Anatomy of a Great Prompt
About this resource
This takes one genuinely strong prompt and pulls it apart, labelling every component and explaining what each one contributes. Where 6 Techniques shows before-and-after pairs and CRAFT gives you a mnemonic, this shows the anatomy of a prompt as a single annotated specimen, which is a different way of learning the same territory and suits people who need to see the whole before the parts.
The labelled components cover role, context, task, constraints, format, examples, and the success criterion. The second half is a build-your-own section with the same labels blank, so you construct one for a task you actually have. The point of the annotated format is that you can see how much is missing from a typical prompt at a glance rather than being told.
A typical workplace prompt contains two of them. Seeing all seven labelled on one specimen makes the omissions obvious in a way that a list of tips does not.
How to use it
Take a prompt you wrote this week and hold it against the labelled specimen. Most workplace prompts contain the task and nothing else, and seeing the six missing components labelled beside your own is more persuasive than any amount of advice about being specific. Then use the blank half to rebuild that same prompt properly.
Inside the download
Role, context, task, constraints, format, examples, and success criterion, labelled on one specimen.
Compare a prompt you wrote this week against the labelled version.
Not just that a component belongs, but what changes when it is present.
Why context before task and constraints before format produces better results.
Common questions
Role, context, task, constraints, output format, worked examples, and a success criterion. Not every prompt needs all seven, but most workplace prompts contain only the task, which is why the results come back generic.
Context before the task, so the model interprets the request correctly. Constraints and format after the task, so they are the last thing read. Examples near the end, and for long supplied material, put the instruction after the content rather than before it.
Long enough to contain what the model cannot infer, and no longer. Most disappointing prompts are too short rather than too long. Length itself is not the goal: a prompt is finished when a competent stranger could carry out the request from it without asking you a question.
Free. No email list, no follow-up sequence.
More from the resource library, chosen because they pair well with this one.
Six before-and-after prompt pairs that show exactly what changes when a vague request becomes a specific one.
Eight reusable prompt structures that transfer to work you have not encountered yet.
Tokens, context windows, training cutoffs, and why these systems produce confident wrong answers. One page, kept beside a monitor.
Tell us a little about yourself and this document is yours. We don't have an email list. We collect this so we know who's finding our resources useful, and that's it.
You're downloading
this document
Your information is kept confidential and never shared with anyone.
