly the passive-income machine people imagined when the GPT Store was compared with an app marketplace.

Consensus provides a more interesting model. Its GPT passed 5 million conversations, and the company said roughly 10% to 15% of new subscriptions were being generated through it. We do not have enough public information to calculate the exact revenue attributable to those users, but a channel producing around one in every seven to ten new subscribers can clearly be valuable.

The contrast is useful. Villocido tried to monetize the conversation itself through advertising. Consensus monetized the customer's next step.

Freelancers found a third route by charging businesses directly to create GPTs and related AI assistants. Current marketplace listings still advertise this work from well below $100 for basic setups to several hundred dollars or more than $1,000 for implementations involving APIs, company data and automation.

So yes, custom GPTs have made money. The mistake was assuming the GPT Store payout would be the main source.

How much usage can a successful custom GPT actually get?

A successful custom GPT can still reach millions of conversations, so audience size was never the biggest weakness in the model.

Consensus passed 5 million conversations with its research GPT. A 2025 academic study of the GPT ecosystem found even larger numbers among some creators. In the researchers' collected dataset, gptonline.ai had generated more than 48 million conversations from 26 GPTs.

The same study also showed how uneven the market could be. Another creator had published 9,401 GPTs without producing comparable engagement.

That is a striking gap. One creator averaged roughly 1.8 million conversations across 26 GPTs, while simply flooding the marketplace with thousands of products did not guarantee anything close to that level of demand.

It suggests that the GPT Store behaved more like a hit-driven content marketplace than a catalogue where every extra product created another predictable stream of income.

Millions of conversations were achievable. Predictable monetization was much harder.

Can a popular custom GPT still bring in paying customers?

Yes, an existing popular custom GPT can still make money today by sending users into a product or service that you actually charge for.

Consensus gives us the clearest public example. Its GPT generated more than 5 million conversations, while the company told WIRED that roughly 10% to 15% of new subscribers were coming through that GPT.

That means the GPT was doing much more than collecting impressive usage numbers. It was functioning as a customer-acquisition channel.

The same logic can work for a software company, agency, consultant or education business. A GPT can answer part of a user's question, demonstrate expertise, perform an initial diagnosis or help the person discover the paid product that solves the larger problem.

This is a better business than depending on conversation royalties because the company controls what happens after the GPT interaction.

There is one obvious limitation now: relying entirely on GPT Store discovery would be risky when the format is being retired. An existing GPT with traffic can still be useful, but we would move those users toward something portable such as an account, email relationship, SaaS product, service or another owned destination.

The valuable part is the customer relationship created by the GPT.

What 1 million GPT conversations produce	Commercial value
No payout and no next step	Potentially very little direct revenue
Advertising impressions	Some revenue, but usually dependent on large traffic
Qualified consulting leads	Potentially high value with relatively few conversions
SaaS registrations and subscriptions	Potentially very high value
Internal employee usage	Value comes from time saved rather than sales

Could you charge people directly to use a custom GPT?

Custom GPTs never got the simple built-in paywall that would have made them much easier to monetize as standalone products.

A normal builder could not publish a GPT, choose a price of $9.99 a month and receive subscription revenue through the GPT Store. OpenAI instead experimented with usage-based builder payments.

Builders could connect GPTs to external APIs through Actions, which made much richer business models possible. A GPT could interact with an existing product, authenticated system or proprietary database. But billing and customer relationships generally had to live outside the GPT itself.

OpenAI's newer app ecosystem is moving closer to what developers originally needed. ChatGPT apps can connect to external backends, and customers can authenticate into services they already pay for. OpenAI has also launched Instant Checkout in ChatGPT through the Agentic Commerce Protocol for supported commerce flows.

Developers should still be careful about getting ahead of the facts again. OpenAI's current Apps SDK documentation says that broader app monetization details will be shared later, and support for the Agentic Commerce Protocol within the Apps SDK is still described as planned.

After what happened with GPT Store revenue sharing, we would treat future app monetization as potential upside until the economics are actually published and available.

Are custom GPTs too easy to copy to make serious money?

Basic custom GPTs are extremely easy to copy, which makes generic prompt-based GPTs weak standalone businesses.

That was partly the point of the product. OpenAI wanted almost anyone to be able to create a specialized GPT without writing code. More than 3 million appeared within roughly two months.

When millions of people have access to the same underlying models and the core product consists mostly of instructions, an idea can be reproduced very quickly. A useful “SEO assistant,” “PDF summarizer” or “fitness coach” can have dozens of similar alternatives.

The harder-to-copy GPTs usually contain something beyond instructions: proprietary data, a specialized database, customer history, internal company knowledge, APIs or a workflow embedded in the way an organization already operates.

OpenAI's own enterprise research supports that view. The company found that heavily deployed organizational GPTs tended to capture institutional knowledge or automate workflows through connections to internal systems.

BBVA is a strong example. The bank created more than 20,000 custom GPTs, with around 4,000 used frequently. One internal assistant in Peru was used by more than 3,000 employees and reduced the time required to handle certain queries from roughly 7.5 minutes to about one minute, according to OpenAI's BBVA case study.

Nobody is paying for that assistant because its prompt is difficult to reproduce. Its value comes from fitting a real workflow inside a large organization.

That is why the more defensible opportunity now sits further away from the simple GPT builder.

Can freelancers still make money building custom GPTs for clients?

Yes, businesses still pay freelancers for custom GPT and AI-assistant work, although basic GPT setup has become a cheap service.

Current Upwork listings show simple custom GPT packages starting below $100. More involved packages regularly move into the hundreds of dollars, while some offers involving integrations and business workflows reach roughly $750 to $1,250.

Advertised prices do not tell us how many orders each freelancer receives, so we should not mistake them for average earnings. They do show something useful about how the market values different kinds of work.

At the inexpensive end, sellers usually offer instructions, uploaded documents, basic knowledge bases and simple configuration.

Higher-priced packages tend to add APIs, RAG systems, databases, CRM connections, automation, testing or custom integrations. In several current listings we reviewed, moving from basic GPT configuration to the most advanced package increased the advertised price by roughly three to ten times.

This price difference makes sense. Almost anyone can learn to configure a GPT. Understanding a company's process, connecting its systems and making the workflow reliable requires substantially more work.

There is also a current opportunity created by the retirement itself. Companies that accumulated useful GPTs now need to decide what to keep, document their existing workflows, rebuild Actions that will not migrate automatically and test their replacements.

Calling yourself a “custom GPT builder” will probably age badly. Selling AI workflow implementation, migration and integration work gives the same skill set a much longer commercial life.

Current freelance offer we reviewed	Basic package	Higher package
GPT trained on FAQs or documents	$59	$199
Business GPT with integrations	$45	$350
GPT / AI assistant with RAG	$90	$350
Workflow and integration implementation	$100	$750
More advanced business implementation	$400	$1,250

Are companies actually using custom GPTs enough to pay for this work?

Yes, companies are using custom GPT-style workflows at meaningful scale, and this is probably the strongest part of the whole money-making case.

OpenAI's enterprise report found that weekly users of Custom GPTs and Projects increased about 19-fold during 2025. In recent months covered by that report, roughly 20% of ChatGPT Enterprise messages were processed through a Custom GPT or Project.

BBVA makes those percentages easier to picture. The bank had created more than 20,000 custom GPTs, with around 4,000 being used frequently. OpenAI also reported that employees across BBVA were saving about three hours per week on average after the wider deployment of ChatGPT.

One internal assistant in Peru went further. More than 3,000 employees were using it, and query-handling time fell from roughly 7.5 minutes to around one minute.

Those numbers tell us why companies can justify paying someone to build these systems. Even modest time savings become large when a workflow is repeated by hundreds or thousands of employees.

The current retirement of GPTs does not erase that demand. OpenAI is explicitly giving organizations a migration path because many companies have already turned GPTs into useful internal tools.

As seen above, OpenAI found that the most widely deployed ones generally captured institutional knowledge or automated repeated work through integrations. Those problems still exist after the GPT format disappears.

Where is the custom GPT opportunity moving now?

The custom GPT opportunity is currently moving toward plugins, ChatGPT apps and AI workflows connected to real business systems.

OpenAI says useful GPT instructions can move into skills inside plugins, while connected apps can be added to the same plugin. Custom Actions require more work because they do not transfer automatically and may need to be rebuilt using supported connectors or custom MCP servers.

That migration makes the direction of travel fairly obvious. OpenAI wants reusable instructions to work alongside external tools, company data and actions instead of living inside isolated custom chatbots.

ChatGPT apps push this further. Developers can build chat-native interfaces, connect their own backend and let customers authenticate into an existing service. OpenAI has also opened an app directory where apps can be discovered and invoked from ChatGPT.

This is much closer to a conventional software business. The developer can own the backend, maintain a customer account, provide premium features and, in some cases, transact with the user rather than hoping OpenAI eventually decides how much a conversation is worth.

OpenAI still has work to do on app economics. Its current Apps SDK documentation says further monetization details are coming later, even though Instant Checkout already exists for supported commerce flows and the Agentic Commerce Protocol is part of the broader direction.

So we would not assume that the new app directory automatically becomes the App Store opportunity people once expected from GPTs. What looks better this time is the ability to connect ChatGPT distribution to a product and revenue system the developer already controls.

Can one person still build a profitable AI product without much coding?

Yes, one person can still build a small AI business with surprisingly little coding, but simply configuring a custom GPT no longer counts as much of a product.

The useful legacy of custom GPTs is how cheap they made experimentation. Someone could take a narrow problem, write the instructions, attach reference files and test whether people actually wanted the workflow before building traditional software.

That advantage is still here. Plugins, apps, APIs and low-code tools make it possible to prototype specialized AI products much faster than a few years ago.

What the GPT Store experience exposed is everything that still comes after the prototype.

You need people to find the product. You need some reason for them to come back. You need access to information or capabilities that thousands of competitors cannot instantly reproduce. And eventually you need a payment mechanism you control.

A solo builder therefore has a real opportunity in something like a specialized research tool, sales workflow, compliance assistant, industry database or customer-service product where AI makes the product dramatically cheaper to create.

Publishing another generic “marketing expert GPT” and hoping Store discovery does the rest is a very different bet, and these days it is barely a viable bet at all.

Should you start a custom GPT business from scratch now?

No, we would not start a new business today whose core product is a public custom GPT.

Personal accounts can no longer publish new GPTs, OpenAI is preparing to retire the format, and the broad GPT Store revenue-sharing program people expected never arrived.

There are still situations where an existing GPT makes sense. Someone who already has meaningful GPT traffic can keep converting those users while moving the audience elsewhere. A company may continue using an internal GPT while preparing its replacement. A consultant may use the format temporarily while migrating a client's workflow.

Starting from zero is different.

The useful things to build now are portable: good instructions, proprietary knowledge, evaluation examples, customer accounts, integrations, backend logic, payment relationships and distribution you control.

Those assets can survive a move from a GPT into a plugin, ChatGPT app, website, API product or whatever interface comes next.

The public GPT itself cannot.

So, can custom GPTs still make money now?

Yes, custom GPTs can still make money now, but the easy-sounding GPT Store business is effectively finished.

OpenAI did create a real usage-based payment pilot, but it stayed limited to a small group of US builders. The broad creator revenue system announced around the GPT Store never became generally available. New GPT creation has now closed on personal accounts, while OpenAI is actively preparing to retire existing GPTs.

At the same time, we found plenty of evidence that useful GPT workflows themselves have economic value.

Consensus turned more than 5 million GPT conversations into a channel that generated roughly 10% to 15% of its new subscribers. Freelancers currently advertise AI-assistant implementations from tens of dollars for simple work to more than $1,000 for deeper integrations. Inside companies, weekly use of Custom GPTs and Projects grew about 19-fold during 2025, while BBVA alone created more than 20,000 GPTs and regularly used around 4,000 of them.

Put together, those numbers draw a fairly sharp line.

Publishing a GPT and waiting for OpenAI to pay for conversations is no longer a sensible business plan. Existing GPTs can still generate leads, subscriptions and client work, and companies are clearly willing to use specialized AI workflows when they save real time.

For someone starting today, the money has moved one layer deeper. Build the workflow, own the customer relationship, connect useful data and systems, and charge for the product or service around it. Whether the interface is called a GPT, plugin or app matters far less than it did two years ago.