You finally get an AI answer that is useful. Two weeks later, you need the same kind of help and cannot remember what you asked.
A small prompt notebook solves that everyday problem. It is a note or document containing requests you want to reuse, plus the checks that helped you judge the answers. You can start in a notes app, a plain document, or a paper notebook. You do not need a special prompt-management subscription.
Your first exercise: save one reusable prompt, try it with the fictional example below, and record what you changed. Allow about 15 minutes. The aim is a small collection you actually use.
Save the instructions and the checking habit
A prompt is the request you give an AI assistant. A reusable prompt keeps the job and format steady while leaving spaces for details that change.
For example, “Draft a friendly reply under 100 words using these facts” can be reused. A long chat containing a friend’s address, last month’s deadline, and an unrelated conversation makes a poor template.
Clear instructions, relevant context, smaller steps, and follow-up feedback are also principles described in Anthropic’s prompting guidance. The templates below apply those ideas to ordinary tasks. Their usefulness still depends on the task, the information you supply, and your review.
Give each saved prompt five small fields
- Name: something you will recognize, such as “Reply without adding promises.”
- Use it when: the job this prompt is meant to do.
- Prompt: the reusable instructions, with bracketed spaces for new details.
- Check before using: the specific errors that would matter.
- Last tried: the date, assistant used, and one note about the result.
Keep identifying details out of the template. Use [recipient], [date], or [amount] instead. Before sending a new request, replace every bracketed space and remove old facts. A notebook stored in a shared account may itself be visible to others, so keep it as spare as possible.
Prompt 1: a clear reply without extra promises
Use it when: you know what you want to say but want help saying it clearly.
Draft a [friendly / neutral / firm] reply to [recipient or role], under [word limit] words.
Facts I want to communicate: [facts].
What I want to happen next: [request].
Commitments I am willing to make: [commitments, or “none”].Use only these facts. Do not invent a reason, attachment, deadline, apology, or promise. If an essential detail is missing, ask before drafting. After the draft, list any wording that creates a commitment.
Check: names, dates, tone, and promises. A fluent reply can quietly turn “I will consider it” into “I will do it.” See our email-writing practice guide for another worked example.
Prompt 2: a task list without invented deadlines
Use it when: you have a few messy notes and want to see what is actionable.
Turn these notes into a short task list: [notes].
For each task, show the action, the owner if explicitly named, and the deadline if explicitly stated. Write “not specified” for missing owners or dates. Preserve anything marked tentative. Separate reminders and open questions from agreed tasks. Do not add tasks or put anything on a calendar.
Check: that every task exists in the original notes, and that “maybe” did not become a commitment. If you later want suggested priorities, ask for them separately and keep them labeled as suggestions.
Prompt 3: compare options without choosing for me
Use it when: you have two or three options and want to see the trade-offs clearly.
Compare these options using only the details I provide: [options and facts].
My priorities are [priorities]. Create a compact comparison showing the known facts, relevant differences, and unanswered questions. Keep missing details marked “unknown.” Do not treat missing information as a disadvantage or invent prices, features, reviews, or guarantees. End with the three questions most useful to answer before deciding.
Check: the same criteria are used for each option, totals are correct, and an unanswered question has not turned into an assumed fact. Our secondhand-shopping exercise shows why a missing measurement can matter more than a small price difference.
Test one prompt before adding more
Try the task-list prompt with these fictional notes:
Casey agreed to bring folding chairs to the picnic. Jamie will confirm the headcount by Thursday. We might move indoors if rain is forecast; no decision yet. Need to ask who is bringing drinks.
A useful answer should identify Casey’s chair task with no stated deadline, Jamie’s headcount task with Thursday as the deadline, the indoor location as tentative, and drinks as an unanswered ownership question.
It should not assign drinks to Casey, decide the weather, or tell everyone to arrive on Thursday. This example gives you something specific to check instead of judging an answer only by how polished it sounds.
Now change the notes: “Jamie is unavailable; nobody has agreed to confirm the headcount.” Run the revised example as a fresh request. Does the answer keep the owner unresolved, or carry Jamie forward from the earlier version?
These are teaching examples and expected checks, not a report of a measured product test. You are testing whether your chosen assistant follows the prompt in your own conversation.
Improve the small collection you use
If an answer misses the point, add one precise instruction that addresses the problem. “Keep tentative ideas separate” is more useful than repeatedly adding “be accurate.” Then try the revised prompt again with different details.
Save a short note such as: “October 10: separated tentative plans from agreed tasks; still need to check dates.” Use the date you actually tried it. You do not need a complicated scoring system.
When changing assistants or returning after a long break, try a simple example again. A prompt that helped last month is a starting point, not a guarantee of identical behavior.
Your next small win: keep the best one of these three prompts, give it a name, and use it for one ordinary task this week. Add another only when you have a reason to reuse it.
AI disclosure: This guide was created with AI assistance. Sources checked October 5, 2026. The picnic scenario is fictional. Templates require your review and do not guarantee accurate answers. See how we review our guides.
Illustrative photograph by Lauren Mancke on Unsplash.
