← Back to Resources
    Resource · Cheat sheet

    The AI trigger code cheat sheet

    20 one-word prompts that change how any AI answers, grouped by the job you're doing. Type one word into a chat and the answer comes back completely different. Same question, different brain.

    They work in Claude, ChatGPT, Gemini and Copilot, because none of them depend on a feature. Nothing to install.

    Last updated: 25 August 2026

    What a trigger code actually is

    A trigger code is one word that stands in for a paragraph of instructions.

    There's no hidden menu inside any of these tools. No back door, no developer mode, nothing the AI companies left lying around for you to find. You define a word at the top of a chat, then type that word instead of re-writing the instruction every time.

    Which sounds small. It isn't, and here's why.

    This is the smallest system you'll ever build

    Every real AI system does the same thing at a bigger scale. You define how the work should be done once, store it somewhere, and call it by name whenever you need it. That's what an agent instruction is. That's what a workflow is. That's what we build for clients.

    A trigger code is that, shrunk down to something you can set up in ten seconds without any tools. Same shape, same logic, same payoff: you stop re-explaining yourself.

    So treat these as practice. If you find yourself typing the same instruction three times a week, that's a system waiting to be built. A code is the first version of it.

    Some of these words already mean something

    Worth knowing, because it explains why a few work better than others. REDTEAM, PREMORTEM, STEELMAN, 80/20 and the Feynman technique are real methods with decades behind them. They're all over the material these models learned from, so the AI picks them up even without a definition.

    The six marked one of ours below are labels we've attached to a behaviour, and they mean nothing until you define them. Paste the starter line either way. It's the difference between a code that works reliably and one that works when you're lucky.

    What changes between tools

    The codes themselves are portable. Three things aren't, and it's better to know them upfront than to blame yourself when something doesn't land.

    Bluntness varies

    STRAIGHT works everywhere, but some tools are more willing to be harsh with you than others. Enterprise Copilot is the politest of the four. It'll still tell you the truth, just with more cushioning.

    Definitions drift

    Over a long conversation, every model eventually slides back towards its default tone. Some hold on longer than others. The fix is in every starter line below: the instruction to keep going until you say stop.

    The persistent settings differ

    Each tool has somewhere to store your codes permanently, and they're in different places. There's a table further down for that.

    How to set one up (do this once per chat)

    Three steps. Takes about ten seconds.

    01

    Open a fresh chat

    Codes live inside a single conversation. Define them at the top where you can scroll back and find them.

    02

    Paste the starter line

    A plain sentence telling the AI what the word means. Paste it, send it, and you'll usually get a short confirmation back.

    03

    Type the code at the end of your message

    Write your real message, then drop the code on the end of it. The answer comes back in that mode instead of the default one.

    Example starter line

    From now on, when I type STRAIGHT, give me your most honest answer with no softening. No flattery, no caveats, no “great question”. Tell me the thing I don't want to hear.

    All 20 codes, grouped by the job

    Pick by the job you're doing, not by the name. Every code lists what it changes and when to reach for it. 3 complete starter lines are open below — the rest unlock with your email.

    Jump to a job

    Work out what to do next

    Six codes for when you're stuck between options, or staring at a list with no idea which end to pick up.

    DIRECTOR

    One of ours

    Give it a task and it splits the work three ways: what an AI can run unsupervised, what needs you in the loop, and what shouldn't be automated at all. The third category is the one people skip, and it's the one that saves you.

    Use it when you're about to hand something to AI and haven't thought about where the line is.

    80/20

    Starter line included

    Names the one thing that'll make the biggest difference. The 20% of effort behind 80% of the result. You get the first domino instead of a twelve-step plan.

    Use it when you're staring at a to-do list and don't know where to start.

    When I type 80/20, identify the single highest-leverage thing I should do first, the 20% of effort that will drive 80% of the result, and tell me to start there. Don't give me a long list.

    MARGIN

    One of ours

    Forces the commercial answer before the creative one. What it costs to deliver, what it earns, where the margin actually sits. Cuts a surprising number of good ideas off at the knees, which is the point.

    Use it when something sounds exciting and you haven't done the sums yet.

    FIRSTPRINCIPLES

    Ignores how things are normally done and rebuilds the problem from what's actually true. Strips out the copied best-practice and the assumptions nobody's checked.

    Use it when the standard approach feels wrong and you can't articulate why.

    LINDYMODE

    Only advice that has held up for decades. Ideas that worked in 1995 and will still work in 2035. Filters out the hype layer entirely, which in AI is most of the layer.

    Use it when you're making a decision you don't want to redo in six months — pricing, positioning, business model.

    YCOMBINATOR

    Evaluates your idea the way a startup investor would in an interview. Who's the customer? Is this a real problem? Why now? Fast, and focused on whether the thing deserves to exist at all.

    Use it when you've got a new product or service idea and want a sharp first gut-check.

    Pressure-test a decision

    Five codes for the moment before you commit. Each one attacks from a different angle, so read the “use it when” line to pick.

    STRAIGHT

    Starter line included

    Stops the AI being nice for the sake of it. No flattery, no hedging, no “that's a great question”. Just the answer, including the part you were hoping to avoid. The one we use most.

    Use it when you suspect everyone around you is being too polite about your idea, offer or copy.

    From now on, when I type STRAIGHT, give me your most honest answer with no softening. No flattery, no caveats, no “great question”. Tell me the thing I don't want to hear.

    REDTEAM

    Turns the AI into your sharpest critic, hunting for holes. Weak spots, risks, the parts that fall over under pressure. Stress-test it yourself before the market does it for you.

    Use it when you're about to launch, sign or commit, and you want the flaws found now.

    PREMORTEM

    Pretends your plan has already crashed six months from now, then works backwards to explain why. Gets far more honest answers than “what could go wrong?”, because the failure is assumed rather than argued about.

    Use it when you've committed to a plan and want the risks on the table before you're deep in it.

    STEELMAN

    The mirror image of REDTEAM. Builds the strongest possible case for the view you disagree with, argued better than its actual supporters would argue it.

    Use it when you need to understand the other side properly before a negotiation or a hard conversation.

    PRODUCTION

    One of ours

    REDTEAM attacks the idea. PRODUCTION attacks the operational reality. It assumes the thing works in a demo, then tells you what breaks at volume, on messy inputs, and the day the person who built it leaves.

    Use it when something worked beautifully once and you're about to make it a process.

    Check it before it ships

    Three codes for getting reliable answers consistently, before anything leaves the building.

    SOURCED

    One of ours

    Separates what the AI knows from what it's inferring from what it's estimating. Models blend all three by default, in the same confident tone, and that blending is where most bad decisions start.

    Use it when you're about to act on numbers, dates or facts an AI gave you.

    HUMANCHECK

    One of ours

    Run it on anything before it goes out the door. Names which specific parts need a human to verify, and what goes wrong if nobody does. Turns “you should review the output” into an actual list.

    Use it when anything is about to reach a client, a customer or a public feed.

    PLAINENGLISH

    One of ours

    Rewrites something technical for a board, a client or a sceptical CFO. ELI10 is for you understanding something. This is for you explaining it to someone whose time you can't waste.

    Use it when you understand it and now you have to present it.

    Understand something hard

    Four codes for when the explanation isn't landing. The first simplifies. The other three make you do the work, which is slower and sticks better.

    ELI10

    Starter line included

    Takes anything complicated — a tax concept, an AI term, a clause in a contract — and explains it like you're ten years old. Plain words, everyday analogies, zero jargon.

    Use it when you've read something three times and still don't get it.

    When I type ELI10, explain the topic like I'm 10 years old. Simple words, everyday analogies, no jargon. Assume I know nothing about it.

    FEYNMAN

    Makes you explain the concept back in your own words, then catches the bits you got wrong or fuzzy. The Feynman technique, and the fastest way to find the holes in what you think you understand.

    Use it when you're about to teach, present or defend something and want to check you actually know it.

    SOCRATES

    Asks you questions, one at a time, until you reason your way to the answer yourself. Slower than being told. Much better for learning something or untangling a decision you're circling.

    Use it when you want to think more clearly rather than be handed a conclusion.

    RUBBERDUCK

    The AI mostly listens while you talk a problem through, reflecting back what it hears and asking the occasional clarifying question. Half the time you solve it yourself mid-sentence. That's the point.

    Use it when you're tangled up in something and need to hear yourself say it out loud.

    Get words on the page

    Two codes for the writing itself. One gets you started, one fixes the tone.

    ROUGH

    A deliberately unpolished first draft. Something to react to instead of a blank page. Perfectionism off, momentum on.

    Use it when you've been staring at an empty document for twenty minutes.

    VOICE

    Paste a few samples of your real writing, then use VOICE on anything. The AI studies your rhythm and word choices, then rewrites the piece to sound like you. The antidote to copy that smells like AI.

    Use it when you want something in your voice and you're sick of editing the AI tone out of it.

    Free unlock

    Get every starter line, the combos, and the build-your-own method

    • The remaining 17 copy-paste starter lines
    • Four stacking combinations that work better than any single code
    • Where to store your codes permanently in Claude, ChatGPT, Gemini and Copilot
    • The method for writing your own codes

    Where this stops working

    Codes are the shallow end, and we'd rather say that than pretend otherwise.

    They change how an AI answers a question you asked. They don't do the work while you're asleep, they don't read your inbox, they don't chase an invoice or draft next week's content or catch the client email that needed a reply on Tuesday. The moment you want any of that, you've moved from prompting to building, and that's a different set of problems: permissions, escalation, who's accountable when it gets something wrong.

    Start here anyway. Set up four codes tomorrow morning and use them for a fortnight. You'll get better answers immediately, and you'll get something more useful than that: a feel for how much of the result comes down to how precisely you define the work.

    That's the skill the systems run on.

    Frequently asked

    Next step

    Want your whole team working like this?

    Trigger codes are day one of AI fluency. We train teams to use AI on their real work, then build the systems that keep it running. Book a free 30-minute consultation.

    Book a free consultation