A prompt pack is someone else's wording, frozen
It was written for a task that was not yours, on a model that has since changed, by someone whose reasoning you cannot see. It does not know what you are working on today, and it will not learn. That is the whole product.
Try it free, no signup See how it picks
No account, no email, no card · one prompt takes about ten seconds
- Wording written for someone else's task
- Frozen on the day it was published
- No reasoning you can inspect
- No source, no measurement
- Same file for every buyer
- 6,235 techniques, each carrying the paper it came from
- 10 specialists read the full papers behind your candidates
- A judge ranks the Top-3 and shows its reasoning
- 1,420 effects checked against the papers that reported them
- Grows every month as new research is screened
What you are paying for, and what ages
Marketplaces and prompt libraries sell wording. Wording is the least durable part of a prompt. Models get retrained, the phrasing that carried a task last year stops carrying it, and the pack sits there unchanged because nobody is paid to revisit it.
The durable part is the method: show worked examples before asking, ask for reasoning before the answer, break the task into verifiable steps, make the model criticise its own output against stated criteria. Researchers name these, test them, and publish what they measured. Those papers keep coming out. A pack cannot follow them; an index can.
Where the 6,235 come from
Far more prompting research is published than anyone can read, and most of it is noise. Our source screens that stream at scale and keeps the work worth keeping. What we index is that selected part, not the raw pile: 6,235 techniques, each with the paper behind it, what it does, when it helps and when it does not.
Then, for one task of yours, almost all of it gets thrown away. 150 close candidates. 10 read in full by separate AI specialists. 3 ranked against each other by a judge that shows its reasoning. You pick the winner and it arrives as a prompt you can paste. One method out of six thousand, chosen for the job in front of you rather than for a category someone guessed at.
The part most of this category will not show you
We check the numbers. Of the measured effects in the index, 1,420 have been verified against the papers that reported them, and where a paper contradicts the claim the technique loses its page rather than quietly keeping the number. The whole table is public: read the evidence page.
Nobody selling a prompt pack shows you that, because there is nothing behind the wording to check.
Try it before you decide anything
The single-pass version is open on this site. No account, no email, no card. Describe a task, get a prompt built on a cited technique, and see whether the output beats the pack you were about to buy.
Try the free generator Create a free account
The free tool runs one retrieval pass on the smallest model we have. The full run, with ten specialists and the judge, is what an account gives you, and the first one is on us.
Frequently asked
Is this a prompt marketplace?
No. Nothing here is sold by other users and there is no catalogue of wording to browse. There is an index of published methods and a pipeline that picks one for your task.
Can I still use my existing prompts?
Yes, and keep the ones that work. This is for the tasks where they do not.
What does it cost?
The generator on this site is free and needs no account. A full search costs 20 credits, and joining gives you enough for one, plus the prompt at the end of it, free. Credits come in monthly plans starting at $10.
Does the index keep up?
It grows every month as new research is screened and added. That is the part a static pack structurally cannot do.