ChatGPT prompt writing tips

Help me design a prompt that consistently gets useful results for this task: [TASK TO IMPROVE]. I will use [AI TOOL] and the output will be used by [INTENDED USER OR AUDIENCE]. Here is an example of a weak result or current prompt: [CURRENT PROMPT OR BAD OUTPUT]. My available source material and non-negotiable constraints are: [SOURCES AND CONSTRAINTS].

First diagnose the failure in the current prompt or output. Identify which of these are missing or underspecified: role or expertise, task boundary, source inputs, output structure, evaluation criteria, uncertainty handling, audience, and constraints. Do not claim a capability the tool may not have, such as live web access, file access, or guaranteed factual accuracy.

Then produce:
1. A short diagnosis with the three highest-impact changes.

2. A reusable prompt of 180-260 words, written in first person, with square-bracket placeholders for inputs I must supply.
3. A compact input checklist explaining what each placeholder needs.
4. Two realistic example prompts for the same task: one quick version and one high-stakes version.
5. A “quality gate” of 4-6 checks I can ask the AI to perform before it answers.
6. Three follow-up messages for improving an initial answer without restarting.

Make the prompt operational: specify a deliverable, audience, format, decision criteria, and how to label missing evidence. Avoid empty instructions such as “make it engaging” unless you turn them into observable standards. Before answering, check that every recommendation relates to my stated task and that the proposed prompt does not request invented facts, citations, calculations, or credentials. Ask up to 3 clarifying questions only if a required input is missing.

Audit this prompt for [TASK]: [EXISTING PROMPT]. The intended reader is [AUDIENCE], and the model may use only [ALLOWED SOURCES]. Return a line-by-line critique identifying ambiguity, missing inputs, conflicting instructions, unsupported assumptions, and output requirements that cannot be checked. Then provide a revised version that preserves my intent and uses clear placeholders. Include a short test input and the acceptance criteria I should use to judge the answer. Do not add a persona unless it changes domain decisions. Check that every edit fixes an identified issue.

Create a team prompt template for [REPEATED TASK], used by [TEAM] for [AUDIENCE]. Inputs currently available are [INPUTS], and the deliverable must follow [OUTPUT STANDARD]. Produce: a copy-ready master prompt, an input form with required versus optional fields, examples of a complete and incomplete submission, and a reviewer checklist. Include instructions for handling missing evidence and confidential information. Avoid steps that depend on undocumented company knowledge. Check that two people using the template would produce outputs with the same required sections.

Build a small test set to evaluate prompts for [TASK]. The candidate prompt is [CANDIDATE PROMPT], and success means [SUCCESS CRITERIA]. Create five representative test cases covering normal input, incomplete input, conflicting source material, edge cases, and a request that should be declined or escalated. For each, provide the input, expected output characteristics, failure signals, and a scoring rubric from 0-2 per criterion. Do not invent proprietary facts; use neutral fictional examples. Check that every case measures a stated success criterion rather than writing style alone.