Graded Prompts
Prompts Marketplace, Every prompt human-tested on its real m
Buying a prompt is a strange transaction: the product is the one thing you can't be shown. Any other digital good can be previewed before purchase, but a prompt IS its own preview — reveal it and there's nothing left to sell. So the buyer pays on faith, and the market has no natural way to police itself. A seller can attach the four best images they ever generated, or images the prompt didn't actually produce, and nobody finds out until the money has moved.
That asymmetry is the whole reason Graded Prompts exists. If buyers can't inspect, someone has to inspect for them. So nothing here auto-publishes: before a prompt goes on sale, free or paid, a person reads the listing for accuracy, runs the prompt on the model it was written for with varied inputs, and checks the seller's examples against what the model really produces. Falls short, it goes back with feedback. Passes, it gets the Graded Badge.
The cost of a bad prompt was never the $3, either. I lost an evening trying to get Kling to shoot a clothing ad — a model and a boot staged beside a car — and no matter how I rewrote the prompt, it kept putting both of them on the wrong side of the car. I burned every credit I had and never got the shot. That's the actual price: not the few dollars, but the credits and the hours you spend working out whether you're the one holding it wrong.
The other half is that the people who are genuinely good at this should get paid properly: sellers keep 85% of every sale, no listing fees, no tiers. And we sell as merchant of record, so refunds and support land on us instead of in an individual creator's inbox.
The honest downside: human review doesn't scale, and I've built the business around the one step that can't be automated. I'd still make the same call.
3 Comments
The Kling example is exactly right — the real cost is never the $3, it's the two hours you spend wondering if you're the one doing it wrong.
I build apps with Claude and have learned that prompt quality makes an enormous difference in what actually gets built. A bad prompt costs you an afternoon of broken code. A good one gets you a working feature in 20 minutes.
The human verification angle is the right call. Automated testing can check if a prompt runs — it can't check if it produces what was actually promised.
The “buyer pays on faith” problem is a really sharp way to frame prompts, and the fact that you actually run them with varied inputs makes the Graded Badge mean something. This feels like it could be pulling in a lot more organic traffic than it probably is right now, especially while you’re building trust around the marketplace. Have you noticed which kinds of prompts people are most eager to verify?
The trust problem feels like the real product here, not the prompt marketplace itself. If the grading consistently predicts what a buyer will actually get, the badge could become more valuable than the prompt listing. The hard part will be proving that consistency without making review too slow or expensive.