Selling prompts is real income and a small amount per unit. A Claude skill file listed on KissMySkills sells at $14.99 and pays its creator 35% of the net sale price; a $9 prompt pack pays 30%. That is roughly $5 and $2.70 a sale, so the business is decided by how many people ever find the listing, not by how well it was written. The packs that keep selling are the ones carrying knowledge a model cannot guess at on request.
Revenue shares and third-party fees checked August 2026. Vendor terms change - confirm on the platform's own page before you price anything.
This is the seller's side. If you are buying rather than selling, the AI prompt marketplace guide covers vetting a pack and installing it.
What does one sale actually pay?
Start with the arithmetic, because it changes what you should build.
On KissMySkills, a skill file submission earns 35% of the net sale price and a prompt pack earns 30%, paid 35 days after the sale. The 35 days is deliberate: it clears the 30-day money-back window, so a refund is never clawed back out of money already sent to you. There is a separate affiliate arrangement paying 10% on anything you refer through your own link, including products that are not yours. "Net" is defined in the terms that come by email on acceptance - assume payment processing comes off the top before the split.
Run the numbers on the actual price points:
- A $14.99 skill file → about $5.25 a sale.
- A $19 flagship skill → about $6.65 a sale.
- A $9 prompt pack → about $2.70 a sale.
So $1,000 from one skill file is roughly 190 sales. That is the number that should drive every decision after this point. Not "is this good" - it is whether 190 people with the same problem will ever see it.
The shape of a role skill file at the standard tier: one named job, a working method, and the judgment calls written down. Read one before you write yours - it is the fastest way to see how much is in a file that sells for $14.99.
View Serge →What do the other platforms take?
Three different bargains, and the difference is who does the finding.
| Platform | You keep | Who finds the buyer |
|---|---|---|
| PromptBase | 80% on marketplace sales, 100% through your own link | Mostly you |
| Gumroad | Everything less 10% + $0.50 through your link; less 30% if the buyer came from Gumroad's marketplace | You, unless discovered there |
| KissMySkills | 35% of net on skills, 30% on prompt packs | The store - listing, product page and traffic |
Sources: promptbase.com/sell and gumroad.com/pricing, both checked August 2026.
Reading that table as "keep 80%, obviously" is the mistake. 100% of a listing nobody reaches is nothing, and 0% fees on your own link only helps if you already have an audience to send. The right question is not which cut is smallest but which platform is going to produce sale number 190 - and whether you would rather spend that effort writing the next file or running your own funnel. Both are defensible. Only one of them is a writing job.
What separates a pack worth paying for?
Be honest about the competition: a buyer can ask a current model to write a prompt pack about almost anything, and it will produce something serviceable in a minute for free. If your product is a tidy list of well-formatted prompts, that is what you are competing with, and you will lose.
What survives that test:
- The thresholds a professional carries in their head. Not "analyse this contract" but which clause you check first, what number makes you walk away, what the model should flag rather than resolve. A model can produce the shape of the task; it cannot produce the specific line you learned by getting it wrong once.
- Order of operations. Real work has sequence - you do not price before you scope, you do not draft before you know the objection. Generic packs list tasks. Good ones enforce an order and say why.
- The negative rules. Half the value of a well-built file is what it forbids: the defaults it turns off, the phrasing it bans, the situations where it tells the user to stop and go find a fact instead of generating one. This is the part that is hardest to reverse-engineer from the output, and it is the part free drafts never have.
- Failure handling. What the user does when the output is wrong. Most packs assume success and go silent at the point the buyer actually needs help.
- Maintenance. Prompts degrade quietly when a new model ships - no error, just gradually worse output. A file with a visible update history against model releases is worth more than a better file that was last touched two releases ago.
The common thread: everything above comes from having done the job, not from being good at prompting. The prompt craft is the smaller half. If you want the mechanics of writing a single strong prompt, the prompt template examples cover the structure.
The $19 tier is where roles with a real body of practice sit - a close sequence, a board narrative, the numbers a CFO checks before anyone else sees them. The extra $4 is knowledge, not extra prompts.
View Victor →What does the work actually consist of?
Writing the prompts is the part that feels like the job and takes the least time.
- Pull what you already use. Your product is sitting in your chat history. The gap is that you supply half the context from memory every time and never wrote it down.
- Watch someone else run it cold. Your prompts work partly because you patch them instinctively. Give them to somebody outside your role and note every point where they stop and ask you a question - each one is a missing line in the file.
- Write the documentation. When to use it, when not to, what a good result looks like, what to do when it is wrong. A prompt without this is software without a README, and it is the most common reason a refund gets requested.
- Go narrower than feels safe. "Marketing prompts" competes with everything. "Claude skill for B2B SaaS demand-generation managers" competes with almost nothing and reaches someone who will pay.
- Re-test on release. Every major model release is a maintenance event. This is the recurring cost of the business and it never stops.
A prompt pack is quicker to build and pays a third as much per sale as a $19 skill. It is the right format for one repeated task - and the wrong one if you are hoping a single product carries the income.
View AI Product Description Writer →How should you price it?
Against what it replaces for the buyer, never against what it cost you to make. Nobody pays for your weekend. A pack that saves a freelancer one billable hour is priced against that hour, and the ceiling is low because the buyer can see the product is a text file and knows roughly what a model would produce for free.
The way out of the ceiling is scope, not price. One task is a $9 pack. A whole role with method and rules attached is a $14.99 to $19 skill file. Several files a team standardises on is a bundle. Anyone telling you a solo prompt pack supports a $79 price is describing a market that is not showing up in the catalogue.
Where does this business fail?
- Discovery, not supply. On the large open marketplaces most listings never sell at all. Assume nobody stumbles across you; either the platform brings the traffic or you do.
- A text file copies perfectly. There is no DRM on a paragraph. Price and refund terms are the only real protection, and you should expect your best file to end up somewhere you did not put it.
- Refunds are the buyer's only trial. Nobody can test a text file before paying, so a generous refund window is what makes the sale - and it means unclear documentation costs you the revenue, not just the goodwill.
- The price ceiling is real. Low ticket, easy comparison, no lock-in. Volume is the only lever, and volume is a distribution problem.
- The model improves underneath you. This is the structural one and there is no hedge for it. Every capability release moves some of what your file provides into the model's default behaviour, for free. Files that encode scaffolding the model now supplies itself lose their value; files that encode a specific person's judgment about a specific job lose it slowest. Build for the second kind and accept that the ground is moving.
Is it worth doing anyway?
As a business on its own, for most people, no - 190 sales for $1,000 is a marketing job, not a writing job. As a way to turn work you have already done into something that keeps paying while you do other things, yes, and the setup cost is one file and an email.
KissMySkills accepts submissions both ways: sell a skill file at 35% of net, or sell a prompt pack at 30%, with the listing, product page and distribution handled at this end. Full terms come by email on acceptance.
In summary:
A skill file pays its creator 35% of net and a prompt pack 30%, paid 35 days after the sale so refunds are never clawed back. Before you build one, read a live example at each tier: Serge - SEO Specialist at $14.99, Victor - CFO at $14.99, and the AI Product Description Writer pack at $9. Then submit through Sell a skill or Sell a prompt.
1,000+ Claude skill files at $14.99, 158 prompt packs at $9, and 55 agents at $32 - indexed by job function. The fastest way to judge whether your file is finished is to read three that are already listed.
See the skill files →Browse all prompts →