Read this first
I tested 30+ of the codes floating around. Killed the gimmicks. The 20 below are the ones I fire 3+ times a week running Wright Mode — and they all work in any LLM (ChatGPT, Claude, Gemini, Copilot, whatever you use).
A few codes work slightly better in one model than another. Where that's true, I've flagged it on the code. Otherwise — paste, run, get a better answer. No prompt engineering degree required.
How to actually use these (3 steps)
Save them once
Drop the 20 codes into your "system prompt" or "custom instructions" in ChatGPT, and a Project in Claude. Now they're always available — no copy-paste from a screenshot at 11pm.
Stack them
These work best in combos. Run any prompt → then fire PARETO to cut the fluff. Or run BRAIN DUMP → REVERSE OUTLINE → VOICE MATCH for a publish-ready post in 15 minutes. The codes aren't standalone — they're moves in a workflow.
Fire at the right moment
Each code has a "fire when" line. Most people don't fail at AI because their prompts are bad — they fail because they use the wrong tool for the moment. BRUTAL on a half-formed idea kills it. STEELMAN on a half-formed idea sharpens it. Same week, different reads.
Codes 1–7: Pressure-test your thinking
When you need the AI to push back, not just agree. These are the ones I fire before any decision over $5k or 5 hours.
What it does: Strips the agreeableness and forces honest critique.
Fire when: You've drafted a plan, pitch, or decision and want it stress-tested.
Paste this:
BRUTAL: Stop agreeing with me. Be brutally honest. If I'm wrong, lazy, or kidding myself — say it. Challenge my logic. No softeners.
What it does: Explains anything like you're five — strips jargon, uses analogies.
Fire when: You're in a meeting and someone drops a term you can't ask about without losing face.
Paste this:
ELI5: [paste topic or concept]
What it does: Attacks your idea from every angle so weak spots surface before you ship.
Fire when: Before a launch, a pitch, or any "I'm pretty sure about this" moment.
Paste this:
RED TEAM this: [paste your plan/idea]. Find the 5 biggest holes, the assumption that's most likely wrong, and the failure mode I'm not seeing.
What it does: Builds the strongest version of the OPPOSITE position so you're not arguing with a strawman.
Fire when: You're 100% sure about something. (That's exactly when you need this.)
Paste this:
STEELMAN the opposite of what I just said. Make the strongest case against my position — the one a smart critic would actually make. Then tell me which parts I should take seriously.
Works noticeably better in Claude than ChatGPT — Claude's reasoning chains tend to be more rigorous here.
What it does: Fast-forwards 12 months and works backwards from disaster.
Fire when: Before any decision over $5k or 5 hours of your time.
Paste this:
PRE-MORTEM: Assume this fails 12 months from now. Walk me through the 3 most likely reasons why. Be specific. What would the failure actually look like? Don't reassure me.
What it does: Maps the consequences AFTER the obvious consequences.
Fire when: Hiring, pricing changes, launching anything that shifts your business model.
Paste this:
SECOND ORDER: If I do [decision], map first-, second-, and third-order consequences over the next 12 months. What unintended outcomes should I prepare for?
What it does: Surfaces the gap in your thinking you didn't know was there.
Fire when: Three times a week, on anything that matters.
Paste this:
What am I missing here? What am I being overly optimistic about? What's a macro event that could totally reverse this?
The simplest prompt on this list and one of the highest-leverage. Lenny Rachitsky fires this on every PRD.
Codes 8–14: Learn + decide faster
For when the AI should make you smarter, not lazier. These force YOU to do the thinking, with the AI as scaffolding.
What it does: Teaches you by asking questions instead of giving answers.
Fire when: You're stuck and reaching for the AI for a shortcut. Use this instead — you'll actually learn it.
Paste this:
SOCRATES MODE: Don't answer me. Ask me questions until I work it out myself. One question at a time.
What it does: Forces YOU to teach the concept back so you actually retain it.
Fire when: You're learning something you need to remember in 6 months, not 6 minutes.
Paste this:
TEACH BACK: Explain [topic] to me in 3 short steps. Then ask me to explain it back to you in my own words. Correct me where I'm wrong.
What it does: Ranks every output by leverage — what's 80% of the result.
Fire when: You asked for "all the options" and got 12. Now you need the 2.
Paste this:
PARETO: Of everything you just told me, which 20% gives me 80% of the result? Cut the rest.
Rescues every overlong AI response. Pair it with any other prompt as a one-line follow-up.
What it does: Forces structured comparison instead of a wall of text.
Fire when: You need to decide but the AI keeps giving you essays.
Paste this:
Give me 3 different approaches to [problem]. Compare side by side: pros, cons, biggest risk. Then recommend ONE based on my constraints: [list them].
What it does: Voice-dump everything in your head → AI sorts it into themes + next steps.
Fire when: Your brain is loud and you can't think straight. Sunday night planning. Post-client-call processing.
Paste this:
BRAIN DUMP: I'm about to talk in a stream of consciousness. Don't interrupt. When I say "done," organise everything into themes, surface what I said but didn't realise I said, and give me 3 concrete next steps.
Use voice mode for this one (both ChatGPT and Claude have it). The "surface what I said but didn't realise I said" line is the whole game.
What it does: Compresses a long doc into a paragraph that loses nothing.
Fire when: You have a 40-minute Fathom transcript and need to know what mattered.
Paste this:
Apply CHAIN OF DENSITY to this: [paste content]. Write an initial 80-word summary. Then rewrite it 4 more times — each pass adding 1-2 missing key entities or concepts without making it longer. Show me only the final version.
The only summary prompt that's been peer-reviewed (Salesforce/MIT). The best summary technique that exists.
What it does: Paste your draft → AI shows you its skeleton → you see what's missing.
Fire when: You've written something and it's nearly there but you can't see why it's off.
Paste this:
REVERSE OUTLINE this: [paste your draft]. For every paragraph, give me one sentence on what it actually says. Then tell me: what's repeating, what's missing, what's out of order?
Codes 15–19: Write as yourself
For when AI is helping you write — but you don't want it to sound like AI. These are the codes that protect your voice while still saving you hours.
What it does: Replicates your actual writing voice from real samples.
Fire when: You're outsourcing first drafts to AI and they keep sounding like AI.
Paste this:
Here are 3 samples of how I write: [paste 3]. Extract 5 specific patterns in my voice — sentence length, signature phrases, what I never do. Then write [new piece] using those patterns.
Examples beat instructions every time. This is what replaces the "make it sound human" gimmick.
What it does: Loads the AI with context so it thinks strategically, not generically.
Fire when: You need real strategic input, not a Google search.
Paste this:
CHIEF OF STAFF MODE: Before answering, ask me 5 questions about my business, this week, and my actual constraints. Then answer with that context — like a chief of staff who's been with me 6 months.
The "5 questions first" line is what makes it strategic. Without that, it's just role-play.
What it does: Paste something that worked → AI extracts the formula → you reuse it.
Fire when: You see a competitor's post go viral and want to know what they actually did.
Paste this:
REVERSE PROMPT this: [paste a post/email/script that worked]. Show me the underlying structure — hook, format, voice moves, what made it land. Give me a reusable template.
![WHAT WOULD A [X] ASK prompt example](/slurp/images/slurp-18-whatwould.png)
What it does: Borrows a specific expert's questioning brain without the bro-coach roleplay.
Fire when: Pitching, pricing, positioning. Before you go public with anything.
Paste this:
What questions would a [skeptical investor / experienced operator / customer who almost churned] ask before believing this? List the 7 hardest ones.
Asks for the QUESTIONS not the verdict. Subtle but the difference between "act as Hormozi" (generic) and this (specific).
What it does: Filters out what's trending and gives you what's been proven to work for years.
Fire when: You're tempted to chase the latest hack and need a reality check.
Paste this:
OLD MONEY answer: Give me approaches to [topic] that have worked for 10+ years across multiple businesses. Skip anything trendy or new. Time-tested only.
The bonus 20th: ONE LIE
This one isn't on anyone else's list. I built it because the women founders in Wright Mode kept asking the wrong AI question — "what should I do?" — when the more useful question is "what am I doing that's quietly wrong?"
What it does: Surfaces the most common piece of advice in your industry that's quietly wrong.
Fire when: You're about to do "what everyone does" — and you want to check whether that's actually still good advice.
Paste this:
ONE LIE: What's the most common piece of advice given in [my industry] right now that's quietly wrong or outdated? Tell me why people still believe it, what's actually true now, and one specific thing I should do differently.
This is Wright Mode's bespoke 20th. It's saved me from following a lot of received wisdom that stopped working in 2024.
The 8 codes I tested and CUT
The brief from me was "20 but make sure they don't suck." Here's the proof I held the line. These 8 codes are in every viral "ChatGPT secrets" post going around. They didn't earn a spot.
Why cut: Name-prompts produce generic bro-coach output. The model doesn't actually know what Alex Hormozi would say in your situation — it pattern-matches on his public soundbites. You get a watered-down version of a person, not their actual thinking.
Why cut: Inconsistent. Sometimes makes output WORSE — overcompensates with em-dashes and clichés that read as "trying to sound human." Code 15 (VOICE MATCH) does it properly by feeding the model your actual samples.
Why cut: Vibey. The AI doesn't know your future self. Output is generic life-coach pablum that could apply to anyone. Pretty sells, useful doesn't show up.
Why cut: Only works in Claude Code (the terminal-only version of Claude that developers use). Does literally nothing in regular ChatGPT or claude.ai. Fails the cross-platform test.
Why cut: Jailbreaks. Against terms of service, brittle (patched within weeks), embarrassing in a professional context. Not what we're doing here.
Why cut: Debunked. Modern reasoning models don't respond to these the way 2023 GPT-3.5 did. Research from Sander Schulhoff (and others) shows these tricks no longer add value. Including them is teaching 2023.
Why cut: Research now shows role prompting is largely ineffective beyond surface tone. The examples + context approach (Codes 15 and 16) does the same job properly.
Why cut: Gimmicky one-word keywords. The IDEAS underneath them are good but the bare word doesn't reliably trigger anything in either ChatGPT or Claude. The codes that earned a spot bake the actual instruction in.
You see "10 ChatGPT secrets" posts everywhere. Most of the codes in them are gimmicks. The reason I cut more than I kept is because the magnet only works if every code earns its slot.
If you've ever felt vaguely cheated by an AI tips post — this is why.
Particularly valuable for women founders
Not every code is universal. The ones that earn their slot SPECIFICALLY for women running businesses (the Wright Mode audience) — the ones I'd say "if you only use 6, use these":
BRAIN DUMP (12)
Most useful for founders running on chaos with no team to delegate the processing to. Treats the AI like a coach, not a notetaker.
CHIEF OF STAFF (16)
Replaces the strategic thinking partner you can't afford yet. Better than hiring badly.
BRUTAL (01)
Counters the politeness training most of us got that bleeds into our business decisions. The AI you actually need.
WHAT WOULD A [X] ASK (18)
Gets you the room-full-of-investors stress test without needing the room. Or the cost.
PRE-MORTEM (05)
Particularly valuable when you're emotionally invested in a launch (which, let's be honest, is always).
VOICE MATCH (15)
For founders writing as themselves while running everything else. This is the AI use case for solo founders.
AI isn't going to build your business for you. But the right codes — fired at the right moment — will give you a thinking partner that pushes back, surfaces your blind spots, and protects your voice while you do the work.
The women in Wright Mode use these codes weekly. Now you have them too.
Ready to use these weekly?
The codes are the start. Here's where they get real.
Wright Mode Membership
Where these codes get used weekly with a room of women founders building real AI workflows together. Live calls, templates, ongoing build support.
Claude Masterclass
The full Claude system — not just the codes. Prompting fundamentals, voice docs, projects, the whole stack I run Wright Mode on.
Coding Camp
A full day building with these codes (and more) in person. Gold Coast, Perth, Sydney, Melbourne. For non-tech founders who want to build, not just consume.