The ten checks
Ten questions to ask of any prompt before you send it, shown on one request: grouping customer feedback for a planning meeting.
The questions, with a weak and a fixed phrase
| Check | Ask of the prompt | Weak | Fixed |
|---|---|---|---|
| 1. Task | Does it say exactly what to produce or do? | Can you look at this feedback? | Group the feedback below into themes and summarize each one. |
| 2. Purpose | Does it say who or what the result is for? | No reader named | It is for the planning meeting that picks next quarter’s two fixes. |
| 3. Material | Does it give the facts the model cannot guess? | Feedback about our new release. | Comments from the two weeks after version 4.2 moved the Export button. |
| 4. Scope | Does it set a size: length, count or depth? | List the main themes. | The five most common themes, two sentences each. |
| 5. Form | Does it name the format or structure? | No format named | A table: theme, number of comments, one quote. |
| 6. Limits | Does it say what to leave out? | Nothing excluded | Leave out pricing comments, which finance is reviewing, and customer names. |
| 7. Standard | Does it say what a good result looks like? | Make it clear and useful. | Someone who has not read the feedback can rank the problems. |
| 8. Reference | Does it give an example to match? | No example | Example row: Export is hard to find, 14, “I spent ten minutes looking.” |
| 9. Doubt | Does it say what to do when something is unclear or missing? | Nothing said about gaps | Put comments too vague to assign under “Unclear” instead of guessing. |
| 10. Fit | Is it laid out the way the target model reads best? | One paragraph, data in the middle | Material first, then one labeled line per part. For Claude, XML tags. |
The ten levels
A prompt’s level is the number of checks it passes, from 1 to 10. Low levels leave the AI guessing; at 10 it has nothing left to guess. A prompt that passes none still reads 1.
- Guess
- Hunch
- Notion
- Idea
- Outline
- Plan
- Brief
- Spec
- Blueprint
- Superintelligent
Brief, at 7, is good enough for most work. Aim higher when a prompt runs many times or other people rely on it. The scale is an order, not a measurement: a 9 means nine questions answered, not 90% good.
How to use it
- A check passes only when the prompt itself says it. Anything the model would have to guess counts as a fail.
- Fix the cheapest failures first: a precise verb, a reader and a size.
- Not every check matters for every request. A quick one-off question is fine at Guess.
- A blank such as {{audience}} counts as supplied, because you fill it in before you send.
- For a free score as you type, paste the prompt into the Grader at thinkhard.si/grader/. The full guide is at thinkhard.si/guides/ten-tests/.