Chapter 4 — Prompting as Professional Communication

“Garbage in, garbage out.” — a computing adage

“A word fitly spoken is like apples of gold in a setting of silver.” — Proverbs 25:11 (ESV)

Why This Matters

Somewhere in the last three years a myth took hold: that there is a secret language for talking to AI. That somewhere on the internet lives the magic phrase, the perfect incantation, the string of words that unlocks the good answers everyone else seems to be getting. People collect these phrases the way medieval alchemists collected recipes for gold. “Act as a world-class expert.” “Take a deep breath and think step by step.” “I’ll tip you $200.” They paste them in, hope for magic, and are quietly disappointed when the results are still mediocre.

Here is the truth that this whole chapter is going to earn: prompting is not incantation. It is communication. The reason your prompt got a vague answer is almost never that you missed a secret word. It is that you did not say what you wanted clearly enough for anyone — human or machine — to give you a good result. The vague prompt “write an email to my customers about the new system” would stump a brilliant new hire, too. Not because they lack talent, but because you did not tell them who you are, what the system is, why the customer should care, or how long the note should be. You did not communicate. So they guessed. And guesses are bland.

This reframe changes everything, and it is good news for you specifically. You do not need to become a programmer or memorize a spellbook. If you can brief a colleague — if you can hand a task to a capable assistant and tell them what “done well” looks like — you already have the core skill. The rest of this chapter just makes that skill deliberate. We are going to name the parts of a good brief, practice each one, and pack them into a repeatable structure you will use for the rest of your career: Role + Objective + Context + Examples + Format.

And here is where the thread that runs through this whole book runs through this chapter too. You choose the tool. You own the verdict. The assistant will draft, generate, and accelerate — but a clearer prompt does not make the output true, only more useful. A well-briefed assistant hallucinates just as confidently as a badly-briefed one; it simply does it in your preferred tone and length. So communication is half the job. Verification is the other half, and it never leaves your desk. AI is an assistant, not an authority.

Coach’s Note — If you are joining this book fresh, here is the house rule you will hear all term: learning is a sport, and the keyboard is the gym. You do not get better at prompting by reading about it any more than you get fit by reading about squats. Every chapter ends with Reps and a Project. The reading is the chalk talk. The reps are the game. Show up for both.

This week’s deeper question is a question about words. Scripture calls a word rightly spoken “apples of gold in a setting of silver” — a thing of real worth, fitted to its place. We spend our working lives trying to be understood, and mostly failing at the edges. So: what is a word “fitly spoken” — and why does clarity, the simple act of being truly understood, carry such weight? Hold that. We will give it its due near the end.

4.1 — Prompting Is Communication, Not Incantation

Let us kill the myth properly, because it is the single biggest thing standing between you and good results.

An AI assistant is not a search engine and it is not a genie. The closest everyday thing it resembles is a fast, widely-read, eager-to-please new colleague who has never met you, does not know your job, cannot see your screen, will not ask clarifying questions unless you invite them, and will confidently make something up rather than admit they are stuck. Read that sentence twice. Every technique in this chapter is just a sensible response to one of those traits.

• The colleague has never met you → so you tell it who to be and what your world is (Role, Context).
• It does not know what “good” means here → so you tell it the objective and the shape of the answer (Objective, Format).
• It learns fast from a sample → so you show it an example (Examples).
• It will not ask → so you either front-load what it needs, or you invite the questions (“Ask me anything unclear before you start”).
• It will confidently make things up → so you verify, always, no matter how clean the prose looks.

Notice what is not on that list: secret words. “Act as a world-class expert” helps a little, not because it is magic, but because it is a crude way of doing the Role step. You are going to do the Role step on purpose and far better. The tips-and-tricks people are stumbling toward the same handful of communication moves you are about to learn deliberately.

Coach’s Note — The industry adopted the phrase “prompt engineering,” and it did the field a disservice. “Engineering” makes it sound like tuning a machine to a spec. What you are actually doing is briefing a colleague — a human skill you have been building your whole life. When you feel intimidated by a prompt, drop the word “engineering” and ask instead: “If I were handing this to a sharp assistant on their first day, what would I need to tell them?” Answer that out loud and you have written your prompt.

4.2 — Start With a Clear Objective

Every good prompt begins with a decision you must make before you type: what does “done well” actually look like?

Most weak prompts fail here, at the very first step, and no amount of clever wording downstream can rescue them. “Write something about our new scheduling portal” has no objective — it names a topic, not a target. Compare:

• Topic (weak): “Write about the new portal.”
• Objective (strong): “Draft a short email that makes our patients feel reassured about the new portal and makes crystal clear the one action they must take.”

The second version names the reader’s end state (reassured, clear on the action) and the outcome that means success. That is an objective. The assistant can now aim. Without it, the assistant optimizes for the only thing it can infer — sounding vaguely on-topic — and you get filler.

A useful habit: finish the sentence “After reading this, the reader should ______.” Should feel what? Should know what? Should do what? If you cannot finish that sentence, you are not ready to prompt — you are still figuring out what you want, and no assistant can do that part for you. That is your judgment, and it is the part the machine cannot supply.

Objectives also carry the constraints that matter most: “under 180 words,” “no jargon,” “do not imply we are raising prices.” Constraints are not nagging — they are you doing the reader’s thinking in advance. We will formalize the “how long, what shape” constraints under Format in 4.6; the point here is that the objective is where you decide what a good answer even is.

4.3 — Give Context: The Half of the Message You Forgot to Send

Here is the failure mode I see most often, and it is subtle because it does not feel like a failure. You write a perfectly reasonable prompt, get back something generic, and conclude “AI just isn’t good at this.” Nine times in ten, the real problem is that you knew things the assistant could not possibly know, and you did not say them.

The assistant has read a large fraction of the public internet and knows nothing about your Tuesday. It does not know your company’s name, your patients’ age range, that you are emphatically not raising prices, that the deadline is March 1, that “the portal” is a scheduling tool and not a payroll system, or that your boss hates exclamation points. All of that lives in your head. Until you send it, the assistant is writing blind and filling the gaps with plausible guesses — which is exactly where invented “facts” come from.

Context is the cure. Before you prompt, ask: what would a competent colleague need to know to do this well? Then supply it:

• Audience — who reads this, and what do they already know? “Patients aged 40+, not especially tech-savvy” changes every word choice.
• Facts it cannot guess — names, dates, numbers, and constraints. The deadline. The one required action. The dollar figure. Get these in.
• What must be true, and what must not be said — “nothing is lost in the switch”; “do not imply a price increase.”
• Your voice — warm, formal, plain, our-house-style. If you have a sample of how you normally sound, that is gold (see 4.4).

There is a discipline here that this book will repeat until Week 16: context is not a place to paste secrets. A consumer chatbot is not a private notebook. Never paste customer records, patient health information, employee data, passwords, or anything confidential into a public AI tool. You can give the assistant enough context to write well without handing it anything that would get you in trouble — say “a patient aged around 65,” not a real name and chart. We treat this properly in Chapter 14 and in Appendix C; mark it now.

Coach’s Note — When an answer comes back generic, resist the reflex to blame the tool or hunt for a magic phrase. Ask instead: “What did I know that I didn’t tell it?” Nine times in ten, the fix is a sentence of context, not a cleverer verb. The best prompters are not the wittiest writers. They are the ones who reflexively include the three facts everyone else leaves in their own head.

4.4 — Show, Don’t Just Tell: Examples and Few-Shot Prompting

You can describe the tone you want with a pile of adjectives — “warm but professional, concise but friendly, confident but not salesy” — and still get something off. Adjectives are slippery. Showing is exact.

If you paste one short sample of the voice you are after — a past email you liked, a paragraph in your house style, one product description that nailed it — the assistant will match it far more reliably than it will match a stack of descriptors. This is the single highest-leverage move most professionals never make.

When you give the assistant one or more worked examples of the input-and-output you want, the technique has a name: few-shot prompting (“a few shots at the target”). Giving none — just describing the task — is “zero-shot.” Most everyday prompting is zero-shot, and it is fine. But the moment you care about a specific format or voice, a couple of examples are worth a paragraph of instruction:

• Tone matching: “Match the voice of this note I sent last month: [paste].”
• Format by example: “Format each item like this: Name — one-line benefit — price. Here are two done for you: [paste two]. Now do the other eight.”
• Category by example: “Sort each message as URGENT, ROUTINE, or IGNORE. Examples: ‘server is down’ → URGENT; ‘lunch menu’ → IGNORE. Now sort these twenty.”

The pattern is always the same: don’t just tell it the rule, show it the rule in action. One good example teaches faster than three sentences of description, because it removes the ambiguity that words alone leave behind. The provided file code/prompt-before-after.txt is itself an example in this sense — it shows you what a weak prompt and a strong prompt look like side by side, so the lesson lands in your eye and not just your ear.

4.5 — Assign a Role

A role is a shortcut. When you begin with “You are an experienced office manager who writes warm, plain-English notes to patients,” you have, in one sentence, set a voice, a standard, and a point of view — the assistant now knows roughly what vocabulary to reach for, what to emphasize, and what a good version looks like in that role.

Roles work because they let you borrow a whole cloud of expectations with a few words. Watch how the same request bends under different roles:

None of these is “correct” in the abstract — the right role is the one that fits your objective and audience. That is your call to make, which is the recurring lesson of this book: the tool can play any role you name, but choosing which role serves this reader is judgment, and judgment is yours.

A word of caution the tips-and-tricks crowd never mentions: a role changes the voice, not the facts. “You are a world-class cardiologist” makes the assistant sound authoritative. It does not make it right, and it does not give it a medical license. A confident role on a shaky fact is the most dangerous combination in this whole book — fluent, authoritative-sounding, and wrong. Assign roles to shape tone and framing. Verify substance regardless of the costume you dressed the assistant in.

Coach’s Note — The role is a costume, not a competence. Dressing the assistant as an expert improves how the answer reads; it does nothing for whether the answer is true. Treat a role-prompted answer with exactly the skepticism you would give a very polished stranger who claims to be an expert: pleasant, useful, and unverified until you check.

4.6 — Specify the Format

You would not tell a designer “make it look good” and walk away. Yet people routinely ask an assistant for help and leave the shape of the answer entirely to chance — then feel let down when they get five dense paragraphs and they wanted three bullets.

Say the shape. It is the cheapest, most reliable upgrade to any prompt:

• Length: “under 180 words,” “one paragraph,” “exactly three options.”
• Structure: “a subject line and a body,” “a bulleted list,” “a two-column table,” “headings for each section.”
• Must include / must end with: “end with our phone number,” “include a one-line summary at the top,” “give me three subject-line options.”

Format instructions do something subtler than tidy the output: they constrain the assistant toward your actual need. “Give me three subject-line options, each under eight words” forces variety and brevity you would otherwise have to ask for in a second round. “Put it in a table with columns for Option, Pro, and Con” forces the assistant to actually think in comparisons instead of mush. The shape you request shapes the thinking you get back.

This is also where you make the output usable without rework. If the thing needs to drop into a slide, ask for short phrases, not sentences. If it goes into an email, ask for a subject line. If you will paste it into a spreadsheet, ask for a table or comma-separated values. A little format discipline turns “nice draft I now have to reformat” into “done.”

4.7 — The Skeleton: Role + Objective + Context + Examples + Format

Now assemble the five moves into one repeatable structure. This is the backbone of the whole chapter, and it is worth committing to memory:

Role · Objective · Context · Examples · Format — R-O-C-E-F.

You will not use all five every time — a quick “summarize this in three bullets” needs only an objective and a format. But when the stakes are real, walking the five in order guarantees you have said everything a competent colleague would need. The provided file code/prompt-skeleton.txt is this skeleton as a fill-in-the-blank template; the provided code/prompt-before-after.txt shows a vague prompt rebuilt on the skeleton and how much it improves.

Here is the skeleton on one concrete task, so you can see it whole:

ROLE:      You are an experienced office manager at a small dental
           practice who writes warm, plain-English notes to patients.

OBJECTIVE: Draft a short email telling patients we switch to a new
           online portal on March 1. Goal: they feel reassured and
           know the one action they must take.

CONTEXT:   - Riverbend Family Dental; signed "Maria, Office Manager."
           - Patients are mostly 40+, not very tech-savvy.
           - Only action: click a setup link we email the week of
             Feb 24, then set a password.
           - History and insurance carry over; nothing is lost.
           - We are NOT raising prices — do not imply that we are.

EXAMPLES:  Tone to match: "Good news — we've made a small change to
           make your life easier..." Warm, "we," no jargon.

FORMAT:    Subject line + body, under 180 words, one short bulleted
           list of what changes, end with a friendly sign-off and
           our phone number.

Read the difference against “write an email to my customers about the new system.” Same assistant. Same afternoon. The gap between the two outputs is entirely the gap between those two messages — which is the entire thesis of this chapter stated in one example. You did not get better at magic. You got better at communicating.

Coach’s Note — Do not treat R-O-C-E-F as a form to fill out rigidly every time — that way lies stiffness. Treat it as a checklist you run in your head, the way a pilot runs a pre-flight check. Most days you will glance down the five and realize you already covered four of them; the value is catching the one you dropped, which is almost always Context or Format. The dropped one is why the answer disappointed you.

4.8 — Ask It to Show Its Work: Chain-of-Thought

You learned this in grade school: on a hard problem, “show your work.” It turns out the same instruction helps an AI assistant, and understanding why — at a plain-English level, no math — makes you better at knowing when to reach for it.

When you ask an assistant a question that needs several steps of reasoning — a comparison, a calculation, a decision with tradeoffs — and you let it answer in one leap, it often blurts a confident conclusion that skipped a step. But if you add “Think it through step by step before you give your answer,” or “First list the factors, then weigh them, then recommend,” the assistant tends to reason more carefully and land better answers on exactly those multi-step problems. The industry calls this chain-of-thought prompting. You do not need the jargon; you need the move: for anything that requires reasoning, ask it to reason out loud first, then conclude.

Two professional payoffs, beyond a better answer:

1. You can check the reasoning, not just the verdict. When the assistant shows the steps, you can see where it went wrong — it used the wrong number in step two, or weighed the wrong factor. A bare conclusion hides its own mistakes. Visible reasoning is auditable reasoning, and auditing is your job.
2. It surfaces hidden assumptions. Asked to lay out its thinking on “should we switch vendors,” the assistant will often reveal it assumed a budget, a timeline, or a priority you never gave it — assumptions you can now correct.

A caution, because this chapter keeps handing you sharp tools: showing plausible-looking work is not proof the work is correct. A chain of reasoning can be fluent, orderly, well-formatted — and built on a fabricated fact in step one that poisons everything after it. Chain-of-thought makes the reasoning visible, which makes it checkable. It does not make it true. That is still on you. We spend all of Chapter 8 on exactly this: how confident, well-reasoned-looking AI output can still be wrong, and how to catch it.

4.9 — Iterate: The Follow-Up Is Where the Value Lives

If you take one habit out of this chapter, take this one: the first answer is a draft, and the conversation is the tool. Most people write one prompt, judge the result, and either accept it or give up. Both are mistakes. The professionals getting real value from these tools treat the first output as the opening move, not the final word.

The assistant remembers the conversation. You do not have to re-explain — you just say what to change:

• “Warmer. It reads like a form letter.”
• “Half as long.”
• “You invented a 10% discount. We never said that. Remove it.”
• “Keep the second paragraph exactly; redo the rest.”
• “Give me three more subject-line options, punchier.”
• “Now rewrite that same email for our staff instead of our patients.”

Each follow-up is cheap and specific, and two or three of them will beat the most elaborate single prompt you could have written up front — because you cannot foresee everything you want until you see a draft to react to. This is exactly how you would work with a human assistant: hand off, review, redirect, review again. Iteration is not a sign your first prompt failed. It is the method.

There is a deeper reason iteration matters, and it points at the next chapter. The follow-up is where your judgment enters the loop. “You invented a discount — remove it” is you catching a fabrication. “Warmer, this reads cold” is you applying taste the assistant does not have. Every correction you make is a small act of ownership — the human deciding what is true and what is good. Delegate the drafting all you like. The corrections are the part that is unmistakably, accountably you.

Coach’s Note — Watch for the trap of the “good enough” first draft — the one that is 80% right and would take three quick follow-ups to make excellent. Fatigue whispers “ship it.” Don’t. The gap between the 80% draft and the finished piece is precisely the gap between “used AI” and “used AI well,” and your reader can feel it. Three follow-ups is thirty seconds. Spend them.

4.10 — Toward Reusable Prompts: Templates You Keep

Notice something about the good prompt in 4.7: most of it is reusable. The structure — role, objective, context, examples, format — is identical whether you are announcing a portal, a price change, or a holiday closure. Only the specifics change. That observation is the seed of a professional superpower we build fully in Week 12 (Chapter 12): your own prompt library.

The idea is simple. Once you have written a prompt that reliably produces good work — a solid “rewrite this email for a specific audience” prompt, a dependable “summarize this document into five bullets and a risk” prompt — you do not throw it away and reinvent it next Tuesday. You save it, with blanks where the specifics go:

You are a {{ROLE}} writing to {{AUDIENCE}}.
Draft {{WHAT}} so that the reader {{DESIRED OUTCOME}}.
Facts you must use: {{FACTS}}.
Match this tone: {{SAMPLE}}.
Format: {{LENGTH / SHAPE}}.

Fill the blanks, run it, done. You have turned a one-time effort into a reusable asset — a tool you built, sharpened once, and now reach for again and again. A team can share these. You can version them, improve them, and pass them to the next person who joins.

For now, in Week 4, you do not need a whole library — you need to feel the power of one good reusable prompt. The provided code/prompt-skeleton.txt is your first template. Keep it. Start a plain document — call it my-prompts.txt — and every time you write a prompt that works well, paste it in with a one-line note about what it is for. By Week 12 you will have a real library and the habit to maintain it. The wise, Scripture says, lay up knowledge; a saved good prompt is knowledge laid up.

4.x — Interactive Lab: Prompt Upgrader

Below this chapter on the website you will find an interactive panel called the Prompt Upgrader. Go use it now — it is not decoration, it is the rep that wires this whole chapter into your hands.

The panel starts you with a deliberately vague prompt — something like “write an email about the new system.” Along the side are five toggles: Role, Objective, Context, Examples, Format. Flip each one on and watch the prompt transform in front of you: switch on Role and a “You are a…” line appears; switch on Context and the facts the assistant could never guess snap into place; switch on Format and the shape of the answer gets pinned down. As you build it up, the panel shows you how the likely output sharpens from generic filler toward something you could actually send.

What it teaches is not a fixed template to memorize — it is the felt cause and effect between what you put in and what you get back. Toggle Context off and watch the prompt go blind again. Toggle it on and watch it see. After a few passes you will stop believing in magic words, because you will have watched, with your own hands, that every real improvement came from saying more of what a colleague would need to know — never from a secret phrase.

Run it twice. The first time, turn the toggles on one at a time and read how each one changes the prompt. The second time, start from the vague prompt and try to predict which single toggle will help most for that particular task — then flip it and see if you were right. Predicting before you measure is how the lesson moves from your eyes into your instincts.

In BoodleBox — The widget trains your eye; now take the skill into the tool Concordia actually gives you. BoodleBox is the campus AI platform — sign in with your Concordia account at box.boodle.ai. When you want help shaping a prompt instead of just answering one, start a chat and open the bot picker (type @) to bring in PromptBot, a helper built to draft and sharpen prompts with you. You can also switch on Coach Mode from the prompt-modes control (labels drift, so as of 2026 look for the coaching or “prompt modes” option) and it will nudge you toward the R-O-C-E-F piece you left out as you write. Draft a real prompt, let PromptBot or Coach Mode push back, and you will feel the same cause-and-effect the widget just showed you — this time on a prompt you actually intend to send. No BoodleBox access? The same move works in any free assistant: paste your draft prompt and ask it to critique it against Role, Objective, Context, Examples, and Format.

4.11 — A Word Fitly Spoken

Now the week’s question, given its due. What is a word “fitly spoken” — and why does clarity, being truly understood, carry such weight?

“A word fitly spoken is like apples of gold in a setting of silver” (Proverbs 25:11, ESV). Sit with the image, because the writer chose it with care. Not gold alone — gold set in silver, the right thing in the right setting, fitted to its place so that both are more beautiful for the fitting. The Hebrew behind “fitly” carries the sense of a word spoken upon its wheels — timely, turned to the moment, shaped for this hearer on this occasion. A true thing said clumsily, at the wrong time, to the wrong person, is not a word fitly spoken. Aptness is part of the worth. And Scripture rates that worth astonishingly high: precious metal, craftsmanship, a thing of real and lasting value. The Bible does not treat clear, fitting speech as a soft skill. It treats it as a form of gold.

That should reframe what you did all chapter. When you stopped to name your audience, to fit the tone to a 40-year-old patient who is a little nervous about technology, to say the one true thing plainly and cut the padding — you were not “optimizing a prompt.” You were practicing the ancient discipline of the fit word: the right thing, shaped for this hearer, in its proper setting. The skill this chapter teaches is, underneath, a moral skill as much as a technical one. To be understood is to serve. To be careless with words — to fire off the vague, the padded, the misleading, and make your reader do the work of decoding you — is a small failure of love toward the person on the other end.

The Lutheran tradition names this directly. Luther’s explanation of the Eighth Commandment in the Small Catechism does not stop at “do not lie”; it presses us to “defend” our neighbor, “speak well of him, and explain everything in the kindest way.” Explain everything in the kindest way. That is a communication standard, and it is higher than mere accuracy. A word can be technically true and still fail the neighbor — cold when it should be warm, confusing when it could be clear, careless of the reader’s fear. Clarity, in this light, is not efficiency. It is charity. The reason being understood carries such weight is that a person is on the other end of every message, and words are how we reach them — or fail to.

Which is exactly why the AI cannot own this part for you, and it is worth seeing precisely where the line falls. The assistant can format kindness — it will happily make a note “warmer” on command. But it does not know your neighbor. It does not know that this patient’s husband just died, that this employee is frightened about the reorganization, that this word will land on a wound. Fitting the word to the real person in front of you — that is the setting of silver, and it is yours to craft. The tool can polish the gold. Only you know the setting it goes into. This is the spine rule wearing its Sunday clothes: you choose the tool, you own the verdict — and the verdict here is whether the word was truly fit for the soul who receives it.

And there is a deeper reason a Christian takes words seriously, worth naming plainly. Our whole faith rests on a God who did not stay silent or aloof but communicated Himself — who “spoke” creation into being and then, at the fullness of time, spoke the final Word by taking on flesh so that we could understand (John 1:1, 14). The God we confess is a God who condescends to be understood, who fits His word to His hearers, who meets us where we are. When you labor to be clear — to say the true thing in a way your reader can actually receive — you are, in a small and ordinary way, imaging that. The apples of gold are worth the setting of silver. Do not let a fast tool talk you out of caring how the word lands.

4.12 — Common Pitfalls

Pitfall: Hunting for a magic phrase instead of communicating. Example: You paste “act as a world-class expert and take a deep breath” in front of a vague request and are baffled when the answer is still generic. Fix: There is no secret word. Ask what a competent colleague would need to know and say that — Role, Objective, Context, Examples, Format. The phrase-hunting is a symptom of a missing brief.

Pitfall: Leaving out the context that lives only in your head. Example: “Write an email about the new system” — no deadline, no audience, no “we are not raising prices” — and the assistant invents details to fill the void. Fix: Before you send, ask “what do I know that I haven’t told it?” Supply the audience, the facts it cannot guess, and what must not be said. Generic output is almost always a context problem, not a talent problem.

Pitfall: Accepting the first draft. Example: The opening answer is 80% right; you are tired; you ship it — misjudged tone and all — instead of spending three follow-ups. Fix: Treat the first output as the opening move. Say what to change: “warmer,” “half as long,” “remove the part you invented.” The follow-up is where the value lives; iteration is the method, not a failure.

Pitfall: Trusting a confident role or a tidy chain of reasoning as if it were proof. Example: “You are a top financial advisor” produces a fluent, authoritative recommendation with a specific figure — which turns out to be fabricated. Fix: A role changes the voice, not the facts; visible reasoning is checkable, not true. Verify every fact, number, name, and citation regardless of how expert or well-reasoned the answer sounds. Costume is not competence.

Pitfall: Pasting confidential or personal information into a consumer AI tool for the sake of “context.” Example: To get a better draft, you paste a real patient’s name, chart, and insurance details into a public chatbot. Fix: Give enough context to write well without the sensitive specifics — “a patient around 65,” not a real record. Never paste PII, PHI, credentials, or confidential material into public AI. See Appendix C.

Pitfall: Describing tone with adjectives instead of showing an example. Example: You ask for “warm but professional, concise but friendly” and get something that is none of those, because those words mean different things to you and to the model. Fix: Paste one short sample of the voice you want and say “match this.” One good example beats a paragraph of adjectives — that is what few-shot prompting is for.

4.13 — Reps

The work is in the exercises. The keyboard is the gym; the reading was the chalk talk, and this is where Week 4 gets into your hands. A preview of what is waiting:

• Upgrade a vague prompt into a full Role-Objective-Context-Examples-Format brief, and run both to feel the difference.
• Add context on purpose — take a generic answer and fix it with one sentence of the right context, not a cleverer verb.
• Match a voice with an example — paste a sample and make the assistant sound like you (few-shot).
• Iterate a draft across three deliberate follow-ups and log what each one bought you.
• Save your first reusable template into a my-prompts.txt — the seed of your Week-12 library.

Every rep that uses an AI tool ends with an honest one-line AI usage note: what you asked, what it got wrong or invented, and what you verified. That note is not busywork — it is the habit that keeps you the author and the AI the assistant. The human owns the verdict.

A short Check Your Reps quiz is embedded on this page, right under the chapter. Take it before you move on — five questions, grounded in exactly what you just read.

4.14 — This Week’s Project

Your project is P4 — “One Email, Three Ways,” specified in Project 4. You will take one real workplace message and prompt an AI assistant to write it three ways for three different audiences — the same underlying news, fitted to three very different readers — then judge which version lands and defend why. You will build each version on the R-O-C-E-F skeleton, iterate it with follow-ups, and verify that no version quietly invented a fact.

At a high level: Normal tier produces the three tailored emails, each with the prompt that made it, plus a short reflection on what changed between them and why. Medium tier adds a subject-line A/B test and a from-scratch reusable template. Hard tier is the part no assistant can do for you: a one-page memo recommending which version you would actually send to which audience and defending the call — the judgment that fits the word to the hearer. Full spec, starter files, and rubric are in Project 4.

4.15 — Coach’s Final Word

Here is what I want you to carry out of Week 4. You came in half-believing there was a secret language for talking to AI, a spellbook you were missing. There isn’t. What there is — and what you now have the shape of — is communication: telling a fast, willing, sometimes-wrong colleague who to be, what you want, what it cannot guess, what good looks like, and what shape to hand it back in. Role, Objective, Context, Examples, Format. Then the follow-up, where your judgment does its real work. That is not a trick. It is the oldest professional skill there is, pointed at a new kind of colleague.

And it is durable in a way the rest of this book’s contents are not. The model names in your fact brief will roll over by the time you read this twice; the prices will drift; the tools will merge and rename. The skill of being clear will not. A person who can say the true thing plainly, fitted to the one who needs to hear it, was valuable before any of these tools existed and will be valuable long after the current crop is forgotten. You are not learning to operate a 2026 product. You are sharpening the one thing the product cannot supply.

So do not chase phrases. Chase clarity. Say more of what a colleague would need. Show, don’t just tell. Iterate. Verify. And remember whose word it is when it goes out the door — because a word fitly spoken is worth gold, and the setting it goes into is yours to craft. You choose the tool. You own the verdict.

Now go do the reps. The Prompt Upgrader is waiting right below this page, the skeleton is in code/, and Project 4 is where it all comes together.

See you on Monday.

Up next: Read the exercises and do all of Week 4’s reps, then build Project 4 — Project P4: One Email, Three Ways. If you have not signed in to BoodleBox yet, do it now from Appendix A (it also covers the free fallbacks); the tool directory is Appendix B; the rules for what you may never paste into a public tool are in Appendix C; and any unfamiliar term — token, few-shot, chain-of-thought — is defined in Appendix D. Then Chapter 5 — Choosing the Right Tool for the Job.