Create a quality-control checklist for AI output on a recurring task

Create a reusable quality-control checklist for AI output on a task my team does again and again. The recurring task: [e.g. "AI-drafted customer replies", "AI summaries of sales calls", "AI first drafts of job descriptions"] What a great output looks like: [DESCRIBE, or PASTE a strong example] Common problems we've seen: [e.g. invented details, wrong tone, missing steps, too long] Who reviews and how much time they have: [DETAILS] Stakes if something slips through: [DETAILS] Build the checklist in these groups, each with yes/no checks a reviewer can run quickly: 1. Accuracy - facts, names, numbers, dates, and links match the source. 2. Completeness - every required element is present. 3. Instructions followed - format, length, structure, and constraints. 4. Audience fit - tone, reading level, brand voice, no jargon the reader won't know. 5. Risk - no confidential data, no promises we can't keep, no claims that need approval. 6. Telltale AI problems - generic filler, overconfident statements, made-up specifics. Return: 1. The checklist (12-20 checks), each with a pass/fail definition. 2. Which checks are "must pass" (stop and fix) vs. "should pass." 3. A spot-check rule: how many outputs to review in full and when to increase reviews. 4. A simple log format to track failures, so we can fix the prompt instead of fixing the same error repeatedly. 5. A one-page version for reviewers.