Rate your prompts so the best ones stand out
A quick star rating, a one-line note on what worked, and the model you used turn a pile of prompts into a library you trust - kept light enough to actually do.
Prompts don’t get better by accident. You write one, it works well enough, and you forget the dozen small adjustments that got it there. Next month you tweak it again, undo a change that was actually helping, and wonder why the output got worse. A simple rating fixes this by giving your prompts a memory: a quick verdict on how well one performs, the model it works best with, and a note on why. It turns “this feels better” into something you can actually see.
What to capture
You only need three things, and none of them should break your flow:
- A star rating - one to five, your honest read on how well the prompt does its job.
- The model it works best with -
gpt,claude, a local model, with the size if it matters. - A one-line note on why: what’s good, what’s off, what you’d change next.
That’s it. The point is not to build a research dataset; it’s to leave yourself breadcrumbs. When you next open the prompt, the rating and note tell you whether it’s a keeper and which direction was working, sparing you from relearning it. The Promptsson workspace keeps the rating attached to each prompt, so the verdict travels with the thing it describes rather than living in a separate file you forget to update.
Rate honestly, and don’t agonise
A star rating is only useful if you give it fast. Don’t invent a rubric or argue with yourself over whether something is a three or a four - your gut read a second after seeing the output is the signal you want. Five stars means you’d reach for it again without thinking; one or two means it needs work before you trust it. The number is a bookmark, not a grade.
The real value shows up across your library. When your strongest prompts carry four and five stars, they’re easy to spot and reuse, and the low-rated ones flag exactly where your next bit of editing should go. You don’t need precision on any single prompt; you need enough of a signal to sort the keepers from the works-in-progress.
Notes are where the value lives
The rating tells you whether a prompt is good; the note tells you why, and the why is what lets you improve it. Good notes are specific and short:
- “Too formal for a customer email - needs a warmer opening.”
- “Ignored the 80-word limit again. Make the limit the first line, not the last.”
- “Great on Claude, rambled on the local model. Stick to
claude.”
Notice that each note points at a concrete change. A note that just says “meh” helps no one. The discipline is to finish every note with the verb you’d act on: shorten, add, remove, reorder, constrain. Those notes accumulate into a clear to-do list for the next revision.
When you do revise, the change usually means adding one constraint or pulling a detail out as a variable - exactly the kind of edit the reusable prompt structure is built to absorb without breaking the parts that already work.
Turning ratings into better prompts
The rating earns its keep when you act on it. A simple loop:
- Use and rate. Run the prompt for real work, then set a star rating and a one-line note. Don’t break your flow to write an essay.
- Look for the weak spots. Scan your lower-rated prompts and the notes attached to them. The repeat complaints are your priorities.
- Make one change. Address the most common complaint with a single edit, then use the prompt a few more times and see if the rating climbs.
Changing one thing at a time is the whole trick. If you fix three things at once and the result improves, you won’t know which fix mattered - and you’ll carry two useless changes forward. One edit per round keeps cause and effect visible.
Keep it lightweight or you won’t keep it
The fastest way to abandon the habit is to make it a burden. A few habits keep it sustainable:
- Rate at the moment of judging. You’ve already decided whether the output was good - set that star right then, not later.
- Skip the boring ones. If a prompt is unremarkable, a bare rating with no note is fine. Save the writing for the prompts that surprised you.
- Record the model when it matters. Noting which model a prompt works best with pays off the day it behaves differently across two of them.
Used this way, the rating becomes the quiet engine behind a library that improves on its own. Pair it with sensible categories and folders so trusted prompts are easy to spot - and rest easy knowing it all stays on your device, per the local-first approach. Browse the starter library for prompts worth putting through this loop.