World Model Readiness
Engraved personal-practice instrument

For you

Personal Practice

Module · Repeatable craft or daily improvisation

The Prompt Craft Check

Most people improvise every prompt from scratch, which means they solve the same problem again every morning. Craft is the opposite: it accumulates. This check looks at the five habits that turn prompting from improvisation into a skill that compounds: whether you reuse what works, whether you iterate on purpose, whether you give the tool what it needs, whether you control the shape of the answer, and whether you know when not to prompt at all.

Question 1 of 5 · You reuse what works

When a prompt works well, do you keep it, or type it fresh next time?

A prompt that worked is a small tool you built. If you retype it from memory each time, you rebuild it worse every day. A saved library, even a scruffy notes file, is where prompting starts to compound.

Question 2 of 5 · You iterate on purpose

When the first answer is not right, do you refine the prompt or just try again?

Retrying the same prompt hoping for luck is not iteration. Craft is reading why the answer missed and changing one thing: adding a constraint, an example, a correction. The second and third pass is where the quality lives.

Question 3 of 5 · You give it context

Do you give the tool the context it needs, or expect it to guess?

The model knows nothing about your audience, your constraints, your prior work unless you tell it. Most weak outputs are not the model failing, they are a prompt that withheld what a competent human would have needed too.

Question 4 of 5 · You control the format

Do you tell the tool what shape the answer should take?

Left to itself the model returns an essay. Craft is asking for the form you actually need: a table, five bullets, a diff, a draft in your voice. Controlling the output format is often the difference between usable and something you have to rewrite.

Question 5 of 5 · You know when not to

Do you know which tasks are not worth prompting at all?

Not everything should go through AI. Some tasks are faster by hand, some are too high-stakes to delegate, some need a judgement only you can make. Knowing when to close the tab is as much a part of the craft as knowing how to prompt.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

What does your collection of reusable prompts look like?

I do not keep any
A few in my head
Scattered notes
An organised library
Shared with my team

On a typical task, how many times do you refine before you accept the answer?

I take the first
Once or twice
Three to five
As many as it takes
Depends on the stakes

What do you most often use AI for?

Writing and editing
Coding
Research and analysis
Planning and thinking
Routine admin

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

You reuse what works

  1. 1Retype every time
  2. 2A few in memory
  3. 3Saved somewhere messy
  4. 4A real library
  5. 5Library, refined over time

At the low end: Retyping your best prompts is like rewriting a spreadsheet formula every time you need it. Start one notes file today and paste in the next prompt that works well; that file is your first tool. What good looks like: A refined prompt library is a genuine productivity asset. Prune it as the models improve; prompts that were necessary a year ago are often just clutter now.

You iterate on purpose

  1. 1Accept the first
  2. 2Reroll and hope
  3. 3Rephrase randomly
  4. 4Refine deliberately
  5. 5Diagnose then adjust

At the low end: Taking the first answer means you are seeing the tool at its worst, because the first draft is rarely its best. Build one habit: when an answer misses, tell it exactly what was wrong and ask again. What good looks like: Diagnosing the miss and adjusting deliberately is the core of the craft. Notice which fixes recur; those belong in your library as defaults so you stop rediscovering them.

You give it context

  1. 1One-line prompts
  2. 2Minimal context
  3. 3Context when I remember
  4. 4Context by default
  5. 5Right context, every time

At the low end: A one-line prompt asks the tool to guess everything that matters, and it guesses average. Before you ask, add who it is for, what it is for, and any constraint; that single change lifts most outputs. What good looks like: Supplying the right context by default is what separates craftsmen from dabblers. Bank your standing context (your role, your style, your constraints) so you are not retyping it each time.

You control the format

  1. 1Take what comes
  2. 2Reshape by hand
  3. 3Ask sometimes
  4. 4Specify the format
  5. 5Format plus voice, precise

At the low end: Accepting whatever shape it returns means reformatting by hand afterwards, which throws away the time you just saved. Tell it the format you want: a table, a list, a length; ask up front. What good looks like: Controlling both format and voice means the output lands ready to use. Save your best format instructions as snippets; a precise 'answer as X' is reusable across a hundred prompts.

You know when not to

  1. 1Prompt everything
  2. 2Rarely question it
  3. 3Sometimes step back
  4. 4Choose deliberately
  5. 5Sharp instinct for when

At the low end: Routing everything through AI wastes time on tasks that were faster by hand and risks the ones that needed your judgement. Ask one question before each prompt: is this actually the right tool for this? What good looks like: A sharp instinct for when not to prompt is the most underrated part of the craft. It is what keeps AI a lever rather than a detour; protect the judgement calls that are yours to make.