AI prompt checklist

A model answers the request you wrote, not the one you meant. Run these eight checks before you send a request that matters. Each one closes a gap the model would otherwise fill with a guess.

The checklist

1. 1

   Name the reader

   Say who will read the output and what they already know about the subject.

2. 2

   State the purpose

   Say what the output is for: a decision, an update, a first draft, or a reply.

3. 3

   Fix the shape

   Give the format, the length, and the sections. Name the columns if you want a table.

4. 4

   List the hard limits

   Write down the dates, figures, names, and rules the output must respect.

5. 5

   Swap adjectives for an example

   Replace words like “professional” with one sentence written in the voice you want.

6. 6

   Say what to leave out

   Name the topics, claims, or details the output must not include.

7. 7

   Say what to do with gaps

   Tell the model to mark missing information as missing instead of inventing it.

8. 8

   Give feedback that names the fault

   When a draft misses, point to the exact sentence and show what right looks like.

Red flags

• Your request would fit on a sticky note, and the task would not.
• You asked the model to “make it better” and got the same draft back with new words.
• Every output needs the same edit, and the request never mentions it.
• You are on the fourth rewrite of a long request without knowing which instruction fails.

At each band

E1 · Foundational

States the reader, purpose, format, and limits instead of a one-line ask.

E2 · Proficient

Fixes a weak draft in one or two rounds with precise feedback and an example.

E3 · Distinguished

Isolates the one instruction that fails and keeps a library of requests that work.

E4 · Exceptional

Writes standing instructions a whole team can reuse, including what to do when inputs are missing.

Summarized from the Emeri study guide Asking well, free at emeri.org/learn/asking-well.