A prompt that works is rarely just the text. The words on the page are the easy part to copy. What's harder to recover later is the surrounding record: what job it was for, which model it ran on, the shape of output you expected, and one example input that produced a good reply.
Without that record, reopen becomes guesswork. You paste the same wording into a new chat and get something close, or something odd, and you can't tell whether the drift came from the model, the missing example, or a constraint you forgot to restate.
A practical checklist, kept beside the text:
- Job: one line on what this was for (summarising a long thread, say, or drafting an FAQ answer).
- Model: which chat it ran in when it worked.
- Output shape: bullet list, table, short paragraph, JSON-ish block, or whatever you actually needed.
- Example input: one real-ish sample (stripped of private details) that previously produced a solid reply.
That's enough to judge whether a later paste is still the same tool, or a different one wearing the same words.
For context, I build PromptAlo, a Chrome extension that saves prompts to a library, adds AI tags to them, and sends any saved one to Claude, ChatGPT or Gemini in one click. The four notes above are still worth writing yourself, because they describe how you used the prompt, and that part only you know.
If you already keep prompts somewhere: which of those four notes do you write down, and which stay only in your head?
https://promptalo.conceptgame.co.uk
Disclosure: this article was written with the help of AI and reviewed and edited by me before publishing.
Top comments (0)