Instruction Clarity, Positive Framing & Iterative Refinement

Instruction Clarity Beats Cleverness

Be specific about the desired output and format rather than relying on the model to infer intent from a clever or terse instruction. Ambiguity in the instruction produces variance in the output.

Positive Instructions Over Negative Lists

A positive instruction -- "respond only in JSON matching this schema" -- is more reliable than a long list of "don'ts" (don't add commentary, don't use markdown, don't include extra fields, ...). Positive framing tells the model what to do; negative lists only narrow what not to do, leaving the actual target under-specified.

Placement Matters

With long inputs, place the key instruction after the material, near the end of the prompt, rather than before it. This reduces the chance that the instruction is diluted by everything the model processes afterward.

Deliberate Iteration

Prompt improvement is empirical, not intuition-driven:

1. Change one thing at a time
2. Measure the change against an eval set
3. Keep what wins, discard what doesn't

A prompt change without a measured comparison is a guess, not an improvement.

Common exam traps

• Assuming a longer or more detailed instruction is automatically clearer -- clarity is about specificity and structure, not length.
• Iterating on multiple prompt changes at once, which makes it impossible to attribute a quality change to a specific edit.