This episode teaches the job of prompting, tool-agnostic across whatever the current ChatGPT, Claude, or Gemini happens to be. The central idea: a prompt is a brief, the same brief you'd hand a freelancer or a new junior hire. [Anthropic's docs](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/be-clear-and-direct) frame the model as "a brilliant but new employee who lacks context on your norms and workflows." Vague brief in, generic slop out. Specific brief in, on-brand draft out. The five-part anatomy: - **Audience** — exactly who reads it: role, knowledge level, priorities, funnel stage, objections. Paste your ideal customer profile (ICP) or buyer persona. [OpenAI](https://platform.openai.com/docs/guides/prompt-engineering) notes specifying role and audience gives the most relevant results. - **Goal** — one job-to-be-done, one action, plus success criteria. One goal per prompt. - **Format** — the container: structure, exact length, channel limits (email subject ~50 chars, Google ad headline 30, meta description ~155, X post 280). Tell it what to do, not what not to do. - **Voice** — how it sounds. Show, don't tell: paste examples, don't list adjectives. - **Constraints** — hard limits, must-includes, banned phrases, factual grounding (write [TK] rather than inventing numbers). The highest-leverage move is few-shot (multishot) prompting: 3-5 of your real past pieces that performed, labeled and diverse. Anthropic recommends examples that are relevant, diverse, and structured. The one pitfall: slop, the 2025 Word of the Year ([Merriam-Webster](https://www.merriam-webster.com/wordplay/word-of-the-year) and the American Dialect Society). With no constraints, the model returns the statistical middle of everything it's seen. The diagnostic: if you can swap your company name for a competitor's and the copy still fits, it's slop. Mnemonics covered (all repackagings of role + task + context + format + examples): RTF, RACE, CRISPE, RODES, TRIM. Copywriting frameworks for the copy itself: AIDA, PAS, BAB. Common mistakes and fixes, plus a reusable skeleton to save and reuse per channel. Sources: Anthropic and OpenAI prompt-engineering documentation; "slop" as 2025 Word of the Year. Vendor percentage stats are not stated as fact.