Make something new. These words move the machine past information and into making — drafts, stories, versions, and voices.
Everything before this part was about getting the truth out of the machine in a usable shape. This part is about getting something out of it that did not exist before: a draft, a name, a story, a script for the hard phone call. Creation is where most people underrate the machine most — they use it as a library when it is also, and often better, a workshop.
The catch is that machines create the way water flows: along the path of least resistance, toward the average. Every slogan sounds like every slogan; every story opens the way stories open. Creation words are the dams and channels: invent, don't recall; vary on purpose; write it in a voice; combine the two things nobody has combined. And above all, iterate — because almost nothing worth making arrives finished, and the loop of version, complaint, and revision is where your taste does its work. The machine supplies the clay at industrial speed. These words keep your fingerprints in it.
What it makes the AI do. It asks for a working first version with the clear understanding that it will be revised — which frees the machine from trying to be final and you from trying to start.
Weak prompt: "I need to write an email to my landlord about the broken heater."
Same prompt, with the word: "Draft an email to my landlord about the heater that's been broken for five days: polite, firm, mentions the two times I already called, and asks for a repair date in writing."
Why it worked. A blank page is the hardest part of any writing. A draft — even a flawed one — gives you something to push against, and pushing is faster than starting.
Another example.
Weak prompt: "I need to thank my neighbor for helping after the storm."
Same prompt, with the word: "Draft a short thank-you note to the neighbor who cleared my driveway after the storm: warm, specific about what he did, and ending with a real offer to return the favor."
Why it worked. The draft carried the feeling across the hard part. All that was left for the human was the signature — and a stamp.
One more.
Weak prompt: "I need to apologize to a client for the delay."
Same prompt, with the word: "Draft the apology: own the delay in the first sentence, give the new date, name what changed so it won't repeat — no groveling, no blaming the supplier by name."
Why it worked. Clients forgive delays with explanations attached. The draft had the explanation in sentence three, where it belongs.
Watch for. A draft's danger is its competence: it reads finished, so it gets sent — heater email dispatched with the placeholder date still sitting in sentence two. Drafts are scaffolding, and scaffolding gets checked before anyone moves in. Read it once, aloud, as its recipient.
Placed well: "Draft the landlord email: polite, firm, both earlier calls mentioned, and a repair date asked for in writing."
What it makes the AI do. It licenses originality: not the known options, but new ones — names, ideas, combinations nobody has listed yet.
Weak prompt: "What should I name my food truck?"
Same prompt, with the word: "Invent fifteen names for a food truck selling grilled cheese in the Midwest — some funny, some warm, some clever about cheese — none of them a name already famous."
Why it worked. "What should I name it?" earns the familiar five. "Invent" plus a count and flavor directions earns a page worth choosing from — choosing being the job you actually wanted.
Another example.
Weak prompt: "What should I name my dog?"
Same prompt, with the word: "Invent twenty names for a stubborn, dignified old rescue beagle — some noble, some funny, none of them from the ten most common dog names."
Why it worked. Twenty fresh candidates turned naming from settling into choosing. Abundance is a different errand than scarcity, and a much pleasanter one.
One more.
Weak prompt: "What should I call my Wi-Fi network?"
Same prompt, with the word: "Invent fifteen Wi-Fi names for a house full of teachers: some bookish, some funny, all safe to read aloud to a grandmother."
Why it worked. The constraint about the grandmother did more for the list's quality than the request did.
Watch for. Invented names arrive suspiciously available: fifteen fresh truck names, three of them already painted on trucks two states away — the machine doesn't check, and fame is a local thing. Before falling for one, search it. Originality is a claim; verify it like any other.
Placed well: "Invent fifteen names for a grilled-cheese truck — funny, warm, or clever — none of them already famous."
What it makes the AI do. It converts facts and timelines into a story with a thread through it — the form human memory is actually built to hold.
Weak prompt: "Tell me what happened in the 2008 financial crisis."
Same prompt, with the word: "Narrate the 2008 financial crisis as a story following one ordinary family's house — bought in 2005, refinanced in 2006, in trouble by 2008 — explaining each big event as it reaches them."
Why it worked. Event lists slide off the mind. A story with people in it stays — and the big forces become understandable when you watch them land on one kitchen table.
Another example.
Weak prompt: "Tell me about the factory closing in my town."
Same prompt, with the word: "Narrate the closing of our town's mill through one worker's last year — the rumor, the announcement, the final shift — so my grandchildren understand what it actually cost."
Why it worked. Dates become history when someone inside the story has a name. The economics arrived attached to a life, and stayed.
One more.
Weak prompt: "Tell me how email was invented."
Same prompt, with the word: "Narrate the invention of email through one engineer who just wanted the machines in his building to talk — and let the technology arrive as it reaches him."
Why it worked. Protocols are forgettable; a man leaving notes for colleagues on a shared computer is not.
Watch for. Narration picks a hero, and the picking is an editorial act: follow the mill worker and the closing is a tragedy; follow the mayor and it's a transition. Neither is false, both are partial. Ask, once, whose kitchen table the story never visits.
Placed well: "Narrate the crisis through one family's house — bought in 2005, refinanced in 2006, in trouble by 2008."
What it makes the AI do. Beyond languages: it converts anything written in one dialect of difficulty — legal, medical, technical, teenage — into the dialect you actually speak.
Weak prompt: "What does this insurance letter mean?"
Same prompt, with the word: "Translate this insurance letter into plain everyday language, sentence by sentence, and after each sentence tell me whether it means I owe something, I must do something, or nothing at all."
Why it worked. Jargon is often just plain news wearing a uniform. Translation strips the uniform — and the "so what is it to me" line after each sentence is where understanding turns into action.
Another example.
Weak prompt: "What does this after-visit summary from my doctor mean?"
Same prompt, with the word: "Translate this doctor's note into plain language, and after each item tell me plainly whether it is something to do, something to watch, or just information."
Why it worked. Plain words plus a "so what" turned a frightening page into a short to-do list — and the watch items stopped disguising themselves as emergencies.
One more.
Weak prompt: "What does this text from my teenager mean?"
Same prompt, with the word: "Translate this text from my daughter into plain parent language, phrase by phrase, and after each tell me the temperature: fine, joking, annoyed, or actually upset."
Why it worked. Slang was the easy half. Temperature was the translation that mattered — and it read "joking," thank heaven.
Watch for. Translation loses things on purpose — the insurance letter's menace lives partly in its formality, and a plain-language version can make a serious deadline sound like a friendly note. Keep the original beside the translation, and let the translation carry the original's temperature markings.
Placed well: "Translate this letter sentence by sentence, marking each: I owe, I must act, or nothing is required."
What it makes the AI do. It keeps the content and changes the delivery — tone, length, audience, temperature — which is most of what editing actually is.
Weak prompt: "Make this email better." (pasted email)
Same prompt, with the word: "Rewrite this email to my coworker so it says the same things but sounds calm instead of irritated. Keep the deadline and the request exactly as they are."
Why it worked. "Better" is a lottery. Naming the one quality to change — temperature — and the two things to preserve turns the machine into a careful editor instead of a random one.
Another example.
Weak prompt: "Fix this text to my sister."
Same prompt, with the word: "Rewrite this text to my sister so it sounds warm instead of annoyed — but keep the boundary exactly where it is: I am not hosting Thanksgiving this year."
Why it worked. Naming what must survive — the boundary — kept the warmth from becoming surrender. Both things arrived in the same short message.
One more.
Weak prompt: "Make this note to the babysitter better."
Same prompt, with the word: "Rewrite the babysitter note so it sounds relaxed instead of anxious, but keep every time, number, and rule exactly as written."
Why it worked. The sitter got the same instructions at a lower heart rate — and so, reading it back, did the parents.
Watch for. Rewrite long enough and the content quietly migrates: the boundary you said must survive comes back softer, the deadline "exactly as written" returns as "soon." After any rewrite, diff it against the original yourself — what changed besides the thing you asked to change?
Placed well: "Rewrite this text warm instead of annoyed, but keep the boundary exactly: I am not hosting Thanksgiving."
What it makes the AI do. It asks for the same thing in deliberately different directions — not ten near-copies, but a spread of real choices.
Weak prompt: "Give me slogans for my bakery."
Same prompt, with the word: "Give me slogans for my bakery, but vary them on purpose: three that are funny, three that are nostalgic, three that are short enough for a sign, and three that mention bread specifically."
Why it worked. Machines repeat their first idea with synonyms. Varying by category forces genuinely different attempts — and a real spread is what makes the good one visible.
Another example.
Weak prompt: "Write a birthday message for my mom."
Same prompt, with the word: "Write birthday messages for my mom, but vary them on purpose: one funny, one grateful, one short enough for a card, and one that mentions her garden."
Why it worked. Four genuinely different doors into the same sentiment. One of them turned out to be exactly the one he had been trying to write for years.
One more.
Weak prompt: "Give me a caption for this photo of the lake."
Same prompt, with the word: "Give me captions, varied on purpose: one poetic, one funny, one plain factual, and one that's just three words."
Why it worked. The three-word one won. It would never have appeared in a run of similar tries, because machines, unasked, do not dare to be brief.
Watch for. Variety by category can still be variety in costume only: the funny slogan, the nostalgic slogan, and the short slogan all selling the same single idea the machine had first. If the spread feels familiar, name the idea you keep seeing — and ban it.
Placed well: "Vary these captions on purpose: one poetic, one funny, one factual, and one that's just three words."
What it makes the AI do. It makes creation a loop instead of a lottery ticket: version, reaction, version — each round building on your specific complaint about the last.
Weak prompt: "That logo description isn't right. Try again."
Same prompt, with the word: "Let's iterate on this logo description. Keep the mountain, change only the colors to feel more like morning than night, and make the lettering bolder. Everything else stays."
Why it worked. "Try again" throws away the 80% that was working. Iterating names what to keep and what to change, so each round compounds instead of restarting.
Another example.
Weak prompt: "This soup is bland. Fix it."
Same prompt, with the word: "Let's iterate on the soup. Round one: change nothing except seasoning — give me three seasoning directions. I'll pick one, and only then do we talk about texture."
Why it worked. Changing one variable at a time meant the final recipe could be repeated — and remembered. Random improvements can't be cooked twice.
One more.
Weak prompt: "My presentation isn't landing. Redo it."
Same prompt, with the word: "Let's iterate instead of redoing: keep the data slides exactly, change only the opening — give me three openings. Round two, we fix the ending with whichever opening wins."
Why it worked. The data was never the problem. Protecting it meant the energy went to the two minutes that were.
Watch for. Iteration drifts. Each round preserves "everything else," but everything else shifts a degree per round — the mountain migrates, the morning darkens — until round six is a stranger to round one. Keep the original beside the latest, and compare them directly every few rounds.
Placed well: "Iterate on this logo: keep the mountain, change only the colors toward morning, nothing else moves."
What it makes the AI do. It raises the register: rough notes, texts, and spoken thoughts become proper documents — letters, proposals, policies — dressed for the room they're entering.
Weak prompt: "Turn my notes into something I can send to the city council."
Same prompt, with the word: "Formalize these rough notes into a one-page letter to the city council about the potholes on Elm Street: respectful, specific about locations and dates, and ending with one clear request."
Why it worked. Ideas get ignored when they arrive dressed casually for a formal room. Formalizing keeps your substance and changes only the suit — which is often the whole difference.
Another example.
Weak prompt: "Turn these texts into a complaint to the airline."
Same prompt, with the word: "Formalize my notes into a complaint letter about the canceled flight: dates, flight numbers, what the agent promised me, and the refund I am requesting — firm, factual, and with no insults anywhere."
Why it worked. A formal letter enters a system that is built to answer formal letters. The same facts in a furious text enter a bin.
One more.
Weak prompt: "Make my complaint about the hotel sound professional."
Same prompt, with the word: "Formalize this complaint: the dates, the room, the noise complaint I made at 1 a.m. and what was promised, and the one night's refund I'm requesting — courteous throughout."
Why it worked. Hotels refund documented guests. The formal version was a document; the refund arrived with the reply.
Watch for. Formalizing can embalm a complaint: dates perfect, register impeccable, and the one human sentence that would have moved a reader filed off as unprofessional. Formal is a suit, not an embalming — tell the machine which sentence keeps its heartbeat before it dresses the rest.
Placed well: "Formalize these notes into a one-page council letter: respectful, specific, ending with one clear request."
What it makes the AI do. It suspends judgment on purpose: quantity first, criticism later, so unusual ideas survive long enough to be seen.
Weak prompt: "Give me some video ideas for my gardening channel."
Same prompt, with the word: "Brainstorm thirty video ideas for a small gardening channel — quantity first, no judging, wild ones welcome — then group them by type so I can pick."
Why it worked. Asking for "some ideas" gets the five obvious ones. Thirty forced the machine past its defaults, and the good idea was sitting at number twenty-two, where it usually lives.
Another example.
Weak prompt: "What should we do for our anniversary?"
Same prompt, with the word: "Brainstorm twenty-five anniversary ideas across five kinds — adventurous, quiet, silly, generous, and new-to-us — five of each, budget-blind for now."
Why it worked. Categories kept the brainstorm from producing twenty-five dinners. The budget could edit later; it could not have invented the hot-air balloon.
One more.
Weak prompt: "Ideas for my YouTube channel about woodworking?"
Same prompt, with the word: "Brainstorm forty video ideas, quantity first — builds, repairs, mistakes, history, viewer projects — and don't stop at the ones a channel like mine usually makes."
Why it worked. Idea thirty-one — rebuilding viewers' worst mistakes — became the series. It was never going to appear in a list of five.
Watch for. Quantity has a hangover: forty ideas, and the choosing — which was the job — now takes longer than the brainstorm did. Cap the flood by deciding the sieve in advance: three finalists, chosen against two named tests, before the list is allowed to cool.
Placed well: "Brainstorm thirty video ideas, quantity first, no judging — then group them so I can choose three."
What it makes the AI do. It writes in a named voice — a kind of person, a tone, a temperament — instead of the machine's default voice, which sounds like a brochure.
Weak prompt: "Write a note to customers about our holiday hours."
Same prompt, with the word: "Write the holiday-hours note in the voice of a friendly small-town shopkeeper: short sentences, warm, one joke, and the hours impossible to miss."
Why it worked. The default machine voice is competent and forgettable. A named voice gave the note a person to be — and customers write back to people.
Another example.
Weak prompt: "Write the instructions for watering my plants while I'm away."
Same prompt, with the word: "Write the plant instructions in the voice of a kind neighbor who's done this before: plain, reassuring, and tell her which plant is the drama queen."
Why it worked. A housesitter follows instructions she enjoys reading. The voice carried the care instructions past the fridge door and into actual watering.
One more.
Weak prompt: "Write a welcome message for my rental cabin."
Same prompt, with the word: "Write it in the voice of a mountain innkeeper who's seen every kind of guest: warm, unhurried, a little dry humor, and the Wi-Fi password woven in gracefully."
Why it worked. Guests began mentioning the note in their reviews. A voice, it turns out, is hospitality too.
Watch for. A named voice can tip into costume: the small-town shopkeeper acquires a corn-pone accent nobody requested, the innkeeper starts performing folksiness at the Wi-Fi password. Voice should season, not swamp — if you notice the accent more than the hours, dial it back one notch.
Placed well: "Write the holiday-hours note in a friendly shopkeeper's voice: short, warm, one joke, hours unmissable."
What it makes the AI do. It forces two unlike things together — a format and a subject, two hobbies, two tools — and makes the machine find the working hybrid instead of a safe single answer.
Weak prompt: "I like chess and I like cooking shows."
Same prompt, with the word: "Combine them: invent a show where two chefs compete like chess players — timed moves, an opponent's dish to respond to, a clock. Give me the format, the rules, and a sample episode."
Why it worked. Neither interest alone produced anything. The collision did — because formats are portable, and a chess clock turns out to be drama you can put in a kitchen.
Another example.
Weak prompt: "Help me exercise and learn history at the same time."
Same prompt, with the word: "Combine my daily walks with learning history: design a walking route plan where each week's walks come with one era to listen through, in order, with the total distance matched to the episodes."
Why it worked. Two goals were competing for the same hour. Combined, they stopped competing — the hour now pays both debts at once.
One more.
Weak prompt: "I want to practice guitar and spend time with my daughter."
Same prompt, with the word: "Combine them: design a Saturday ritual where she picks the songs and I learn them to play together — her requests, my practice, one hour, with a simple plan for the first month."
Why it worked. Practice acquired an audience of one who requests encores — the most durable motivation a guitar has ever had.
Watch for. Combinations fail charmingly: the chess-cooking show is a delight to describe and a misery to film, because collisions produce novelty faster than practicality. Before adopting a hybrid, ask the machine the unglamorous question: what does Tuesday look like, doing this, in week three?
Placed well: "Combine my walks with history: one era per week, episodes matched to my actual route distances."
What it makes the AI do. It writes the exact words for a moment you will actually face — the call, the ask, the apology — so you rehearse lines instead of improvising under pressure.
Weak prompt: "I have to call the insurance company about a denied claim."
Same prompt, with the word: "Script the call for me: my opening line, the calm version of my argument, what to say if they say no, and a polite closing that keeps the door open. Keep every line short enough to say out loud."
Why it worked. Under pressure, nobody thinks of the good sentence. A script moved the thinking to a calm afternoon, where it belonged.
Another example.
Weak prompt: "How do I tell my boss I need Fridays off?"
Same prompt, with the word: "Script it: the exact first two sentences for asking my boss about Fridays off, one sentence for the reason, and a ready answer for 'what about deadlines?'"
Why it worked. The whole conversation frightened him; the first two sentences were the actual mountain. Scripted, they turned out to be easy to say.
One more.
Weak prompt: "I need to apologize to my brother."
Same prompt, with the word: "Script the apology: first sentence that owns it without excuses, one sentence about what I'll do differently, and a closing that doesn't demand he make me feel better."
Why it worked. The script protected the apology from its two natural enemies — rambling and self-defense.
---
Watch for. A script is a rehearsal, not a possession: memorize it too rigidly and the unexpected reply strands you, line perfect and nowhere to go. Script the opening and the closing word-perfect; for the middle, script your intentions — the points you won't leave without making.
Placed well: "Script my opening line and my answer to 'what about deadlines?' — leave the middle as intentions."