Selling custom GPTs and Claude skills as a freelancer in 2026
The store listings are the funnel. The paid work is private builds, packaged skills, training and upkeep for a client's team.
On this page
Freelancers sell custom GPTs and Claude Skills as services: private builds for a client's team, packaged skills for a product, training, and upkeep. The store listing is a way to be found, not the product. Charge a fixed fee for scoped work and add a retainer for upkeep.
Both formats teach a model how a particular business works. A custom GPT is a configured assistant inside ChatGPT. A Skill is a folder a Claude agent loads when a task calls for it. Different containers, same job, and the same project mistakes.
Two formats, one skill
A custom GPT is a ChatGPT assistant with its own instructions, uploaded knowledge files and optional Actions that call outside APIs. It runs inside OpenAI's product. When OpenAI announced the GPT Store alongside ChatGPT Team, it said users had already created over 3 million custom GPTs, that Team customers get a private section for GPTs published inside their workspace, and that a builder revenue program would pay creators by engagement (SD Times, OpenAI announces GPT Store and ChatGPT Team). I found no recent confirmation of how that program works today, so do not build a business plan on store payouts.
An Agent Skill is, in Anthropic's words, an organized folder of instructions, scripts and resources that an agent discovers and loads dynamically. At minimum it is a directory containing a SKILL.md file with a name and description (Anthropic engineering, Agent Skills). The agent loads only the name and description at startup and reads the rest when a task matches. Anthropic published the format as an open standard on December 18, 2025, with partner-built skills from companies including Atlassian, Canva, Notion, Figma, Stripe and Zapier (SiliconANGLE). The standard now has its own site, which describes it as adopted by a growing number of agent products (Agent Skills).
Treat the "works everywhere" claim with some care. An analyst quoted by SiliconANGLE was skeptical that the merits of the format guarantee adoption. Build for the client's actual tools first.
Where the paid work sits
The demand signal is clear even if the price data is thin. Upwork's In-Demand Skills 2026 report found that demand for skills tied to AI grew 109% year over year, against 23% for other skills, with AI integration up 178% (Upwork). Publicly sourced rates for GPT and Skills projects are scarce, so this post gives no price ranges. Ask three clients what they pay for similar work and price against the value of the hours saved.
These are the engagement types that tend to come up again and again:
- Private workspace GPTs. An operations or sales lead wants an assistant that knows the company's playbooks and tone, locked to the workspace. The work is scoping, instruction design, knowledge curation, Actions design and testing.
- Vertical workflow assistants. A GPT or Skill for one job in one industry, such as drafting a first-pass contract for a certain kind of client or auditing a product listing. You build it once, then sell access or setup separately. The store listing is the funnel.
- Skill packs for a team. Skills that capture how a company drafts proposals, reviews contracts or writes reports. The buyer is usually an operations lead or a CTO.
- Skills for a software product. A vendor wants an official skill so customers using Claude can connect to their product. This is product engineering and overlaps with MCP server work, because vendors often need both.
- Training. A short engagement that teaches a client's team to build and maintain their own GPTs or skills.
- Upkeep. A monthly retainer to refresh knowledge, adjust instructions after a model update and run a set of test prompts.
The upkeep line matters most. Models change, and a skill that worked in March can drift by June. The person who built the test prompts is the person the client calls.
Know where each format lives
Delivery details decide how you scope. Anthropic's documentation says custom Skills do not sync across surfaces. A skill uploaded to claude.ai has to be uploaded separately to the API, and Claude Code skills are filesystem-based and separate from both. In claude.ai, custom skills belong to the individual user and are not shared organization-wide, while on the API they are workspace-wide (Claude docs, Agent Skills overview).
For a client with ten staff, that changes the job. "Build us a skill" turns into "build it, package it for each place it runs, and write the install steps." Put those steps in the quote, or you will do them for free.
Security is part of the service too. The same documentation warns that a malicious skill can direct the model to run code or call tools in ways that do not match its description, and tells readers to use skills only from trusted sources. A buyer who reads that will ask what is inside yours. Keep scripts readable, avoid fetching from external URLs inside a skill, and hand over a short note explaining what each file does.
Scope and price the work
Three pricing mistakes show up in early projects.
The first is treating the build as an hour of configuring. A real internal assistant needs a discovery call, a pass through the client's documents, a decision about what the assistant must refuse to do, and a round of test prompts. Quote for all of it.
The second is billing by model usage. Clients dislike it, their finance teams dislike it more, and you carry the risk when the model's pricing changes. Use a fixed fee for the build and let the client pay the model provider directly.
The third is open-ended support. "I will fix any issues for six months" turns into unlimited work. A short warranty after delivery, then an optional monthly retainer with a stated scope, keeps it clean.
Write acceptance criteria into the quote. For a GPT: the ten questions it must answer correctly, the two it must decline. For a Skill: the tasks it must complete from a set of sample prompts. Our guides to the statement of work and to change orders apply directly.
Run it like any client project
This work attracts people who treat AI projects as loose and creative. Resist that. Use a written contract, a scoped statement of work, a change-order process and a sign-off for each milestone. Buyers are often early adopters who will later need to show a manager or an auditor what was built and who approved it. The paper trail is part of the deliverable.
How Delivvo runs a GPT or Skills project
Delivvo handles the business side of this work, from first contact to final invoice, so the build is the only part that takes creative effort.
- Put an inquiry form on your site. Fields can ask for the tools the client uses, who will use the assistant and what it must never do. Each submission lands on your lead board.
- Send a quote built from packages and add-ons, such as a discovery sprint, a build, a training session and a monthly upkeep add-on. The client signs the contract and pays the deposit on one link.
- Create the project and deliver each milestone in the client portal: the discovery notes, the instruction document, the skill folder, the test results. The client opens the portal with a one-time code, no password.
- Ask for approval on each milestone inside the portal. The approval is recorded, which gives the client the audit trail and gives you protection against "that is not what we agreed."
- Bill through a payment schedule (up to 12 payments) for the build, and a recurring invoice for the upkeep retainer. Clients pay through your own gateway and Delivvo takes 0%.
Contracts keep a signing record with a SHA-256 hash, IP address and device, and an archived PDF. Every document carries a verification code. Contracts and invoices are templates you edit, so have a lawyer check your standard terms, especially around who owns the finished instructions and skills. Start with the proposals tool and the client portal.
FAQ
Can I make money directly from the GPT Store?
When OpenAI launched the store it announced a builder revenue program based on engagement, to start in the US. I could not confirm how it works in 2026, so treat store payouts as a bonus. Most freelancers earn by selling builds and upkeep to clients privately.
What is the difference between a GPT, a Skill and an MCP server?
A GPT is a configured assistant inside ChatGPT. A Skill is a folder of instructions, scripts and resources that an agent loads when a task matches. An MCP server is a standard way to expose tools and data to AI clients. A skill can call tools from an MCP server.
Do I need to code to sell custom GPTs?
No. You can build useful GPTs from instructions and knowledge files. Actions, skills with scripts and test suites need some technical skill, and that is where larger projects sit.
Are Agent Skills only for Claude?
Anthropic created the format and released it as an open standard, and the standard's site says other agent products have adopted it. Check which tools your client uses before you promise portability.
How should I price a GPT or Skills project?
Quote a fixed fee for each scoped stage and keep model usage out of your price. Add a warranty period and an optional monthly retainer. Published market rates are thin, so benchmark against what a similar project costs the client in staff time.
Delivvo runs the whole client cycle for service businesses, from the first inquiry to the quote, contract, deposit, invoice and delivery, so every AI build is scoped, signed, approved and paid for in one place. Clients pay you directly through your own gateway, and Delivvo takes 0%. Start free