07/24/2026
here are some high-impact LLM prompting pro tips:
Be specific about the output format: “Return JSON with keys: …” / “Use a table with columns …”.
Define the role and task in the first lines: “You are a legal editor. Rewrite for clarity…”.
Provide constraints: length limit, tone, must/avoid items (“Do not mention X”, “Include 3 examples”).
Give a rubric (how it should decide): “Prioritize accuracy over brevity. If uncertain, say so.”
Use examples (few-shot) when the style is important: show 1–3 good input→output pairs.
Ask for assumptions when needed: “If info is missing, list assumptions before answering.”
Break complex tasks into steps: “First outline. Then draft. Then finalize.”
Use delimiters for source text: wrap content in """...""" so the model doesn’t mix it with instructions.
Avoid ambiguous pronouns: replace “it/they” with the actual nouns.
Control verbosity: “Answer in 5 bullets” / “One paragraph only”.
Request uncertainty handling: “If you’re not sure, provide best-effort with confidence level 0–1.”
For coding: specify language, libraries, and whether to include tests; say “Explain briefly after code”.
Iterate with targeted edits: change one thing in the prompt at a time and re-test.
