Reviews and feedback: improving a prompt you've already published

An author has published a prompt that prepares lesson plans for trainers. Three reviews come in: "5/5, very useful," "2/5, too long for what I wanted," "3/5, doesn't work well with my tool." He's tempted to read all three as a verdict on the quality of the prompt. Reading them again, he sees three different issues: a prompt people like, a listing that promised something other than what the buyer wanted, and a question of compatibility with a tool.

A review usually mixes several things. Before you change anything, separate what belongs to the prompt, what belongs to the listing, and what belongs to the buyer's expectations.

What reviews show you

On YouPrompt, only people who have bought a prompt can leave a review: a rating from 1 to 5, with an optional comment. These are reports from real use, so they matter. They're rare at the start, and a single very harsh or very enthusiastic review isn't enough to draw conclusions. Look for the patterns that come back.

Sort feedback into four families

This sorting keeps you from complicating the prompt to meet an expectation the listing never set, and from defending a prompt that really does have a flaw. When you have several reviews, the AI can help you classify them, as long as you remove first names and check the result:

Here are buyer reviews of a prompt that [WHAT THE PROMPT DOES]:

[REVIEWS PASTED, WITHOUT NAMES]

Sort each review into one of these families: flaw in the prompt, mismatch with the description, depends on the tool or the case, personal preference. For each review, quote the sentence that justifies the classification. Then list the patterns that come up at least twice. Do not invent any feedback, and do not suggest any changes until I ask for them.

Fix things without breaking what worked

Resist the urge to rewrite everything. One change at a time, as in iterating on a prompt:

1. Reproduce the problem on a case close to the buyer's, in a fresh conversation. If you can't reproduce it, write down the conditions: tool, input length, type of case.
2. Keep the current version in a file, so you can compare and go back.
3. Change a single thing: one instruction, one variable made more precise, one example for the missing case.
4. Rerun the old cases and the new one. A fix that solves one case and breaks another improves nothing. The testing method describes the protocol.
5. Check the listing. The excerpt, the sample result, and the description must still match what the prompt produces.

You edit a published prompt from your dashboard, with the "Edit" button.

Adjust the listing rather than the prompt

Often the answer to a review fits in two sentences in the description. If two buyers thought the prompt was aimed at a wider audience, write "for adult trainers, not for schools." If it works better with certain tools, name the ones you've tested. A buyer who is told about the limit before paying has no reason to complain about it afterward. The listing that sells covers the writing in detail.

The sample result must remain a real result, obtained with the published version. If you change the prompt, redo the sample, as described in the article on the sample result. Don't pester an unhappy buyer to change their rating: a review obtained under pressure teaches you nothing. And don't change what the prompt is for: for a different use, publish a different prompt.

Tonight

Reread all the reviews in one sitting and note the family of each one. If a pattern comes back, make a single change: reproduce the flaw, fix it, test on the old cases, and update the listing. If not, leave everything alone and come back to it once you have a few more reviews.