State exactly what you want and what success looks like. Be specific about what you want the output to accomplish.
Template: "I want to [TASK] so that [SUCCESS CRITERIA]."
0 chars
2
Say Who It Is For and Why
Name the reader and what they already know. Then say what the output is for: what happens with it once it is written. A model that knows the reader and the purpose makes better choices in everything you did not specify.
Template: "The reader is [WHO], who already knows [WHAT]. It will be used to [PURPOSE]."
0 chars
0 chars
3
Describe Your Context & Files
Reference any files, documents, or background material you are uploading. Tell the model to read them completely before responding. Move long background knowledge out of the prompt and into files, which frees the prompt for instructions only. If you have no file to attach, paste the text into the second box.
e.g. "First, read these files completely before responding: [brand-guide.pdf], [previous-emails.docx]"
0 chars
0 chars
4
Show Reference Examples
Provide concrete examples of excellent output. Two or three that differ teach the pattern. A single example tends to be copied too closely, wording and all. Then say what to take from them: do not say "something like this"; name the tone, structure, and patterns you want.
Paste each example in its own box, or name the attached files. Then say what to copy from them.
0 chars
0 chars
0 chars
0 chars
5
Write the Brief
Specify the output format, length, and scope. Also describe what the output should not be. Avoiding the wrong direction is as useful as pointing toward the right one.
Include: output type · target length · what to avoid.
0 chars
6
State Rules, and the Reason for Each
List the hard rules for this task: legal or compliance limits, style mandates, things the model should always or never do here. Give the reason with each rule. A model that knows why follows the rule better, and applies it sensibly to cases you did not think of.
e.g. "Never mention competitors by name, because legal reviews every comparison. Use metric units, because the readers are in Europe."
0 chars
7
Say What Done Looks Like
Describe the finished result so the model can tell when it has it. Then tell it to check its own work against that description before answering.
e.g. "Done means: under 200 words, one call to action, every claim found in the brand guide. Before you answer, check the draft against each of these and fix what fails."
0 chars
8
Say What to Do When Unsure
Tell the model what to do when it does not know something, or when something it needs is missing. This is the main guard against invented facts.
0 chars
9
Request Clarifying Questions First
Instruct the model to ask questions before producing output. This surfaces assumptions, catches ambiguities, and results in far better first drafts.
Leave this off for a short task.
e.g. "Before writing anything, ask me up to 5 clarifying questions if you need them."
0 chars
10
Confirm the Plan Before Executing
Ask the model to restate the approach in its own words before starting. This is distinct from asking questions: it is a verification step that catches misunderstandings. Nothing should be executed until both sides agree on the plan.
Leave this off for a short task, or when the model will work by itself with nobody there to answer.
e.g. "Summarize your plan in 2 or 3 sentences and wait for my go-ahead before writing."
0 chars
Generated Prompt
–
0 of 7 steps filled0%
What would make this prompt stronger
✦
Fill in at least one step on the left to generate your prompt.
Your draft is kept in this browser only. Clear removes it.