After the third consecutive week of rewriting school notices and client status notes at the kitchen table in Kirkland, I stopped treating prompts as disposable experiments. I began keeping a short permanent list of the ones that actually reduced evening work. The fifteen prompts below are the survivors. Each has been run multiple times from clean sessions with ordinary inputs—messy meeting notes, lecture fragments, research PDFs, and the small administrative tasks that otherwise fragment attention between client deadlines and family logistics.
These are not magic spells. They are structured requests that consistently produce a usable first draft when the bracketed sections are replaced with real material. Every one of them still requires a human to read the output against the source and to correct the places where the model smooths, omits, or mildly invents.
Work Prompts That Reduce Rewriting Time
Turning messy meeting notes into action items
Prompt: “Here are my raw meeting notes: [paste]. Extract clear action items. For each item list the owner if mentioned, the deadline if mentioned, and one sentence of context. Flag any item that lacks an owner or deadline. Use a simple numbered list.”
Expected output is a short, scannable list. The model usually catches explicit tasks and sometimes invents soft owners when none were stated. Always verify ownership before sending.
Drafting a concise client status update
Prompt: “Write a 150-word status update for a client. Project: [name]. Progress this week: [bullets]. Risks or blockers: [bullets]. Next steps: [bullets]. Tone: professional, direct, no hype.”
The result is typically close enough to send after one pass for accuracy. Watch for over-optimism on incomplete work; the model tends to smooth rough edges.
Rewriting an email to be shorter and clearer
Prompt: “Rewrite the following email so it is under 120 words, keeps every factual commitment, and removes filler. Preserve the original request and deadline. Email: [paste].”
This reliably cuts length. It occasionally drops a nuance that mattered to the recipient. Compare side-by-side before sending anything with legal or financial weight.
Creating a one-page project brief from scattered notes
Prompt: “Using only the notes below, create a one-page project brief with these headings: Goal, Scope, Out of Scope, Timeline, Success Criteria, Open Questions. Notes: [paste]. If information is missing, list it under Open Questions instead of inventing it.”
The structure forces gaps into the open. That alone saves later confusion. Expand the Open Questions section yourself; the model cannot know what you forgot to write down.
Study and Learning Prompts
Explaining a concept at two different levels
Prompt: “Explain [concept] in two versions. Version 1: for a bright high-school student. Version 2: for a first-year college student in the same field. Use plain language and one concrete example in each version.”
Useful when you need to teach or when you yourself are still forming the mental model. Check the example for accuracy; plausible-sounding illustrations are a common weak point.
Generating practice questions from a chapter
Prompt: “Based only on the text below, write eight practice questions: five factual and three short-answer. After each question provide a model answer and the sentence or section it came from. Text: [paste].”
The sourcing requirement keeps the model honest. Questions sometimes stay too close to the wording of the original; rewrite a few to test deeper understanding.
Turning lecture notes into a study sheet
Prompt: “Convert these lecture notes into a one-page study sheet. Use headings, bullet points, and bold for key terms. Include a five-item ‘most likely to be tested’ list at the end. Notes: [paste].”
The “most likely” list is subjective and should be treated as a starting suggestion only. Cross-check against the syllabus or past exams when available.
Creating a spaced-repetition outline
Prompt: “Build a simple spaced-repetition outline for learning [topic]. Give me Day 1, Day 3, Day 7, and Day 14 review blocks. Each block should contain the core ideas and one active recall question.”
Helps structure study time. The content quality depends entirely on how accurate and complete the topic description you supply is.
Research and Analysis Prompts
Extracting key claims with source anchors
Prompt: “From the document below, list the main claims. For each claim quote the shortest supporting passage and note the section. If a claim lacks direct support, mark it ‘unsupported in this text.’ Document: [paste].”
This is one of the highest-value prompts for long-document work. It does not replace reading; it accelerates location of the important sentences.
Comparing two short source documents
Prompt: “Compare Document A and Document B on these points: [list]. For each point note where they agree, where they differ, and where one is silent. Use a table if possible. Do not add outside information. A: [paste] B: [paste].”
Works well when both documents are supplied in full. Outside knowledge sometimes creeps in; the instruction to stay inside the texts reduces but does not eliminate that risk.
Building a research question ladder
Prompt: “I am researching [topic]. Generate a ladder of questions: one broad starting question, three more focused sub-questions, and two questions that would require primary data I do not yet have. Explain briefly why each level matters.”
Useful for early scoping. The “primary data” questions often surface the real cost of the project before you invest too much time.
Summarizing while preserving uncertainty
Prompt: “Summarize the following text in 200 words. Explicitly flag any statements that are presented as fact but lack clear evidence inside the text. Keep the author’s original uncertainty where it exists. Text: [paste].”
Most models default to confident tone. Forcing the uncertainty flag produces more trustworthy short summaries.
Everyday Household and Administrative Prompts
Turning school notices into a family calendar entry
Prompt: “Convert this school notice into a short calendar entry. Include date, time, location, what to bring, and one sentence of context for a busy parent. Notice: [paste].”
Saves repeated re-reading of long PDFs. Always double-check the date and any fee information; those are high-cost errors.
Drafting a polite neighborhood or volunteer request
Prompt: “Write a short, friendly message to neighbors asking for [request]. Keep it under 100 words, state the purpose clearly, and include how to say yes or no easily.”
Tone is the main variable. The model sometimes lands slightly too formal or too casual; adjust to match your actual community voice.
Planning a realistic weekend with constraints
Prompt: “We have these constraints: [list time, budget, energy, child needs]. Suggest one realistic Saturday plan with timing, including buffer for transitions. Prefer low-stress options.”
The usefulness depends on how honestly you list the constraints. Optimistic constraint lists produce optimistic plans that fail by 11 a.m.
How to Keep a Prompt Library Useful Over Time
Record the exact wording that worked
Small changes in phrasing often matter more than people expect. Save the version that produced clean output, not the approximate idea of the prompt.
Note the failure mode beside each prompt
If the model repeatedly adds extra sections or drops a constraint, write that down. The next revision of the prompt becomes faster and more reliable.
Re-test after any noticeable model change
A prompt that was stable for weeks can drift when the underlying system is updated. A five-minute clean run with a familiar input reveals the shift before it affects real work.
Retire prompts that need constant repair
If you are editing the output heavily every time, the prompt is no longer saving effort. Archive it and replace it with a tighter version or a different workflow.
These fifteen prompts do not cover every situation. They do cover the recurring tasks that otherwise fragment attention. Used with ordinary review habits, they reduce friction without creating new dependencies. The clean-run standard still applies: if a prompt stops producing reliable results under normal conditions, it leaves the active list.
