Prompting That Actually Works

Imagine you hired a brilliant freelancer who has read the entire internet but knows nothing about you, your company, your customers, or your taste. You wouldn't email them "write me a blog post" — you'd send a brief. Every good prompt is a brief with up to five components, and when output disappoints it's almost always because one of these is missing: **Role** (who should the AI be), **Context** (what it needs to know — your company, audience, goal, constraints), **Task** (exactly what to do, with a specific verb and deliverable), **Format** (length, structure, tone, a template to follow), and **Examples** (what "good" looks like — paste 1–3 samples).

Compare a weak prompt — "Write a LinkedIn post about AI" — to a strong one that specifies role (a growth marketer), context (writing for B2B SaaS founders), task (argue companies buy AI before fixing process), format (structure, word count, banned words), and an example post to match in style. The strong prompt takes 90 seconds to write and saves 20 minutes of editing.

Of the five components, context moves quality the most. Modern models can infer role and format reasonably well, but they cannot invent facts about your business. Practical ways to inject context: paste it directly (your website copy, proposals, meeting notes handle long pastes well); upload files (PDFs, spreadsheets, decks — both ChatGPT and Claude read them directly); use Projects, which let you create standing instructions and reference files that apply to every chat inside them, so you stop re-explaining your business every time; and let the model ask — ending complex prompts with "before you start, ask me up to 5 questions that would meaningfully improve the output" routinely doubles output quality.

Amateurs treat the first output as the final answer. Operators treat it as a first draft from a junior colleague. Your follow-up vocabulary: **Tighten** ("cut this by 40% without losing the argument"), **Redirect** ("good structure, wrong tone — make it warmer"), **Deepen** ("expand point 2 with a concrete example"), **Variant** ("give me 3 alternative hooks, ranked"), **Critique** ("act as a skeptical customer, tear this apart, then rewrite addressing your own critique"), and **Extract** ("turn this into a reusable template with placeholders").

That last move is important: whenever a conversation produces something great, spend one more prompt converting it into a reusable template or checklist. That's how one-off wins become a system.

**Meta-prompting**: ask the AI to write the prompt — "I want [outcome]. Write the ideal prompt I should give you, ask me for missing information, then execute it." This works because the model knows what information it needs better than you do.

**Chain-of-thought for hard problems**: for analysis, planning, or math, say "think through this step by step" or switch on extended thinking mode. **Structured output**: ask for a table or valid JSON when the output needs to be machine-usable — essential for automation pipelines. **Negative constraints**: models respond well to explicit bans, like "do not use bullet points" or "do not invent statistics." **Style transplants**: paste 2–3 samples of your writing and ask the model to describe your style in 10 bullets, then apply those rules to everything it writes.

You don't need to be creative at 9am on a Tuesday. You need a document with your 10–20 proven prompts, ready to paste.

The course's Appendix A gives a starter library of 25 prompts across email, research, content, analysis, and planning — treat it as seed stock and evolve your own. Store your library where you actually work: a Notion page, a text file, or directly inside Project instructions so you never have to paste them at all.