Home / Stage 01 · Tool
01/03

AI as a Tool.
Prompt well. Use it daily.

Stage One of Three
2–4 weeks to fluency
i

AI is a brilliant intern with a PhD and no context. Set the role, give the situation, define what good looks like, then iterate. Never paste anything you wouldn't email to a stranger.

A · Anatomy

The 4-Part Prompt

5 min read
Framework
Every good prompt has these four.
Miss one and the output is mediocre. Use them as section labels in your prompt for the first 10 tries.
01 · ROLE Who should it be? "You are a..." forces a stance and a vocabulary. "You are a sharp board chair who has seen 50 IPOs."
02 · CONTEXT What's the situation? Audience, stakes, constraints, what you tried. Most of the value lives here. "Midmarket SaaS, $80M ARR, considering a sponsor sale next year."
03 · TASK What exactly? Specific verbs: draft, summarize, critique, score, structure. Avoid "help me with." "Critique this strategy and identify the top 3 risks."
04 · FORMAT What deliverable? Length, structure, tone. Specifying this saves a follow-up every time. "5 bullets, plain English, no jargon, under 200 words."
2026 Landscape
Pick your tool
ToolBest for
ClaudeWriting, reasoning, long docs (1M+ context), careful tone
ChatGPTAll-rounder, voice mode, broad plugins, strong research
GeminiNative in Google Workspace, sees Drive/Docs natively
CopilotNative in M365 (Outlook, Word, Excel, Teams)
PerplexityLive web research with citations, fast lookups
B · Prescriptive

See one. Do one.

10 min to try
A real before/after, then your turn
SEE A vague prompt vs. the 4-part version

Same task. Same model. Wildly different outputs.

Vague (most people stop here) help me write an email to my team about the new strategy

Output: 4 paragraphs of generic corporate-speak. Could have been written about any company.

Sharp (all 4 parts) You are a midmarket CEO writing to a leadership team of 12. Context: we're shifting from product-led to sales-led for the enterprise segment over 2 quarters. Stakes: some product leaders feel deprioritized. Task: draft a 200-word email that names the change, acknowledges the discomfort directly, asks 1 specific question to surface concerns. Format: plain English, no buzzwords, conversational tone.

Why it works: Role plus context plus specific task plus format. Same model, dramatically better output.

DO Your turn, right now

Build your first sharp prompt in 4 steps.

  1. Pick a real task you'd push to Friday afternoon. An email, a summary, a critique. Something that usually takes 30 minutes.
  2. Write the vague version first. One sentence, the way you'd describe it to a colleague at the coffee machine.
  3. Layer in the 4 parts. Add role plus context plus task plus format. Get specific where you'd normally generalize.
  4. Run it. Then iterate. If the output is off, your prompt is off. Edit the prompt, not the output. Two rounds usually does it.

First time it'll feel like overkill. By the 5th time it's faster than writing the task yourself.

C · Patterns

5 patterns worth memorizing

Copy these. Adapt them. Use them weekly. Click "Copy" on any card to grab the prompt.

The Critic

Kill My Idea

PROMPT You are a skeptical board member who has seen this play before. Here's my draft strategy. Give me the 3 sharpest objections, the 3 strongest reasons this fails in 18 months, and the specific evidence that would change your mind. Push hard. Don't be polite.
Use for
Strategy memos, pitch decks, M&A theses, board narratives.
Upgrade
Re-run as customer, then competitor, then your CFO. Three lenses.
The Translator

Audience Shift

PROMPT Take this document and rewrite it for [audience: e.g., a board, a frontline team, a 12-year-old, a CFO who hates buzzwords]. Keep the substance. Change the vocabulary, the level of detail, and the assumed knowledge. Output the same length or shorter.
Use for
Same content, multiple audiences. CEO note → all-hands → press release.
Upgrade
Paste 3 examples of how your CEO writes for tone-matching.
The Synthesizer

Raw → Structured

PROMPT Below is [raw input: call notes, survey responses, customer feedback]. Extract: the top 3 themes, the strongest single quote per theme, what's not being said, and the 5 sharpest questions I should ask next.
Use for
Customer calls, exit interviews, surveys, focus groups, town halls.
Upgrade
Add "Flag anything you're inferring vs. quoting directly."
The Drafter

Notes → Polished

PROMPT Here are my rough notes from [meeting/call]. Draft the [email/memo/post] I'd send. My voice: direct, plain English, no exclamation points, sentence fragments OK. Target length: [X words]. Don't add facts that aren't in the notes.
Use for
Follow-ups, board pre-reads, executive summaries, internal posts.
Upgrade
Paste 3 past examples of your writing. It learns your voice fast.
The Coach

Pressure Test Me

PROMPT I'm about to [make this decision / have this conversation / send this email]. Ask me the 5 hardest questions a smart skeptic would ask before I do it. Don't answer them for me. After I respond, push back on my answers.
Use for
Decision pre-mortems, tough 1:1s, board prep, public statements.
Upgrade
Voice mode in Claude or ChatGPT. Talk it out on a walk.
The Structurer

Mess → Frame

PROMPT I have a messy situation: [paste 2-3 paragraphs of facts, tensions, options]. Propose 2 different frameworks for thinking about this, with pros/cons of each. Then recommend which one fits this specific situation and why.
Use for
Org redesigns, tough hires, pricing changes, partnership decisions.
Upgrade
Force a 2x2. Force a decision tree. Force a force-field analysis.
D · Discipline

What to never do.

Read once, remember always
Data Hygiene · Non-negotiable
Never paste into a consumer AI
  • Customer PII (SSN, DOB, account numbers, health data)
  • Privileged legal docs unless on a private/enterprise instance
  • Identifiable comp or salary data tied to a person
  • Source code from licensed vendors
  • Materials covered by NDA to a third party
  • Pre-public M&A, earnings, or material non-public info
  • Customer data without a DPA in place
  • Anything you wouldn't put in #general on Slack

The safe path: get your company on an enterprise plan (Claude Enterprise, ChatGPT Enterprise, Gemini for Workspace) where your data isn't used for training. Until then, redact aggressively.

Common mistakes
And how to fix them
Click any row to expand.

Don't search. Converse. Iterate. The 4th message in a thread is usually where the value lands, not the 1st. One-shot prompts get one-shot answers.

  • Run a follow-up after every output: "what would you do differently?" or "what's the weakest part of this answer?"

First drafts are 60% there. Your edit makes it 100%. Always edit. If you find yourself shipping the raw output, you're not using AI well; you're using it lazily.

When the output is off, hit the up arrow on your last message and edit it instead of writing "no, do it like X." Saves 10–20x tokens and produces cleaner outputs on long threads.

Force a specific verb: draft, critique, score, summarize, restructure, list, compare. "Help me with the strategy memo" becomes "Critique my strategy memo against these 3 frameworks."

"For a board meeting Friday with our PE sponsor" changes everything about the output. Stakes are context. Context is fuel.

Do this

Try this quarter

  1. Replace your Google searches. For 2 weeks, when you'd Google something complex, open Claude/GPT/Gemini instead. Feel the gap close.
  2. Run "kill my idea" weekly. Pick one draft strategy, memo, or pitch and have AI red-team it before the human meeting.
  3. Build your prompt pack. Save your 5 sharpest prompts in a doc by end of quarter. That doc becomes your operating system.
Next up · Stage 02

AI as a Teammate

Once prompting feels natural, the next move is to stop re-typing context every morning. Build helpers that know your job and get sharper weekly.

Continue →