Steps to build your own "unslop" command

- Identify your sloppiest documents: Start with internal docs that get reused: brand kits, briefs, system prompts, process docs. These compound slop into every output they feed. - Have Claude or ChatGPT access ready: Metz built his command in Claude, but the principles work in any LLM - Gather 2-3 sample documents: You'll need real examples to test your command against

- Decide where AI fits by content type: Not everything should involve AI. Metz's breakdown at Animalz: minimal AI for high-end thought leadership, extensive AI for LinkedIn content, significant AI for AEO audits - Create or update your brand kit: Give the AI access to voice guidelines, strategic positioning, and example content. Without this context, outputs will be generic. - Set up feedback loops: Store AI outputs, track what works, and feed learnings back into your prompts and brand kits

- Write the MECE rule: "Each section should address a distinct topic with no overlap. Together, all sections should fully cover the subject." - Write the DRY rule: "If something has been said once, don't say it again. Eliminate anything that just rephrases earlier content." - Write the Essential rule: "Make things as simple as possible, but not simpler. Cut filler, but don't remove necessary context or examples." - Combine into a single prompt: See the prompt template asset for the full command text - Add output instructions: Tell the model to: (1) identify sections to cut with justification, (2) recommend sections to relocate, (3) rewrite the document

- Run the command on your sample documents: Compare before and after versions side by side - Check for over-cutting: If the output "reads like shorthand," the command is too aggressive. Restore context. - Check for example removal: AI often flags examples as unnecessary, but Metz notes "they're actually very important in the context of content creation" - Run standardized tests: Create a consistent test set. After each change to your command, re-run to confirm quality is improving, not regressing - Iterate on the command: Adjust the prompt based on what the model gets wrong. Be specific: "Don't cut examples" or "Preserve section headers"

- Unslop your brand kit first: Slop in reference docs compounds into every output. Clean inputs before cleaning outputs. - Run on internal docs: Emails, meeting notes, strategy memos, briefs. Metz calls this "workslop"—content that "seems valuable when you first read it, but then has no value." - Run on internal docs: Emails, meeting notes, strategy memos, briefs. Metz calls this "workslop"—content that "seems valuable when you first read it, but then has no value." - Keep humans in the loop: Writers and editors focus on strategic decisions and polishing, not mechanical editing. The unslop command handles the mechanical part. - Give feedback to the model: When you fix something the command missed, tell it why. "I wouldn't say it like that" trains the system better than silently rewriting.