Turn a vague task into a brief a model can act on.
From the guide: Brief an AI like a freelancer, and manage it like one → You are helping me brief a task to a capable freelancer who has never worked with me. The task: {task} Who the output is for: {audience} What they will do with it: {purpose} Write the brief in plain English, under 150 words. Include: the goal, the material I will attach, the format and length of the output, three things to leave out, and one example of what good looks like. If anything is missing from what I have told you, ask before writing.prompts
Prompt library
6 prompts you can copy. Each one comes from a guide that explains why it works.
Answer a customer thread in your own tone, from a pasted example.
From the guide: Writing with AI and still sounding like yourself → Here is a reply I wrote myself that I was happy with. Match its tone, length and structure — do not copy its content. My example reply: {example} Here is the thread to answer: {thread} Write one reply. Under {words} words. Plain English. No apologising unless I was actually late. Do not invent facts, dates or commitments that are not in the thread — if something is missing, leave a bracketed gap for me to fill.Extract owners, dates and open questions from raw notes.
From the guide: AI meeting notes are a draft, not a record → Below are raw notes from a meeting. Turn them into three lists and nothing else. Notes: {notes} 1. Decisions — what was actually decided, one line each. 2. Actions — owner, action, due date. If a note does not name an owner or a date, write "unassigned" or "no date"; do not guess. 3. Open questions — anything left unresolved. Attendees for reference: {attendees}. Do not add items that are not in the notes. If the notes contradict themselves, list both readings under Open questions.Describe the mess, get the formula and the steps.
From the guide: Spreadsheets and AI: let the sheet do the arithmetic → I have a spreadsheet in {app} and I want to fix this: {problem} Here are the first few rows exactly as they appear, including the headers: {sample_rows} Give me, in this order: (1) what is actually wrong with the data, (2) the formula or the steps to fix it, written for {app}, (3) how I check the fix worked on a row I can see. Do not tell me to use a script unless a formula genuinely cannot do it. If the sample rows are not enough to be sure, say what else you need.Ask the right questions before you sign up for a tool.
From the guide: AI vendor lock-in and the exit ramps that actually work → I am considering paying for {tool} to do this job: {job}. Before I sign up, write me the exit questions I should be able to answer, grouped as: my data (what goes in, how I get it out, in what format), my work (what I would have to rebuild elsewhere), my money (what the bill looks like at 3× my current volume), and my exposure (what breaks on the day this vendor disappears). For each question, say where I would look to answer it — a specific page, a setting, or a support request. Do not answer the questions for me and do not guess this vendor's policies.A second pass that checks an AI draft for invented facts.
From the guide: How to check AI work before it ships → Below is a draft. Do not improve it. Check it. Draft: {draft} List every factual claim in it — names, numbers, dates, quotes, prices, capabilities — one per line. For each, mark it: SUPPORTED if it appears in the material I attached, UNSUPPORTED if it does not, or UNCHECKABLE if it cannot be verified from what is here. Do not use your own knowledge to mark anything SUPPORTED. Finish with the single claim that would do the most damage if it were wrong.