Ask precisely. These words replace "tell me about" with an actual assignment.
Before the words, a word about the disease they cure. Most questions asked of machines are not questions at all — they are topics, dropped on the counter like a bag of unmarked parts. "Tell me about mortgages." "Thoughts on retirement?" The machine, eager and literal, does the only thing it can with a topic: it writes the encyclopedia entry, average in every direction, useful in none. Then the human sighs at the blandness and blames the cook for a recipe that said only "food."
Sharpening is the opposite habit. A sharp question has a size, an angle, and a job to do. It knows how many items it wants, in what order, against which standard, and for whom. The eight — now twelve — words in this part are all ways of putting an edge on the ask before the machine starts cutting. None of them requires knowing more about the subject. They require knowing more about your own errand, which, it turns out, you usually do — you just hadn't been asked to say it out loud.
Start here if your answers keep arriving true but useless. That combination is the signature of a blunt question.
What it makes the AI do. It forces the answer to commit to particulars — names, numbers, places, materials — instead of floating at the level of general advice.
Weak prompt: "How do I make my kitchen better?"
Same prompt, with the word: "Specify five changes under $300 each that would make a small galley kitchen easier to cook in — name the exact product type and approximate price for each."
Why it worked. "Better" gave the machine nothing to aim at. A budget, a count, and a demand for exact items turned a lifestyle essay into a shopping list you can act on today.
Another example.
Weak prompt: "Help me plan dinners for the week."
Same prompt, with the word: "Specify five weeknight dinners for two adults: each under 40 minutes, no seafood, chicken used twice at most — and list the main ingredients for every dish."
Why it worked. "Plan dinners" returns enthusiasm and a theme night. A count, a clock, and exclusions return a menu you can write a grocery list from in one pass.
One more.
Weak prompt: "What should I plant in my garden?"
Same prompt, with the word: "Specify a spring planting list for a 4-by-8 raised bed in Ohio: eight vegetables, with spacing and the week to sow each."
Why it worked. A bed has dimensions; the answer finally did too.
Watch for. Specifics can pile up until the prompt becomes a contract nobody can fulfill — twelve exact demands, three of them pulling against each other. Specify the details that decide the job; leave the machine a little room to solve the parts you don't care about.
Placed well: "Specify the exact documents, fees, and forms I need to bring to renew a driver's license in Illinois."
What it makes the AI do. It converts opinions into numbers — costs, times, probabilities, percentages — so claims can be compared and checked.
Weak prompt: "Is it expensive to own an older car?"
Same prompt, with the word: "Quantify the yearly cost of owning a ten-year-old sedan versus a three-year-old one: typical repairs, insurance difference, and fuel, in dollar ranges."
Why it worked. "Expensive" is a feeling. Dollar ranges are a decision tool. Once both options carry numbers, the better choice usually points at itself.
Another example.
Weak prompt: "How long does it take to learn Spanish?"
Same prompt, with the word: "Quantify learning conversational Spanish for an English speaker: the study hours it typically takes, and what 20 minutes a day reaches in 3, 6, and 12 months."
Why it worked. "How long" invites a shrug with a number in it. Hours and milestones turn the answer into a plan with checkpoints you can actually stand on.
One more.
Weak prompt: "Is walking good exercise?"
Same prompt, with the word: "Quantify walking for a sedentary 55-year-old: calories per mile, weeks to reach 8,000 steps a day from 2,000, and the point where the habit typically sticks."
Why it worked. Encouragement is free and worth it. Milestones are a schedule.
Watch for. Numbers can look like knowledge while resting on air — the machine will happily give you "about 30%" for something nobody has ever measured. Treat quantified guesses as estimates with good posture: useful for comparing, dangerous for engraving on a contract.
Placed well: "Quantify the true yearly cost of this car — fuel, insurance, repairs, and depreciation — in dollar ranges."
What it makes the AI do. It breaks a lump answer into a complete, countable list — nothing merged, nothing hidden inside a paragraph.
Weak prompt: "What do I need for a home office?"
Same prompt, with the word: "Itemize everything needed for a basic home office, one line per item, with a budget option and a mid-range option for each."
Why it worked. Paragraphs let the machine skip things gracefully. A list exposes the gaps — you can count twelve items, price them, and notice the one it forgot.
Another example.
Weak prompt: "What should I pack for the hospital?"
Same prompt, with the word: "Itemize a hospital bag for a two-night stay after surgery: documents, clothing, toiletries, and comfort items, one line each, nothing combined."
Why it worked. In a stressful week, paragraphs blur. A countable list can be packed against, line by line, with a pen in hand — and the missing line shows itself before the trip does.
One more.
Weak prompt: "What do I need to renew my driver's license?"
Same prompt, with the word: "Itemize exactly what to bring to renew a driver's license in my state: documents, fees, and what happens with each one if I forget it."
Why it worked. The last column — the cost of forgetting — is the part nobody lists and everybody needs.
Watch for. A list can be complete and still useless: forty items, no order, the trivial parked beside the critical, and two entries secretly the same item in different hats. After itemizing, ask which three lines would hurt most if they went missing.
Placed well: "Itemize everything in my garage-sale pile by room, one line per item, with a rough price for each."
What it makes the AI do. It puts two or more options side by side against the same criteria, instead of describing each one in flattering isolation.
Weak prompt: "Tell me about electric cars and hybrids."
Same prompt, with the word: "Compare a full electric car and a plug-in hybrid for a driver who goes 40 miles a day and takes one 600-mile trip a month: purchase price, fuel cost, maintenance, and inconvenience."
Why it worked. Described separately, everything sounds good. Judged on identical criteria for an identical driver, the trade-offs become visible — and trade-offs are what decisions are made of.
Another example.
Weak prompt: "I have two job offers. Which is better?"
Same prompt, with the word: "Compare these two job offers on salary, commute, health-care cost, vacation, and room to grow — for someone whose top priority is time with family, not title."
Why it worked. "Better" splits five ways at once. Naming the criteria — and naming the priority honestly — lets one offer win for a stated reason instead of a vague feeling.
One more.
Weak prompt: "Should I get a dog or a cat?"
Same prompt, with the word: "Compare a dog and a cat for a nurse working 12-hour shifts in a small apartment: daily time, cost, noise, and what each animal's worst ordinary day looks like."
Why it worked. The worst ordinary day decided it — which is what comparison, done honestly, is for.
Watch for. Comparison quietly fails when the criteria are smuggled in afterward — each option praised for whatever it happens to be good at. Fix the yardsticks before the first answer arrives, and make every option be measured by all of them, especially its weakest one.
Placed well: "Compare these two apartments on rent, commute, noise, and light — the same four measures for both."
What it makes the AI do. It forbids the diplomatic tie. The machine must order the options and, in doing so, reveal what it thinks matters most.
Weak prompt: "What are some good ways to save money?"
Same prompt, with the word: "Rank these six ways to save money for someone earning $45,000 a year — cooking at home, refinancing a car loan, canceling subscriptions, buying store brands, negotiating bills, and a side gig — by dollars actually saved per hour of effort."
Why it worked. Unranked lists pretend all advice is equal. A ranking exposes priorities, and the criterion — savings per hour of effort — is the part worth arguing with.
Another example.
Weak prompt: "What do people look for in a house?"
Same prompt, with the word: "Rank these features by resale value added in a starter home: a kitchen update, a second bathroom, a finished basement, a deck, and a new roof."
Why it worked. A ranking turns taste into an order, and an order is something you can spend money against one line at a time.
One more.
Weak prompt: "What are the best exercises?"
Same prompt, with the word: "Rank these six exercises for a beginner with bad knees and no gym: walking, swimming, cycling, yoga, light weights, and a rowing machine — best first, by benefit per week of effort."
Why it worked. The criterion did the work; the order was merely where it landed.
Watch for. A ranking is only as honest as its criterion, and the machine will choose a flattering one if you don't supply yours. "Best" ranked by novelty looks nothing like "best" ranked by cost per year. Name the measure first; argue about the order second.
Placed well: "Rank these five laptops by battery life per dollar, best first, and show the number behind each place."
What it makes the AI do. Similar to ranking, but aimed at action: it sorts tasks by what should happen first, given limited time or money.
Weak prompt: "My house needs a lot of work. Where do I start?"
Same prompt, with the word: "Prioritize these house repairs for a $4,000 budget before winter: a leaking roof patch, drafty windows, an old furnace, a cracked driveway, and peeling exterior paint. What gets done, what waits, and why?"
Why it worked. Ranking asks "which is best?"; prioritizing asks "which is first, and what does waiting cost?" The second question is the one real life actually asks.
Another example.
Weak prompt: "I have one week to study for my exam."
Same prompt, with the word: "Prioritize these six topics for a week of evening study before my certification exam, weighted by how heavily each is tested and how weak I am in it."
Why it worked. Equal time for everything feels fair and fails quietly. Prioritizing spent the scarce evenings where the points and the weakness actually were.
One more.
Weak prompt: "I owe money on four cards. Help."
Same prompt, with the word: "Prioritize these four debts for payoff: give me the order twice — once by interest rate, once by smallest balance — and say honestly which order a real person is likelier to finish."
Why it worked. Two orders exposed that the best plan is partly arithmetic and partly psychology — and named both.
Watch for. Everything marked urgent means nothing is. Prioritizing fails when the list quietly grows back — every waiting task re-argued as essential by Friday. Protect the order you made: a new "first" thing has to name what it is displacing, out loud.
Placed well: "Prioritize these repairs for a $4,000 budget: what happens before winter, what waits, and what waiting costs."
What it makes the AI do. It demands a concrete instance for every abstract claim, so advice has to survive contact with a real example.
Weak prompt: "How can I be a better manager?"
Same prompt, with the word: "Give me four ways to be a better manager, and exemplify each one with the exact words I could say to an employee in a real situation."
Why it worked. Abstract advice is easy to nod at and impossible to use. An example you can copy — actual sentences, in an actual moment — crosses the gap between knowing and doing.
Another example.
Weak prompt: "Explain what a 'professional tone' means."
Same prompt, with the word: "Exemplify a professional tone: take this furious sentence I want to send my supplier, and show me three professional versions of it — mild, standard, and firm."
Why it worked. Tone is easier to copy than to define. Three versions side by side show the dial, not just the setting — and let you pick where on it you want to stand.
One more.
Weak prompt: "Teach me to give feedback without being a jerk."
Same prompt, with the word: "Exemplify kind, direct feedback: write the exact sentences for telling an employee his report was late and sloppy — firm about the work, clean about the person."
Why it worked. He did not need a philosophy of feedback. He needed Tuesday's sentences.
Watch for. Examples can seduce: one vivid story starts standing in for the whole truth. A manager's great sentence worked in her meeting, with her employee. Ask for the example, then ask what about it actually transfers to yours — and what doesn't.
Placed well: "Exemplify each tip with the exact sentence I could say to a customer who is angry about a late order."
What it makes the AI do. It pins slippery words down before the conversation builds on them, so you and the machine stop using the same term for different things.
Weak prompt: "Explain investing risk like I'm a beginner."
Same prompt, with the word: "Define 'risk' the way an investor means it, then define it the way a gambler means it, and show me one decision where the two definitions give opposite advice."
Why it worked. Half of all confusing answers are two definitions arguing inside one word. Defining first — and contrasting the definitions — clears the fog before it costs you anything.
Another example.
Weak prompt: "Is this cereal healthy?"
Same prompt, with the word: "Define 'healthy' three ways — a dietitian's, a food marketer's, and a diabetic's — then judge this cereal under each definition separately."
Why it worked. The label was doing the arguing. Three definitions exposed whose game the word was playing, and the cereal's verdict changed with the judge — which was the real information.
One more.
Weak prompt: "What does 'organic' mean on eggs?"
Same prompt, with the word: "Define 'organic' as the regulator defines it for eggs, then as the shopper imagines it, and show me the gap between the two in one carton."
Why it worked. The gap between the definition and the picture is exactly where the money is spent.
Watch for. Defining every word in advance turns a simple question into a treaty negotiation, and the machine will cheerfully define terms you both already understood. Reserve the full treatment for the one word the whole answer leans on — usually the vaguest one in your prompt.
Placed well: "Define 'affordable' for this budget as a number, then judge every option against that definition only."
What it makes the AI do. It demands the exact spot — the precise cause, moment, place, or line — instead of a tour of the whole neighborhood where the problem lives.
Weak prompt: "My car is making a weird noise."
Same prompt, with the word: "Pinpoint the most likely source of this noise: a clicking that speeds up with the car, not with the engine, and only when turning left. Give me the two likeliest causes and how to tell them apart."
Why it worked. A noise is a mystery; a noise with three coordinates is a diagnosis waiting to happen. Pinpointing turned "something's wrong" into a short list with tests attached.
Another example.
Weak prompt: "This essay drags somewhere in the middle."
Same prompt, with the word: "Pinpoint where this essay starts to drag — quote the exact sentence where a reader would first check their phone, and tell me why that sentence."
Why it worked. "Somewhere in the middle" cannot be edited. One quoted sentence can — and the reason behind it usually fixes the paragraph around it too.
One more.
Weak prompt: "Sales were down this quarter."
Same prompt, with the word: "Pinpoint the week sales turned, the product line that moved first, and whether the drop started before or after the price change."
Why it worked. A quarter is a fog. A week, a line, and a sequence are an investigation.
Watch for. Precision can be faked. Asked to pinpoint, a machine may name one exact sentence, week, or bolt with total confidence and no real evidence. Treat the pinpoint as its best suspect, not its verdict — and ask what test would confirm the spot.
Placed well: "Pinpoint the exact paragraph where this proposal loses the reader, and quote the sentence where it happens."
What it makes the AI do. It reverses the flow: instead of lecturing you, the machine questions you, one good question at a time, until your own thinking gets sharper.
Weak prompt: "Help me decide whether to go back to school."
Same prompt, with the word: "Probe my thinking before advising me: ask me one question at a time about why I want this degree, what it costs, and what changes if it works — and don't recommend anything until you've asked at least five."
Why it worked. Advice given before the questions is a guess about a stranger. Probing made the machine learn the actual decision first — and half the answer arrived inside his own replies.
Another example.
Weak prompt: "Why do I keep procrastinating?"
Same prompt, with the word: "Probe it with me: ask me, one at a time, what exactly I avoid starting, what I do instead, and what I tell myself right before I stop."
Why it worked. A lecture on procrastination is easy to agree with and ignore. Questions made him produce the evidence himself — and evidence is harder to argue with.
One more.
Weak prompt: "Help me write my resume."
Same prompt, with the word: "Before writing anything, probe me: ask one question at a time about what I actually did in my last job — not my title, my Tuesdays."
Why it worked. "What were your Tuesdays?" produced facts no job title ever contained — and resumes are made of facts.
Watch for. Probing has a cost: questions. Let it run and you'll answer twenty before receiving one piece of advice, some of them circling the same drain. Cap it — five questions, then a summary of what it learned — so the interrogation ends in a decision.
Placed well: "Probe my plan with five questions, one at a time, then tell me what my answers reveal."
What it makes the AI do. It fixes one example, standard, or reference point and makes the machine judge everything else against it, so taste stops drifting from answer to answer.
Weak prompt: "Make this flyer for my bake sale better."
Same prompt, with the word: "Anchor on the church picnic flyer I pasted below — the one that worked. Match its length, its energy, and its layout logic, but write mine for a bake sale."
Why it worked. "Better" wanders; an anchor holds still. Every suggestion could be measured against a real thing that had already succeeded once.
Another example.
Weak prompt: "Is this a fair price for a used piano?"
Same prompt, with the word: "Anchor your judgment on this: the same model sold locally last month for $800 in good condition. Now judge the $1,300 asking price against that anchor, and tell me what would justify the difference."
Why it worked. Prices float until something heavy is tied to them. One real sale turned an opinion into a negotiation position.
One more.
Weak prompt: "How should I price my handmade quilts?"
Same prompt, with the word: "Anchor on the quilt I sold last spring for $300 in four days. Price this larger one against that anchor and tell me what the extra size and time honestly add."
Why it worked. Her own sale was worth more than the internet's opinion — it was the one price the market had already cosigned.
Watch for. Anchors hold things still — including your mistakes. Anchor on a flyer that worked for the wrong audience, or one lucky sale, and every later judgment inherits the error. Pick anchors that earned their authority, and replace them when a better one shows up.
Placed well: "Anchor on the email that got us three replies last spring; match its length and tone for this one."
What it makes the AI do. It aims the answer at one specific reader — their age, knowledge, and situation — instead of the general public, who do not exist.
Weak prompt: "Write something about saving for retirement."
Same prompt, with the word: "Target this at a 28-year-old earning $52,000 who has never saved anything and finds finance writing insulting. Five hundred words, zero jargon, one first step she can do this week."
Why it worked. Writing for everyone reads like writing for no one. A named reader forced real choices — about tone, about difficulty, about what "first step" could honestly mean.
Another example.
Weak prompt: "Explain how to use this new software."
Same prompt, with the word: "Target this explanation at our warehouse staff, who will read it once, on a phone, standing up. Short lines, big steps, no computer vocabulary."
Why it worked. The situation did the editing: one reading, standing up, on a phone. The manual that resulted could actually survive the warehouse.
One more.
Weak prompt: "Write an announcement about the schedule change."
Same prompt, with the word: "Target this announcement at night-shift staff reading it once, half-awake, at 2 a.m.: the new hours in the first line, nothing subtle anywhere."
Why it worked. The 2 a.m. reader edited the announcement more ruthlessly than any editor could.
---
Watch for. A target reader can harden into a caricature — the machine writing down to a "busy mom" or an "old-timer" it invented. Target the reader's situation and knowledge, which are facts, rather than their personality, which is a guess wearing a costume.
Placed well: "Target this at new hires reading it on a phone between shifts: short lines, steps numbered, nothing buried."