Asking well

Turning what you need into instructions a model can act on.

Most disappointing AI output is not a model problem — it is an instruction problem. The model answered the question it was asked, which was vaguer than the question you meant. Asking well is the craft of closing that gap.

The skill is learnable, and it compounds: every other competency gets easier once your instructions carry the audience, the purpose, the format, and the constraints of the real task.

1. Stage 1 · E1 FOUNDATIONAL

   Say what you actually want

   Start every request with four ingredients: who the output is for, what it is for, what shape it should take, and any hard constraints. "Draft a 150-word status update for executives: project is one week behind because of vendor delays, new ETA March 4, mitigation in place. Confident tone, no jargon" beats "write a status update" every single time.

   Vague adjectives are the beginner's trap. "Make it professional" or "make it engaging" means something different to you than to the model. Replace adjectives with examples: one sentence written in the voice you want teaches more than five style adjectives.

   Try it this week

   Take the last prompt you wrote that disappointed you. Rewrite it with audience, purpose, format, and constraints stated explicitly, and compare the two outputs side by side.

2. Stage 2 · E2 PROFICIENT

   Iterate with precision

   When a draft misses, resist "try again" or "make it better." Name what is wrong specifically — "the second paragraph hedges; commit to the recommendation" — and show one example sentence of what right looks like. Precise feedback converges in one or two rounds; vague feedback wanders forever.

   Learn to control output shape as firmly as content. If you need a table, specify the columns. If you need it machine-readable, give the exact schema and one worked example of a correct output. Structure requests are the most reliably obeyed instructions there are.

   Try it this week

   Next time an output is 80% right, write one sentence of feedback naming exactly what reads wrong and one example sentence in the voice you want. Count how many rounds it takes versus your old approach.

3. Stage 3 · E3 DISTINGUISHED

   Debug your prompts like an engineer

   When a long prompt keeps failing on one requirement, do not rewrite the whole thing. Isolate a minimal version that reproduces the failure, fix that one instruction, then reassemble. Prompts are systems; debug them like systems.

   Build a small personal library of prompts that work — with the required inputs, expected output format, and an example of a good result saved alongside each. A prompt that took five rounds to perfect is an asset; treat it like one.

   Try it this week

   Take a recurring deliverable and write the full prompt you would hand a new colleague: context, format, constraints, and one exemplar output. Test it cold in a fresh conversation and refine until it works first try.

4. Stage 4 · E4 EXCEPTIONAL

   Design asks others can reuse

   At this level you are not just prompting — you are writing the standing instructions your whole team runs on. That means anticipating misreadings: state what to do when inputs are missing, what not to include, and how to handle the edge case that burned you last quarter.

   You also know when not to specify. Over-constrained prompts strangle the model's useful range on open-ended work. The exceptional operator states hard constraints firmly, leaves creative room deliberately, and knows which is which for the task at hand.

   Try it this week

   Turn your best personal prompt into a team template with slots for the variable parts, guidance on each slot, and two worked examples. Hand it to a colleague and revise wherever they stumble.

Ready to see where you stand? The free check scores all six competencies in about fifteen minutes.