The iPrompt Library · 60 prompts · Free

AI Prompts That Actually Do the Job.

60 tested prompts for ChatGPT, Claude, and Gemini — organised by the thing you're trying to get done. No fluff, no filler, no "act as an expert" nonsense. Copy, paste, ship.

60 prompts · 10 categories · Works in ChatGPT, Claude & Gemini
No signup, no paywall

Curated by Rene Lauritsen · Co-founder, FrontWave Media ·

60+

Tested Prompts

Prompts that make your writing sharper.

For drafting, cutting, and pressure-testing any piece of writing — blog posts, essays, emails, reports.

Writing · Pre-publish

The Writing Pre-Mortem

Surface weaknesses in a draft before you hit publish.

Show prompt

You are a ruthless editor with 20 years at a top magazine.

Below is a draft I'm about to publish:
<<<PASTE DRAFT>>>

Assume it has failed — nobody shared it, nobody finished it, and the audience quietly disliked it. Work backwards from that failure.

Give me:
1. The 5 specific reasons it failed (be concrete; cite paragraphs).
2. The single sentence in the draft that is most likely to lose the reader.
3. One structural change that would have prevented the failure.

Do not rewrite anything. Diagnose only.

View page →ChatGPT · Claude · Gemini

PrinciplePre-mortems (Gary Klein) invert the question — "imagine it failed" produces sharper critique than "what could go wrong".

Writing · Critique

The Ghost Reader

Read your draft through a skeptic's eyes.

Show prompt

Read the draft below as a smart, skeptical reader who does NOT already agree with the author. You are not hostile — you are unconvinced.

<<<PASTE DRAFT>>>

Produce three columns:
- Claim — every claim the author makes.
- Your reaction — what a skeptic thinks when they read it.
- Missing — the evidence or reasoning the skeptic needs but didn't get.

Stop when you have the table. No summary at the end.

View page →ChatGPT · Claude · Gemini

PrincipleForcing model into a specific reader persona beats generic "give feedback" — role + stance = better signal.

Writing · Editing

The Reductionist

Cut your draft to 60% length without losing meaning.

Show prompt

Cut the following text to 60% of its current length. Rules:
- Preserve every concrete claim and number.
- Keep the author's voice.
- Kill: hedging ("somewhat", "perhaps"), throat-clearing ("it is worth noting"), duplicate ideas, and three-adjective runs.
- Never merge two paragraphs whose ideas aren't linked.

After the cut, list the 3 things you removed that you were least sure about, so I can decide.

<<<PASTE DRAFT>>>

View page →ChatGPT · Claude · Gemini

PrincipleConstraint-driven editing beats "make it shorter" — the rules stop the model from sanding off the interesting parts.

Writing · Hooks

The Hook Doctor

Generate and rank 10 opening lines for any piece.

Show prompt

I'm writing a piece about: <<<TOPIC + THESIS>>>
Audience: <<<WHO>>>
Tone: <<<VOICE>>>

Give me 10 opening lines in 5 different styles:
1. Concrete scene (a specific moment)
2. Surprising number
3. Contrarian claim
4. Question the reader is already asking
5. Overheard quote

Then rank all 10 by "stopping power" — how likely a distracted reader is to keep reading. Explain your top pick in one sentence.

View page →ChatGPT · Claude · Gemini

PrincipleAsking for variety across named styles produces more diverse outputs than "give me 10 hooks".

Writing · Clarity

The Ladder of Abstraction

Say the same thing at 5 altitudes — concrete to philosophical.

Show prompt

Take this idea: <<<PASTE IDEA OR SENTENCE>>>

Rewrite it at 5 levels of abstraction:
1. A specific thing I could point at.
2. A category of things.
3. A principle those things share.
4. The human value behind the principle.
5. A philosophical position.

At each level, keep it to 2 sentences max. The reader should feel they're climbing a ladder, not hopping between islands.

View page →ChatGPT · Claude · Gemini

PrincipleHayakawa's ladder forces you to see where your prose is stuck — most bad writing lives on one rung.

Writing · Structure

The Outline Inverter

Extract the structure of a finished piece — yours or a competitor's.

Show prompt

Here is a finished piece of writing:
<<<PASTE TEXT OR URL CONTENT>>>

Reverse-engineer it. Produce:
- The thesis in one sentence.
- The skeleton outline (sections and sub-sections).
- The underlying argument structure — claim, evidence, implication — for each section.
- The 3 moves the author made that carried the most weight.

Do not critique. This is forensic only.

View page →ChatGPT · Claude · Gemini

PrincipleThe best way to learn a format is to invert a finished piece — not to read "how to write X" advice.

02 — Marketing & Copywriting

Prompts that sell without sounding like it.

For positioning, landing pages, email subject lines, and finding angles your competitors don't own.

Marketing · Positioning

The VoC Translator

Turn raw customer quotes into positioning statements you can use.

Show prompt

Below are verbatim quotes from customer interviews / reviews / support tickets.
<<<PASTE 10–30 QUOTES>>>

Do four things:
1. Cluster them into 3–5 recurring jobs the customer is hiring the product for.
2. For each job, quote the most vivid line (word for word) that captures it.
3. Draft a positioning statement in the customer's language — no marketing-speak.
4. List any jobs you expected to find but didn't.

Prefer customer words over clever rewriting. If a quote is clunky, keep it clunky.

View page →ChatGPT · Claude · Gemini

PrincipleCustomers' actual words out-convert anything a copywriter invents — keep the clunk.

Marketing · Objections

The Objection Inventory

List every reason a prospect wouldn't buy — before they do.

Show prompt

Product: <<<ONE-LINE DESCRIPTION>>>
Target customer: <<<WHO>>>
Price: <<<$>>>
Main competitor / status quo: <<<NAME>>>

Write the 20 most likely objections a prospect will have, in the order they'll come up during a buying cycle. Group them into:
- Rational ("the numbers don't work")
- Emotional ("I don't want to be the person who bought this")
- Political ("my boss will kill me")
- Situational ("not right now")

For each, write the single best one-sentence reframe a salesperson could use.

View page →ChatGPT · Claude · Gemini

PrincipleObjection mapping (Blair Warren) forces the pipeline into view — you can't beat an objection you haven't named.

Marketing · Angles

The Angle Machine

Generate 10 positioning angles for a product, ranked by saturation.

Show prompt

Product: <<<DESCRIPTION>>>
Core benefit: <<<WHAT IT ACTUALLY DOES FOR THE USER>>>
Existing tagline (if any): <<<TAGLINE>>>

Write 10 distinct marketing angles. Each angle should:
- Lead with a different emotion or job (speed, status, fear, belonging, control, clarity, identity…).
- Come with a 6-word headline test and a 1-sentence hook.

Then rank them by "crowdedness" — how many competitors already claim this angle. Recommend the least crowded angle that still matches the product truthfully.

View page →ChatGPT · Claude · Gemini

PrincipleAngles beat features. The least-crowded true angle usually wins.

Marketing · Landing pages

The Landing Page Diagnostician

Audit a landing page against conversion principles.

Show prompt

Below is the copy and structure of a landing page:
<<<PASTE>>>

Audit it against these 12 checks:
1. Does the headline name a specific outcome (not a feature)?
2. Is the first screenful comprehensible in 5 seconds?
3. Is the CTA above the fold?
4. Does the sub-head remove the biggest objection?
5. Is there social proof in the first 2 screens?
6. Are features framed as benefits?
7. Is there a "for who / not for who" section?
8. Does the CTA label match the reader's goal (not "Submit")?
9. Is pricing (or its absence) explained?
10. Is there a risk reverser (guarantee, free trial, unsubscribe)?
11. Are there more than 3 competing CTAs?
12. Does the final block restate the headline?

Score each 0–2. End with the top 3 fixes, ranked by effort-to-impact.

View page →ChatGPT · Claude · Gemini

PrincipleA rubric produces consistent audits. "What do you think?" produces whatever the model felt like today.

Marketing · Email

The Subject Line Split

8 subject lines across curiosity, benefit, urgency, and personality.

Show prompt

I'm sending an email about: <<<TOPIC + ONE-LINE PAYOFF>>>
To: <<<AUDIENCE>>>
Brand voice: <<<3 ADJECTIVES + 1 EXAMPLE SENTENCE>>>

Give me 8 subject lines — two in each style:
- Curiosity gap
- Specific benefit
- Gentle urgency
- Personality / voice

Rules: under 50 characters, avoid ALL CAPS, avoid "" and !!. No emoji unless it's in the brand voice. After each, write one line predicting who opens it and who deletes it.

View page →ChatGPT · Claude · Gemini

PrincipleGenerating across named styles forces useful variance — "give me 10 subject lines" produces 10 flavours of the same line.

Marketing · Competitive

The Competitor Gap Finder

Messaging angles your competitors don't own.

Show prompt

My product: <<<ONE-LINER>>>
My 3 closest competitors and their taglines / hero copy:
1. <<<COMPETITOR A>>>
2. <<<COMPETITOR B>>>
3. <<<COMPETITOR C>>>

Do three things:
1. Map each competitor to the emotional territory they occupy.
2. Name the 3 territories nobody has claimed that my product could honestly own.
3. For the strongest gap, draft a hero headline + subhead that plants my flag there.

Only suggest territories I can defend. Don't invent capabilities.

View page →ChatGPT · Claude · Gemini

PrincipleCategory design wins by naming empty territory — not by beating the incumbent at their own game.

03 — Coding & Development

Prompts that treat AI like a senior engineer.

For reviewing, debugging, refactoring, and catching the edge cases you missed.

Coding · Review

The Rubber Duck

Have the model explain your code back to you — and flag what it can't explain.

Show prompt

Below is a function / file I wrote:
<<<PASTE CODE>>>

Explain it back to me as if you were the author on a code review call. Be concrete:
- What does it actually do, line by line (group trivial lines)?
- What assumption is it making about its inputs?
- Which line would you be most nervous to change, and why?
- What does the naming imply that the code doesn't actually do?

If there's a block you cannot explain with confidence, say so — don't paper over it.

View page →ChatGPT · Claude · Gemini

PrincipleExplaining code surfaces hidden assumptions faster than any linter.

Coding · Edge cases

The Edge Case Scout

List the inputs that will break your function.

Show prompt

Here's a function:
<<<PASTE CODE>>>

List 15 edge cases that could break it, grouped as:
- Empty / null / undefined inputs
- Extreme sizes (0, 1, huge, negative)
- Type surprises (string where number expected, unicode, etc.)
- Concurrency / timing
- External failures (network, disk, API)
- Security (injection, overflow, auth)

For each, write a one-line test case. Mark the 3 most likely to bite in production.

View page →ChatGPT · Claude · Gemini

PrincipleA named taxonomy (empty, extreme, type, timing, external, security) generates edge cases your brain forgets.

Coding · Refactoring

The Refactor Reviewer

Rewrite code for readability — with a changelog you can defend in PR review.

Show prompt

Below is working code. It does not have bugs I know of. I want it clearer.
<<<PASTE CODE>>>

Constraints:
- Do NOT change behaviour.
- Do NOT introduce new dependencies.
- Optimise for: fewer indentation levels, clearer names, earlier returns, narrower scope of variables.

Output:
1. The refactored code.
2. A bulleted changelog — every non-cosmetic change with one-line reasoning.
3. One thing you considered but didn't do, and why.

View page →ChatGPT · Claude · Gemini

PrincipleDemanding a changelog prevents silent behavioural drift — which is where LLM refactors quietly kill you.

Coding · Debugging

The Bug Suspect List

Rank the most likely causes of a failing bit of code.

Show prompt

Symptom: <<<WHAT'S GOING WRONG>>>
Stack trace / error:
<<<PASTE>>>
Relevant code:
<<<PASTE>>>
What I already tried: <<<1–3 THINGS>>>

Rank the 5 most likely causes, from most to least probable. For each:
- One-line explanation of the mechanism.
- The cheapest single test I could run right now to confirm or rule it out.
- What you would check if that test is negative.

Do not suggest a fix yet. Diagnose first.

View page →ChatGPT · Claude · Gemini

PrincipleSeparating diagnosis from fix stops the model from shotgunning code changes at a symptom it hasn't understood.

Coding · Design

The API Designer

Three API shapes for a feature — with honest tradeoffs.

Show prompt

Feature: <<<WHAT IT NEEDS TO DO>>>
Callers: <<<WHO CONSUMES THIS AND HOW>>>
Constraints: <<<LANGUAGE / FRAMEWORK / EXISTING PATTERNS>>>

Propose 3 distinct API shapes — not variants of the same shape. For each:
- Signature(s) in code.
- The mental model the caller needs to hold.
- What's easy, what's hard, what's impossible.
- The class of bug it makes likely vs. prevents.

End with your recommendation for the given callers, and one that you'd pick if callers were different.

View page →ChatGPT · Claude · Gemini

PrincipleAn API's quality is a function of the caller's mental model, not the implementer's preferences.

Coding · Testing

The Test Writer

Generate tests a senior engineer would actually insist on.

Show prompt

Code under test:
<<<PASTE>>>

Write tests in <<<LANGUAGE + FRAMEWORK>>>. Cover:
- The happy path (1 test).
- The 3 edge cases most likely to cause production incidents.
- One property-style test (invariants that should hold across inputs).
- One test that would fail if someone changed a specific line (state which line).

After the tests, write a one-paragraph "what this suite does NOT cover" section — honesty over false confidence.

View page →ChatGPT · Claude · Gemini

PrincipleAsking the model what it doesn't cover is often more valuable than what it covered.

04 — Product & UX

Prompts that make your product less stupid.

For stress-testing user stories, sharpening PRDs, auditing microcopy, and finding drop-off before users do.

Product · Stories

The User Story Stress Test

Pressure-test a user story before it enters the backlog.

Show prompt

Story: <<<AS A... I WANT... SO THAT...>>>
Context: <<<PRODUCT + TARGET USER>>>

Stress-test this story. Give me:
1. The unstated assumption it is quietly making about the user.
2. The edge cases it does not account for (list 5, ranked by likelihood).
3. The "so that" — is the stated reason the real reason? Propose 2 more honest ones.
4. What would change if the user were 10x more technical? 10x less?
5. The smallest version of this story that still delivers value.

End with: ship as-is, refine, or kill. One line of reasoning for each.

View page →ChatGPT · Claude · Gemini

PrincipleA story that can't survive a stress test in a doc won't survive contact with real users.

Product · Microcopy

The UX Copy Auditor

Audit every button, label, and error message for clarity and tone.

Show prompt

Below is every piece of copy from a screen — headers, buttons, labels, placeholders, errors, empty states.
<<<PASTE ALL STRINGS>>>

Brand voice in 3 adjectives: <<<VOICE>>>
Audience: <<<WHO>>>

Audit each string against:
- Does it use the user's words (not ours)?
- Is it doing one job (not two)?
- Does it assume knowledge the user doesn't have?
- Does the button label match the outcome the user expects?
- Does the error tell them what to do next?

Flag the 10 worst offenders with a before/after rewrite and one-line reason each.

View page →ChatGPT · Claude · Gemini

PrincipleMicrocopy is 90% of the UX nobody audits — which is why it's where the compounding wins live.

Product · Onboarding

The Onboarding Drop-off Finder

Predict where new users will quit — before you ship.

Show prompt

Here's the onboarding flow, step by step:
<<<STEP 1, 2, 3…>>>

Target user: <<<WHO, what they know, what they expect>>>

Walk through as that user. For each step:
- What is the user trying to do?
- What is the product asking them to do?
- Is there a gap? If so, what is the drop-off risk (low/med/high)?
- The single change that most reduces drop-off without lengthening the flow.

End with the one step you would cut entirely if you could only cut one.

View page →ChatGPT · Claude · Gemini

PrincipleCutting a step usually beats clarifying one. Friction compounds.

Product · Decisions

The Ship / Kill / Test Call

Decide whether to ship, kill, or A/B test — with written reasoning.

Show prompt

Feature: <<<WHAT IT DOES>>>
Evidence for: <<<DATA + QUOTES>>>
Evidence against: <<<DATA + QUOTES>>>
Cost to build: <<<T-SHIRT SIZE>>>
Cost to maintain: <<<ONGOING>>>

Make the call: ship, kill, or A/B. Do all three:
1. Write the case for shipping (strongest version).
2. Write the case for killing (strongest version).
3. Write the case for A/B testing — including what the test would need to show to tip the decision either way.

Then recommend one — with the condition under which you'd change your mind.

View page →ChatGPT · Claude · Gemini

PrincipleSteel-manning both sides before deciding is the cheapest way to avoid a roadmap full of regret.

Product · PRD

The PRD Sharpener

Turn a vague product brief into a PRD engineering can build from.

Show prompt

Below is a rough product brief:
<<<PASTE>>>

Rewrite it as a PRD with these sections, each under 150 words:
- Problem (in the user's voice — not ours)
- Who it's for (a specific persona, not "users")
- What success looks like (one metric, one threshold, one timeframe)
- What we will build (scope)
- What we explicitly will NOT build (non-scope)
- Constraints (tech, legal, brand)
- Open questions (numbered, assigned to who can answer)

If a section can't be filled in honestly, write "TK" and one line on what's blocking it. Don't invent.

View page →ChatGPT · Claude · Gemini

PrincipleA PRD's job is to force disagreement early. Vague = postponed conflict = wasted sprints.

Product · Competitive

The Competitor Feature Autopsy

Reverse-engineer why a competitor built a feature — and what it teaches you.

Show prompt

Competitor: <<<NAME>>>
The feature they shipped: <<<DESCRIBE IT + SCREENSHOT DETAILS>>>
What I know about their business: <<<USERS, PRICING, FUNDING>>>

Reverse-engineer. Produce:
1. The user job this feature is most plausibly serving.
2. The metric they were most likely trying to move.
3. Three hypotheses that would have justified building it.
4. What it implies they believe about their users.
5. One thing it reveals they're worried about.
6. Whether my product should copy it, counter it, or ignore it — with one line of reasoning.

View page →ChatGPT · Claude · Gemini

PrincipleA competitor's feature is evidence about their strategy. Read it that way, not as "what's missing from us".

05 — Strategy & Decision-Making

Prompts that stop you fooling yourself.

For bets, moats, kill-criteria, and the 2x2 that forces a decision nobody wants to make.

Strategy · Bets

The Bet Register

List every strategic bet a plan is quietly making.

Show prompt

Here is a strategy / plan / OKR:
<<<PASTE>>>

Every plan is a stack of bets. Surface them. Produce a table:
- The bet (a specific claim about the world)
- Why we're making it (one line)
- Confidence that it's true (low / medium / high)
- The cheapest way to test it before we've committed
- What breaks if it's wrong

List at least 10 bets. Include the ones nobody wants to name.
End with: the two bets whose failure would hurt most, and the order I should test them in.

View page →ChatGPT · Claude · Gemini

PrincipleA plan is really a portfolio of bets. Naming the bets is the cheapest risk management you'll ever do.

Strategy · Framing

The 2×2 Forcer

Build the 2×2 that forces a decision out of a muddled debate.

Show prompt

The muddled question: <<<WHAT WE CAN'T DECIDE>>>
Options on the table: <<<LIST>>>
What actually matters: <<<PRIORITIES>>>

Build a 2×2 that collapses this decision. Do three things:
1. Propose the two axes — each must be orthogonal and high-signal. (Avoid "cost/value" — too blunt.)
2. Plot each option in the right quadrant, with one line explaining placement.
3. Name each quadrant honestly, including the ugly ones.

End with: which quadrant are we actually in today, and which quadrant are we trying to get to?
If the 2×2 doesn't force a decision, suggest a different pair of axes.

View page →ChatGPT · Claude · Gemini

PrincipleA good 2×2 doesn't describe the world — it forces a decision. The axes are the work.

Strategy · Exit

The Kill Criteria

Define — upfront — what would make you kill the project.

Show prompt

Project / bet / initiative: <<<WHAT IT IS>>>
Timeline: <<<HOW LONG WE'VE GIVEN IT>>>
Resources committed: <<<PEOPLE + MONEY>>>

Write the kill criteria — the specific conditions that would make us stop. For each:
1. The metric or signal.
2. The threshold (a number, not a vibe).
3. By when we need to see it.
4. Whose call it is to pull the plug.

Include at least 3 lagging and 3 leading indicators. No wiggle words. If the answer is "it depends", replace it with "we'll decide on <date> using <specific data>".

View page →ChatGPT · Claude · Gemini

PrincipleYou can't escape the sunk-cost fallacy on the day — only with rules you wrote when you were sober.

Strategy · Inversion

The Inversion Move

Ask "how do we fail?" — then work backwards.

Show prompt

Goal: <<<WHAT WE'RE TRYING TO ACHIEVE>>>
Plan: <<<HOW WE'RE ATTEMPTING IT>>>

Invert. Imagine we have to guarantee failure. Write the 10 most effective things we could do to fail — cleanly and completely.

Group them by:
- Wrong people
- Wrong metric
- Wrong timeline
- Wrong bet
- Wrong order

Now check the current plan against the list. Which failure modes is it quietly adopting? Mark each with:
✘ = already doing it
△ = at risk
✓ = immune

End with the single most important "stop doing this" the plan needs.

View page →ChatGPT · Claude · Gemini

PrincipleCharlie Munger: "Tell me where I'm going to die, so I don't go there." Easier than solving the positive case.

Strategy · Counterfactual

The Counterfactual Planner

Ask: what would have to be true for this plan to work?

Show prompt

The plan: <<<PASTE>>>
The target outcome: <<<WHAT WE EXPECT>>>

Don't ask "will this work?". Ask "what would have to be true for it to work?"

List every condition — environmental, internal, competitive, behavioural — that the plan silently depends on. For each:
- The condition (phrased as a testable statement).
- Current evidence for it (strong / weak / none).
- What would move the evidence from weak to strong.

Identify the weakest link — the condition most likely to be false — and propose the one experiment that would validate or kill it before we commit.

View page →ChatGPT · Claude · Gemini

PrincipleAmazon's "working backwards" memo. The weakest true-condition is where your plan will snap.

Strategy · Moats

The Moat Audit

What actually defends this position — and for how long?

Show prompt

The business / product / position: <<<DESCRIBE>>>
Claimed moat(s): <<<WHAT WE TELL OURSELVES AND INVESTORS>>>

Audit each claimed moat against the Hamilton Helmer 7 Powers:
1. Scale economies
2. Network economies
3. Counter-positioning
4. Switching costs
5. Branding
6. Cornered resource
7. Process power

For each power, assess: present, emerging, absent, or imaginary. Be honest — most self-reported moats are imaginary.

Then answer: which single power, if we invested in it for 12 months, would best compound? And what's the cheapest experiment to prove it's real before we commit?

View page →ChatGPT · Claude · Gemini

PrincipleMost companies describe their moat as whatever feels true this quarter. The 7 Powers framework stops that.

06 — Research & Analysis

Prompts that read for you — properly.

For synthesising sources, prepping interviews, catching bias, and sanity-checking claims.

Research · Synthesis

The Literature Synthesiser

Turn N conflicting sources into a single honest view.

Show prompt

Topic: <<<WHAT I'M TRYING TO UNDERSTAND>>>
Sources (with excerpts):
<<<1. SOURCE A — key claim + quote>>>
<<<2. SOURCE B — key claim + quote>>>
<<<… etc.>>>

Synthesise. Produce:
1. The one-sentence consensus (what everyone agrees on).
2. The genuine disagreements — cluster them and quote the strongest version of each side.
3. Areas where the sources are silent but should not be.
4. The strongest evidence for the minority view.
5. Your best-guess answer, caveated by what would change your mind.

Do not average the sources. Weight them — and say why.

View page →ChatGPT · Claude · Gemini

PrincipleAveraging sources hides the signal. Weighting them — and showing your working — is the job.

Research · Interviews

The Interview Question Bank

Generate 20 interview questions, grouped by what you're trying to learn.

Show prompt

What I'm researching: <<<TOPIC>>>
Who I'm interviewing: <<<ROLE + CONTEXT>>>
What I need to learn (be honest): <<<3 QUESTIONS>>>

Write 20 interview questions, grouped into:
- Warm-up (2) — low-stakes, get them talking
- Behaviour (6) — what they actually did, with dates
- Mental model (4) — how they think about it
- Pain & workaround (4) — where the current solution leaks
- Stakeholders (2) — who else is involved
- Stress test (2) — questions that pressure-test what they said earlier

Rules: no leading questions. No yes/no questions. No "how important is X to you" — ask about the last time they did it. Mark the 5 most critical.

View page →ChatGPT · Claude · Gemini

PrinciplePast behaviour beats stated intent. "What did you do last Tuesday?" beats "what would you do if…".

Research · Stats

The Sample Size Sanity Check

Sanity check whether your data supports the claim.

Show prompt

The claim: <<<E.G. "CONVERSION LIFTED 12%">>>
The data: <<<N, SPLIT, TIMEFRAME, BASELINE>>>
The test: <<<A/B? OBSERVATIONAL? SURVEY?>>>

Sanity check in plain English. Tell me:
1. Is N big enough for the effect size claimed? (Give a rough minimum.)
2. What is the most likely non-causal explanation?
3. Has the confidence interval been quoted, or just the point estimate?
4. Could the split be confounded (time of day, segment, seasonality)?
5. What would a skeptic with 20 years of stats experience ask?

End with: is this claim safe to act on, or does it need more data / a different test? One sentence of reasoning.

View page →ChatGPT · Claude · Gemini

PrincipleMost "results" in deck form don't survive a confidence interval. Ask for the interval, not the headline.

Research · Surveys

The Survey Bias Finder

Audit your survey for leading questions, ambiguity, and acquiescence bias.

Show prompt

Below are the survey questions:
<<<PASTE>>>

For each, check for:
1. Leading language (does the wording nudge a particular answer?)
2. Double-barrelled questions (two questions in one)
3. Assumed answers (questions that don't let the honest answer out)
4. Scale issues (unequal intervals, missing "don't know", forced-choice when it shouldn't be)
5. Ordering effects (questions that prime the next)
6. Acquiescence bias (easier to agree than disagree)

Mark each question "clean", "rewrite", or "remove". For rewrites, give the fixed version.
End with the 3 most important fixes, ranked by impact on results validity.

View page →ChatGPT · Claude · Gemini

PrincipleA biased survey is worse than no survey — because it produces confidence in the wrong direction.

Research · Prep

The Expert Briefer

Prep for a conversation with a domain expert — without looking clueless.

Show prompt

I'm about to interview: <<<EXPERT + ROLE + FIELD>>>
I already know: <<<WHAT I UNDERSTAND SO FAR>>>
I want to learn: <<<3 SPECIFIC THINGS>>>
Timeframe: <<<MINUTES AVAILABLE>>>

Build me a brief. Include:
1. A glossary of terms I should be fluent in (10 max) — with one-line definitions.
2. The 5 questions an amateur would ask that would signal amateur-hour.
3. The 5 questions that would signal I've done my homework.
4. The 3 open debates in the field I should know exist.
5. The one question I could ask that the expert probably hasn't been asked before.

End with my first 30 seconds — how to open the call so they don't write me off in the first minute.

View page →ChatGPT · Claude · Gemini

PrincipleExperts give their best answers to the questions they haven't been asked a thousand times.

Research · Verification

The Fact Crosscheck

Sanity-check a claim against what you'd need to believe for it to be true.

Show prompt

The claim: <<<SPECIFIC STATEMENT — NUMBER / EVENT / ATTRIBUTION>>>
Where I saw it: <<<SOURCE>>>

Don't verify it yet. Do this first:
1. What would have to be true upstream for this claim to be correct?
2. What would the claim predict we'd also see, if true?
3. Whose interest does it serve to have this claim circulate?
4. What's the simplest alternative that would also fit the evidence?
5. The single lookup I could do in 2 minutes to confirm or undermine it.

End with: how confident should I be, 0–100%, based on shape alone — before I check a source?

View page →ChatGPT · Claude · Gemini

PrincipleMost false claims are shaped wrong — you can smell them before you verify them. Train that reflex.

07 — Sales & Negotiation

Prompts for closing the deal — without sounding desperate.

For discovery calls, follow-ups, deal reviews, pricing defence, and referrals that don't feel cheap.

Sales · Discovery

The Discovery Call Plan

Structure a first call so you learn more than you pitch.

Show prompt

I have a 30-minute first call with:
Prospect: <<<COMPANY + ROLE + WHAT I KNOW>>>
Product / offer: <<<ONE-LINE>>>
Best-case outcome for me: <<<ACTION THEY TAKE NEXT>>>

Plan the call. Produce:
1. The first 60 seconds — how I open without sounding like a vendor.
2. The 5 questions I need answered to qualify. (Each 2-layers deep — the follow-up that exposes the real answer.)
3. The 2 things I should NOT pitch — no matter how much they invite me to.
4. The trial balloon I can float to gauge budget and urgency without asking directly.
5. The closing ask — specific next step with a date.

End with the 3 signals that tell me to walk away and not waste another meeting.

View page →ChatGPT · Claude · Gemini

PrincipleThe best discovery calls feel like diagnostic appointments. Pitch like a doctor, not a vendor.

Sales · Follow-up

The Follow-Up Rebuilder

Rewrite a cold follow-up so it doesn't look like every other cold follow-up.

Show prompt

Current follow-up email:
<<<PASTE>>>

Context:
- What I already sent them: <<<SUMMARY>>>
- What they already know about me: <<<WHAT>>>
- Why this follow-up: <<<NEW ANGLE OR JUST A NUDGE?>>>

Rewrite it against these rules:
- First line not about me.
- No "just checking in", "circling back", "bumping this to the top".
- Must contain something new — a piece of info, a question, or a take. No nudges.
- Under 80 words.
- CTA must be low-cost (a reply, not a meeting, unless a meeting is warranted).
- Ends in a way that makes "no" easy. Lower the stakes.

Give me three versions — softer, sharper, and genuinely interesting.

View page →ChatGPT · Claude · Gemini

PrincipleA follow-up with nothing new in it is a tax on the prospect's inbox. Always carry a gift.

Sales · Lost deals

The Deal Autopsy

Work out why a deal really died — beyond "they picked a competitor".

Show prompt

Deal: <<<COMPANY + DEAL SIZE + LENGTH>>>
Stated reason we lost: <<<WHAT THEY TOLD US>>>
What actually happened (timeline): <<<KEY TOUCHES>>>
Champion strength (1–5): <<<X>>>

Autopsy time. Produce:
1. The stated reason — and why it's almost certainly incomplete.
2. 3 more likely underlying reasons.
3. The moment in the timeline the deal most likely actually died (often weeks before the "no").
4. What the champion didn't say that they should have said on our behalf internally.
5. One thing we'd do differently in the first 2 calls next time — specific, not generic.

End with: is this lesson specific to this deal, or a pattern? How would I check?

View page →ChatGPT · Claude · Gemini

PrincipleDeals die weeks before anyone says no. The autopsy job is to find the actual time of death.

Sales · Champions

The Champion Enablement Kit

Arm your internal champion so they can sell for you without you in the room.

Show prompt

Champion: <<<NAME + ROLE + SENIORITY>>>
Who they need to convince: <<<CFO / VP / COMMITTEE>>>
What the buyer cares about: <<<TOP 3 CONCERNS>>>
Our product in one line: <<<IT>>>

Build the kit. Produce:
1. A 90-word summary the champion can paste into an internal email.
2. The 3 internal objections they'll face — with one-line rebuttals, in their boss's language (not ours).
3. A "what if they ask about X" FAQ (5 items, most senior stakeholder's likely questions).
4. One screenshot or metric they should lead with in the internal deck.
5. A question they should NOT answer on our behalf — and the one line to hand it back to us.

Everything should sound like the champion, not like us. Keep the marketing voice out.

View page →ChatGPT · Claude · Gemini

PrincipleIn B2B, you win when your champion sells for you in meetings you're not invited to.

Sales · Pricing

The Pricing Defence

Defend a price point without blinking — or slashing margin.

Show prompt

Our price: <<<$>>>
Competitor / status quo price: <<<$>>>
What the prospect said: <<<"TOO EXPENSIVE" / "NEED A DISCOUNT" / ETC.>>>
What they actually need (best guess): <<<BUDGET / PERMISSION / COVER>>>

Draft 5 responses, each targeting a different underlying reason:
1. They literally can't afford it.
2. They can afford it but need internal cover.
3. They're testing whether we'll discount reflexively.
4. They're comparing to the wrong thing.
5. They don't see the value — yet.

For each, give the one-line response and the follow-up question that surfaces which reason it really was. Mark the two you think are most likely and why.

Never open with a discount. If a concession is needed, trade — don't give.

View page →ChatGPT · Claude · Gemini

Principle"Too expensive" is rarely about price. Diagnose the reason before you reach for the discount button.

Sales · Referrals

The Referral Ask

Ask for a referral in a way that makes it easy to say yes.

Show prompt

Customer: <<<NAME + HOW LONG + WHY THEY'RE HAPPY>>>
Their network type: <<<INDUSTRY / ROLE / CITY>>>
Who I'd love intros to: <<<SPECIFIC PROFILE>>>

Write the referral request as an email. Rules:
- Don't open with "I hate to ask".
- Give them the exact forwardable message (so forwarding takes 3 seconds).
- Be specific about who I'm looking for — vague asks get vague intros.
- Give them an out — make it easy to say no.
- Thank them in advance for considering, not for the intro (pre-gratitude = pressure).

Produce:
1. The email to the customer.
2. The forwardable blurb (≤80 words) they can paste.
3. One low-key follow-up if they don't reply in 2 weeks.

View page →ChatGPT · Claude · Gemini

PrincipleGive the referrer a forwardable blurb or you'll get a polite "sure, who should I introduce you to?" — and nothing else.

08 — Data & Analytics

Prompts that interrogate the data — not just visualise it.

For SQL, metric trees, chart choice, dashboards, and catching data that's quietly wrong.

Data · SQL

The Query Explainer

Translate any SQL query into plain English — and catch what it's doing wrong.

Show prompt

Below is a SQL query:
<<<PASTE>>>

Schema (if you need it): <<<TABLE + COLUMNS, OR "ASSUME TYPICAL">>>

Explain it in three layers:
1. One sentence: what the query is actually answering.
2. Step-by-step, in plain English, what each clause does — in the order the DB would process it (FROM → WHERE → GROUP BY → SELECT → ORDER).
3. Gotchas: NULL handling, duplicates from joins, implicit casts, performance cliffs, timezone issues, silent exclusions.

End with: the single question a reviewer should ask the author to check whether the query means what they think it means.

View page →ChatGPT · Claude · Gemini

PrincipleSQL lies quietly. Explaining it in plain English surfaces the bugs SELECT statements hide.

Data · Metrics

The Metric Tree Builder

Decompose a top-line metric into the inputs you can actually move.

Show prompt

Top metric: <<<REVENUE / MAU / RETENTION / ETC.>>>
Current value: <<<NUMBER>>>
Target: <<<NUMBER + TIMEFRAME>>>
Business context: <<<1–2 LINES>>>

Build the metric tree. Three levels:
- Level 1: the top metric, and the 2–3 drivers that mathematically produce it.
- Level 2: the 2–3 inputs feeding each driver.
- Level 3: the leverage points — the specific, operational inputs a team can move this quarter.

For each Level 3 node, mark:
- Elasticity — how much moving it moves the top (low / med / high)
- Controllability — how directly we can move it
- Instrumented — do we measure it today (yes / no)

End with: the 3 inputs with best elasticity × controllability that we are NOT currently measuring. Fix instrumentation first.

View page →ChatGPT · Claude · Gemini

PrincipleYou can't move a top-line number. You can only move the inputs. A metric tree finds the inputs.

Data · Dashboards

The Dashboard Declutterer

Decide what to remove from a dashboard — without losing the signal.

Show prompt

Below are the charts on a dashboard:
<<<LIST EACH CHART: TITLE + METRIC + WHO VIEWS IT + HOW OFTEN THEY ACT ON IT>>>

The dashboard's single job: <<<DECISION IT SHOULD DRIVE>>>

Declutter. For each chart, assign:
- KEEP — directly supports the decision
- MOVE — matters but doesn't belong here
- KILL — nobody's acted on it in 60 days
- MERGE — redundant with another chart

Then reorder the KEEPs so the decision-maker sees the right chart first.
End with the 2 charts that are MISSING — the ones the decision quietly needs but nobody built yet.

View page →ChatGPT · Claude · Gemini

PrincipleDashboards die by accretion. If a chart hasn't changed a decision in 60 days, it's furniture — cut it.

Data · Scoping

The Analysis Brief

Scope an ad-hoc analysis before you spend 2 days on the wrong question.

Show prompt

The question as asked: <<<HOW IT WAS POSED, VERBATIM>>>
Who's asking: <<<ROLE>>>
What they'll do with the answer: <<<BEST GUESS>>>

Turn this into a brief. Produce:
1. The real underlying question (often different from the asked one).
2. The decision the answer will drive — if no decision, stop here and go back.
3. The shape of the answer before I run the query (a headline number? a distribution? a time series?).
4. The minimum data I need to answer it.
5. Three assumptions I'm quietly making that I should confirm with the asker before starting.
6. What would make the answer useless (missing segmentation, wrong timeframe, confounded split).

End with: do I have everything I need to start — yes / no / needs confirmation.

View page →ChatGPT · Claude · Gemini

PrincipleMost bad analyses answer the asked question correctly. Good ones answer the question the asker should have asked.

Data · Quality

The Data Quality Inspector

Catch the 10 things wrong with a dataset before you analyse it.

Show prompt

Dataset description: <<<WHAT IT IS, HOW IT WAS COLLECTED>>>
Sample rows / schema / summary stats: <<<PASTE>>>

Inspect. For each dimension, flag anomalies:
1. Nulls — where (and what should/shouldn't be null)
2. Duplicates — at what grain
3. Types — surprises (string where number, dates as strings)
4. Ranges — impossible values (negatives, future dates, >100%)
5. Cardinality — lots of near-duplicate categories
6. Gaps — missing days/users/segments
7. Selection bias — what's not in the dataset that should be
8. Timezones / units — silent conversions
9. Joins — keys that should match but don't
10. Drift — did the schema or definition change mid-period

Produce a ranked punch list: must-fix before any analysis, should-fix, nice-to-have. One-line fix per item.

View page →ChatGPT · Claude · Gemini

PrincipleA dataset that hasn't been inspected is a booby trap. Inspect first; analyse second.

Data · Visuals

The Chart Prescription

Pick the right chart — and argue against the one you were about to use.

Show prompt

The question the chart should answer: <<<ONE SENTENCE>>>
The data: <<<SHAPE — 2 VARS, 5 VARS, TIME SERIES, CATEGORIES>>>
Audience: <<<WHO AND THEIR DATA LITERACY>>>
Medium: <<<DECK, DASHBOARD, EMAIL, ARTICLE>>>

Prescribe:
1. The best chart type — and why.
2. The second-best — and when you'd use it instead.
3. The chart type I was probably about to use — and why it's wrong.
4. Labels and annotations that must be on it for the audience to read it in under 10 seconds.
5. Two elements I should NOT include (gridlines, 3D, pie chart, whatever).

End with: one sentence, in plain English, describing what the chart will show — before it exists.

View page →ChatGPT · Claude · Gemini

PrincipleChart type is a question, not an aesthetic. Pick the shape that answers the question fastest.

09 — Learning & Thinking

Prompts for actually understanding things.

For mapping knowledge gaps, Feynman-checking your grasp, and building study plans that stick.

Learning · Gaps

The Knowledge Gap Map

Find what you don't know you don't know about a topic.

Show prompt

Topic: <<<WHAT I'M LEARNING>>>
What I already know (be honest): <<<5–10 BULLETS OF WHAT I CAN EXPLAIN>>>
My goal: <<<WHY I'M LEARNING THIS>>>

Map the territory. Four quadrants:
1. Known known — what I already know.
2. Known unknown — what I know I don't know (list 5).
3. Unknown known — things I've seen but haven't recognised as important (list 5).
4. Unknown unknown — important sub-topics I haven't even touched (list 5).

For each item in quadrants 2 and 4, add:
- Why it's likely to matter for my goal.
- The single best artefact (paper, essay, video, book chapter) to close the gap — most leverage per hour.

End with my next 10 hours of study, in order. Concrete, not vague.

View page →ChatGPT · Claude · Gemini

PrincipleDunning-Kruger. The fastest path to mastery is naming what you don't know — and quadrant 4 is where the money is.

Learning · Feynman

The Feynman Checker

Find the parts of a topic you can't actually explain simply.

Show prompt

Topic: <<<CONCEPT I THINK I UNDERSTAND>>>
My explanation, in plain English: <<<PASTE — TALKING TO A 12-YEAR-OLD>>>

Audit the explanation as a sceptical 12-year-old. Produce:
1. The sentences where I'm using a technical term to explain a technical term (circularity).
2. The places where I wave hands ("basically", "essentially", "it just…").
3. The example I used — is it doing work, or decoration?
4. The questions a smart 12-year-old would ask that my explanation doesn't answer.
5. The one concept, if I mastered it, that would make the whole explanation 2x simpler.

End with a verdict: do I understand it, or do I know the words? One sentence.

View page →ChatGPT · Claude · Gemini

PrincipleIf you can't explain it simply, you don't understand it — Feynman. The prompt makes you prove it on paper.

Learning · Plans

The Study Schedule

Build a study plan with spaced repetition, not a reading list.

Show prompt

Topic: <<<WHAT>>>
Hours I have per week: <<<X>>>
By when: <<<DATE>>>
End state I want: <<<WHAT I'LL BE ABLE TO DO>>>

Build a study plan — not a reading list. Every week must include:
- New input (reading / watching / doing) — specific resource, specific hour count.
- Retrieval — active recall of what I learned last week + 2 weeks ago + 4 weeks ago.
- Application — a concrete thing I build, solve, or produce.
- Review — 15 minutes at week's end on what I can't yet do.

Prefer fewer resources, used deeply, over many resources skimmed. Cap the reading list at 5 items total.

End with the single habit change that would 2x the ROI on this plan.

View page →ChatGPT · Claude · Gemini

PrincipleReading lists aren't study plans. Retrieval, spacing, and application are what build durable skill.

Learning · Compression

The Concept Compressor

Distil a topic to 5 ideas you could teach from memory.

Show prompt

Topic: <<<WHAT>>>
Source material (summary, chapters, notes): <<<PASTE>>>

Compress. Produce:
1. The 5 ideas that carry the most of this topic — if someone only learned these, they'd understand 80% of it.
2. For each idea: a one-sentence definition, a one-sentence example, and the one mistake beginners make about it.
3. The 3 ideas that beginners think are important but actually aren't. Cross them out.
4. A single diagram (described in text) showing how the 5 ideas connect.
5. The one-paragraph version I could write on a napkin.

If you have to compress below 5 ideas, which one do you drop last? Why?

View page →ChatGPT · Claude · Gemini

PrincipleA topic you can't compress to 5 ideas is one you haven't understood yet. Compression is a test.

Learning · Transfer

The Transfer Finder

Find analogies from what you already know to a new field.

Show prompt

New topic I'm learning: <<<TOPIC A>>>
Fields I already know well: <<<LIST 2–3>>>

Build analogical bridges. For each concept in Topic A:
1. The closest analogue in a field I already know.
2. Where the analogy is strong (fact / mechanism / incentive structure).
3. Where the analogy BREAKS — the places a smart beginner would over-transfer and mislead themselves.

Rank the bridges by explanatory power.
Then flag: which concepts in Topic A have no clean analogue — and therefore need to be learned fresh, from the ground up?

View page →ChatGPT · Claude · Gemini

PrincipleExpertise compounds across fields through analogy. Knowing where the analogy breaks matters more than knowing where it holds.

Learning · Notes

The Note Reviewer

Audit your own notes for what you've actually retained.

Show prompt

Below are my notes on a topic:
<<<PASTE>>>

Topic in one sentence: <<<WHAT>>>

Audit. Produce:
1. The 3 things in the notes that are almost certainly wrong — or stated too strongly.
2. The 3 things that are important but missing (based on what a well-covered set of notes would contain).
3. The 5 questions I should be able to answer without looking at the notes. Write them as flashcards — question / answer.
4. Which note is most at risk of "knowledge decay" in 30 days — i.e. I'll remember the word but not the mechanism?
5. One edit pass that would make these notes 2x more retrievable (structure, recall prompts, examples).

End with: what should I do in the next 20 minutes to consolidate these, given time is limited?

View page →ChatGPT · Claude · Gemini

PrincipleNotes you never retrieve are landfill. The audit forces you to retrieve — which is when learning happens.

10 — Creative & Ideation

Prompts that make you weirder — on purpose.

For constraints, brief inversions, character development, mood boards in words, and 30 variants on a theme.

Creative · Constraints

The Constraint Generator

Invent constraints that force better ideas than a blank page.

Show prompt

Creative project: <<<WHAT I'M MAKING>>>
Current brief: <<<HOW IT'S BEEN FRAMED>>>
What's making it feel stuck: <<<VAGUE, TOO OPEN, TOO PREDICTABLE…>>>

Generate 10 constraints that would force a different result. Mix types:
- Formal (length, format, medium)
- Material (only these inputs, only this palette)
- Voice (in the style of X, avoiding Y)
- Structural (must start with, must end with, must contain)
- Time (made in 90 minutes, shipped by Friday)
- Audience (for one specific person, not a segment)

For each constraint, predict what kind of work it would produce. Flag the 3 that would most likely produce something I couldn't have imagined up front.

Apply the top pick and rewrite the brief.

View page →ChatGPT · Claude · Gemini

PrinciplePerec wrote a novel without the letter E. Constraints don't limit creativity — they manufacture it.

Creative · Inversion

The Brief Inverter

Flip a creative brief to its opposite — and see what survives.

Show prompt

Current creative brief:
<<<PASTE>>>

Invert it. Take each assertion and write the inverse — the brief you'd write if the opposite were true.
- Audience → their opposite
- Tone → the opposite tone
- Outcome → the opposite outcome
- Constraints → removed; non-constraints → added

Now answer:
1. Which inversions actually produce a more interesting brief?
2. Which ones prove the original brief was arbitrary?
3. What part of the original brief survives the inversion — that's the real brief.
4. One hybrid — the original brief taking one useful thing from the inverted version.

End with a 3-line summary of what the brief is now actually about.

View page →ChatGPT · Claude · Gemini

PrincipleMost briefs are defaults in formal dress. Inversion exposes which parts were chosen and which were inherited.

Creative · Direction

The Mood Board in Words

Describe a visual direction precisely enough that three designers produce the same thing.

Show prompt

The thing: <<<WHAT WE'RE DIRECTING — SITE, AD, LOGO, VIDEO>>>
Feeling I want: <<<ONE WORD>>>
References I know work: <<<NAMES / EXAMPLES>>>
References that feel close but wrong: <<<ANTI-REFERENCES>>>

Write the verbal mood board. Cover:
- Colour (named, temperature, saturation, contrast)
- Typography (weights, personality, era)
- Composition (negative space, alignment, hierarchy tension)
- Texture (clean / noisy / tactile / digital)
- Motion (if any — pace, easing, restraint)
- Vibe reference (one thing from another medium — a song, a film, an era of product design)
- "Anti-" brief (5 things it explicitly should NOT look like)

The test: if three good designers read this, their first draft will be in the same room. Iterate until that's true.

View page →ChatGPT · Claude · Gemini

PrincipleA brief is a convergence tool. If two designers produce radically different work, the brief was the problem.

Creative · Volume

The Idea Wall

Generate 30 variants on a theme — ranked by how bad most are supposed to be.

Show prompt

Theme / prompt: <<<CENTRAL IDEA>>>
Format each idea should take: <<<HEADLINE / CONCEPT / TAGLINE / ANGLE>>>

Generate 30 variants. Don't self-edit. Rules:
- 10 obvious ones (first-thought territory).
- 10 diagonal ones (taking the theme in an unexpected direction).
- 10 bad-on-purpose ones (deliberately terrible, overstated, weird, literal).

The bad ones matter — they mark the edges of the space. Often, fixing a bad idea produces the best one.

Then:
1. The 3 I should kill without regret.
2. The 3 I should develop.
3. The 1 I should show a colleague even though I'm embarrassed by it.

If the best 3 all came from the "obvious" batch, keep going — you haven't actually explored yet.

View page →ChatGPT · Claude · Gemini

PrincipleQuantity is how you get to quality. Forcing yourself past 20 ideas is where the diagonals start showing up.

Creative · Characters

The Character Interrogator

Develop a character via forced contradictions — not a backstory dump.

Show prompt

Character sketch so far:
<<<NAME, AGE, ROLE, SITUATION, ONE-LINE MOTIVE>>>

Interrogate. Don't invent a backstory. Instead:
1. Name 3 contradictions inside this character — things they'd do that are inconsistent with who they say they are.
2. For each contradiction, the moment in their past that made it inevitable.
3. The belief they hold that no one in their life knows about.
4. What they pretend to want — and what they actually want.
5. The thing they'd never do, under any circumstances — and the situation where they'd do it anyway.
6. Their specific verbal tic (a word or phrase they over-use).

End with a scene, 150 words, where the contradiction breaks surface. Show, don't tell.

View page →ChatGPT · Claude · Gemini

PrincipleCharacters are interesting where they're inconsistent. Backstory is setup; contradiction is payload.

Creative · Wordplay

The Pun Factory

Generate wordplay — headlines, taglines, names — with quality control.

Show prompt

Subject: <<<WHAT THE WORDPLAY IS ABOUT>>>
Tone: <<<DRY, ABSURD, GENTLE, SAVAGE…>>>
Where it'll appear: <<<HEADLINE, BRAND NAME, SOCIAL POST, PACKAGING>>>

Generate 20 pieces of wordplay. Mix types:
- Double meaning (one word doing two jobs)
- Unexpected collocation (two familiar words next to each other for the first time)
- Sound play (rhyme, alliteration, internal music)
- Reference swap (famous phrase with one word replaced)
- Literal / figurative crossover (treat a metaphor as concrete)

Then quality-control:
- Which 3 actually surprised me? Keep those.
- Which are groan-worthy in a way the tone wants (dad-joke-on-purpose)? Keep those too.
- Which are groan-worthy by accident? Kill those.
- Which would only land for a specific reader? Flag "niche".

End with the 3 I should test on someone else before using.

View page →ChatGPT · Claude · Gemini

PrincipleWordplay is volume × filter. Most writers stop at volume and publish the 3rd idea. The 15th is where the gold is.

Get the best new prompt every Wednesday.

iPrompt is two free AI newsletters: Wednesday delivers the weekly prompt, tool, and 5 AI stories. Friday Signals covers AI & robotics investing. 12,000+ operators. No spam.

One-click unsubscribe · EU-hosted · GDPR-safe · We've never sold an address

Questions, briefly.

Are these prompts actually free?

Yes — all 60 prompts, all 10 categories, no signup, no paywall, no "sign up for the full version". Copy any of them and use them right now. We make money from the newsletter, not the library.

Do they work in ChatGPT, Claude, and Gemini?

Yes. Every prompt has been tested in ChatGPT-4, Claude (3.5 Sonnet and later), and Gemini (1.5 Pro and later). Quality varies slightly across models — Claude tends to be more careful with nuance, ChatGPT more confident, Gemini somewhere in between — but each prompt produces useful output in all three.

How do I use them?

Click "Show prompt" on any card, hit "Copy prompt", paste into ChatGPT / Claude / Gemini, and replace the <<<PLACEHOLDERS>>> with your own content. Every prompt is designed so you can use it verbatim — no setup, no system prompt, no special account needed.

Can I adapt them for my own use?

Please do. These are working tools — not museum pieces. Shorten them, combine them, bend them to your stack. If you find a better version, reply to any Wednesday newsletter and we'll credit you in the library.

What's the catch?

We hope you like them enough to subscribe to the Wednesday newsletter — where we publish one new prompt a week, plus a tool review, a tip, and the 5 AI stories that matter. That's the whole funnel. No upsells, no "pro tier", no affiliate sludge.

How often do you add new prompts?

One new prompt every Wednesday, published first to newsletter subscribers, then added to the library. Over a year, that's another 50+ prompts. Subscribe to see them first.

AI Prompts That Actually Do the Job.

Prompts that make your writing sharper.

The Writing Pre-Mortem

The Ghost Reader

The Reductionist

The Hook Doctor

The Ladder of Abstraction

The Outline Inverter

Prompts that sell without sounding like it.

The VoC Translator

The Objection Inventory

The Angle Machine

The Landing Page Diagnostician

The Subject Line Split

The Competitor Gap Finder

Prompts that treat AI like a senior engineer.

The Rubber Duck

The Edge Case Scout

The Refactor Reviewer

The Bug Suspect List

The API Designer

The Test Writer

Prompts that make your product less stupid.

The User Story Stress Test

The UX Copy Auditor

The Onboarding Drop-off Finder

The Ship / Kill / Test Call

The PRD Sharpener

The Competitor Feature Autopsy

Prompts that stop you fooling yourself.

The Bet Register

The 2×2 Forcer

The Kill Criteria

The Inversion Move

The Counterfactual Planner

The Moat Audit

Prompts that read for you — properly.

The Literature Synthesiser

The Interview Question Bank

The Sample Size Sanity Check

The Survey Bias Finder

The Expert Briefer

The Fact Crosscheck

Prompts for closing the deal — without sounding desperate.

The Discovery Call Plan

The Follow-Up Rebuilder

The Deal Autopsy

The Champion Enablement Kit

The Pricing Defence

The Referral Ask

Prompts that interrogate the data — not just visualise it.

The Query Explainer

The Metric Tree Builder

The Dashboard Declutterer

The Analysis Brief

The Data Quality Inspector

The Chart Prescription

Prompts for actually understanding things.

The Knowledge Gap Map

The Feynman Checker

The Study Schedule

The Concept Compressor

The Transfer Finder

The Note Reviewer

Prompts that make you weirder — on purpose.

The Constraint Generator

The Brief Inverter

The Mood Board in Words

The Idea Wall

The Character Interrogator

The Pun Factory

Get the best new prompt every Wednesday.

Questions, briefly.

One Prompt.Every Wednesday.

One Prompt.
Every Wednesday.