🫙 The Pantry

Place this at the start of any prompt where expertise matters.

Act as a [specific expert role — e.g., "senior copywriter", "experienced GP", "startup CFO"].

[Your actual request follows here]

Why it works: Framing the AI as a specific expert shifts the vocabulary, depth of analysis, and the frameworks it uses to respond. The more specific the role, the more targeted the output.

Examples:

• “Act as a venture capitalist who has reviewed 500 pitches”
• “Act as a UX designer who is opinionated about conversion”
• “Act as a copy editor with no tolerance for passive voice”

Drop this in front of any idea, plan, or decision you’re about to act on.

Play devil's advocate. Your job is to make the strongest possible case against what I'm about to share — not to be balanced, not to list concerns, but to steelman the opposition as convincingly as you can.

[Your idea, plan, or decision]

After making the case against it, tell me: what would need to be true for the opposing view to be wrong?

Why it works: AI tends to validate what you show it. Explicitly assigning the opposing role forces it to find weaknesses it would otherwise gloss over. The final question (“what would need to be true…”) turns the exercise into something actionable.

Use when:

• You’re about to make a significant decision
• You want to pressure-test a strategy before presenting it
• You’re too close to something to see its flaws
• You need to prepare for pushback

Ask for comparison or structured data in table format.

Present this as a table with columns: [Column 1] | [Column 2] | [Column 3]

Why it works: Tables make comparisons scannable and force clarity about categories. Great for comparing options, listing pros/cons, or showing structured data.

Best for:

• Comparing tools, options, or approaches
• Feature matrices
• Pros/cons breakdowns
• Showing data relationships

Tip: Specify your exact column headers to get exactly what you need.

Use this format template for anything going to a decision-maker.

Structure the response as:

**Bottom line:** [1 sentence — the conclusion or recommendation]
**Why it matters:** [2–3 sentences — context and stakes]
**Key points:** [3–5 bullets — the essential detail]
**What I'd do:** [1–2 sentences — a clear recommendation if applicable]

Why it works: Busy readers scan before they read. Leading with the conclusion means they get the most important thing even if they stop there. Everything after supports the lead.

When to use it:

• Status updates to managers or clients
• Recommendations to senior stakeholders
• Anything that might get forwarded without explanation
• Briefing documents

Tip: Add “senior executive who skims” to your audience description and this format becomes even tighter.

Set this up at the start of any complex request where context matters.

Before responding, here's the context you need:

Background: [who you are, what you're working on]
Audience: [who this is for]
Constraints: [time, budget, skill level, tools available]
What good looks like: [what you're hoping for]
What to avoid: [specific things that don't fit]

Why it works: AI gives much better answers when it has complete context. This template ensures you haven’t left out anything important before making your request.

Paste this before any writing task where the audience matters.

My audience:
- Who they are: [job title, background, relationship to you]
- What they already know: [their level of familiarity with this topic]
- What they care about: [their priorities, goals, what keeps them up at night]
- What they're skeptical of: [what they push back on, what sounds like noise to them]
- How they'll read this: [skimming? reading carefully? on mobile?]

Why it works: AI writes for a generic reader unless you tell it otherwise. Defining the audience’s existing knowledge stops it from over-explaining basics to experts or under-explaining to beginners. Defining their skepticism stops it writing things that will get ignored.

Quick version for simple tasks:

Audience: [one sentence describing who they are and what they care most about]

Add this phrase to any prompt requiring careful reasoning.

Think through this step by step before giving me your answer.

Or the longer version:

Before answering, think through:
1. [First aspect to consider]
2. [Second aspect]
3. [Third aspect]

Then give me your conclusion.

Why it works: Chain-of-thought prompting forces the AI to reason through problems rather than pattern-match to a fast answer. This dramatically improves accuracy on complex, multi-step problems.

Add this preamble to slow down fast, shallow answers.

Before you answer, I want you to:
1. State what you know with high confidence about this
2. State what you're less certain about
3. Flag anything I should verify independently
4. Then give me your answer

[Your question]

Why it works: AI defaults to confident-sounding answers even when it’s guessing. Making it explicitly separate confidence levels forces it to surface uncertainty instead of hiding it in fluent prose. You get the answer AND a built-in reliability check.

Shorter version:

Before answering: what's the thing you're least certain about in this question?

[Your question]

When it matters most: medical, legal, financial, or technical questions where a wrong answer has real consequences.

Add this constraint to any prompt where length matters.

Maximum [X] words. [Or: "Keep this to [X] sentences / bullet points / paragraphs"]

Why it works: Without length constraints, AI will often produce comprehensive answers when you need a crisp one. A word limit forces prioritization.

Common length targets:

• Tweet: 280 characters
• Email subject line: 6–9 words
• Quick summary: 50–100 words
• Executive summary: 150–200 words
• Standard summary: 300–500 words

Add this to any prompt where plain language matters.

Write in plain language — no jargon, no acronyms without explanation, no technical terms unless you explain them immediately.

Or more specific:

Avoid: [specific terms to avoid]
If you must use a technical term, explain it in plain language immediately after.

Why it works: AI defaults to domain vocabulary because that’s how most of its training data is written. Explicitly banning jargon forces it to find plain equivalents.

Add this when the default AI tone is too formal or stiff.

Tone: casual and direct. Write like a knowledgeable friend explaining something — not a consultant writing a report. Short sentences. No throat-clearing. No "It's worth noting that..." or "In conclusion..."

Why it works: AI defaults to a formal, hedged, corporate register because that’s what most of its training data looks like. Explicitly specifying casual-direct gives you prose that actually sounds like a person wrote it.

Markers of bad AI tone to watch for:

• “It’s important to note…”
• “In today’s fast-paced world…”
• “Let’s dive in!”
• Excessive hedging (“it could be argued that…”)
• Starting every paragraph with a transition

Variations:

Tone: sharp and opinionated — take a clear position, don't hedge.

Tone: warm and encouraging — like a mentor, not a critic.

Add this when you need writing that can withstand scrutiny.

Tone: formal and precise. Use complete sentences, avoid contractions, and define technical terms on first use. Prefer specific language over general claims — say "23% increase" not "significant increase." Do not editorialise.

Why it works: “Formal” alone still produces vague output. The key addition is precise — forcing specificity stops the AI from substituting confident-sounding adjectives for actual detail.

Good for:

• Legal or compliance documents
• Academic or research writing
• Board-level reports
• Job descriptions or HR documents
• Anything that will be quoted or cited

Add this for extra rigour:

Avoid: subjective adjectives (great, significant, robust) unless you can back them with data. If you can't quantify a claim, say so.