Project P4 — One Email, Three Ways

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

Chapter: 4 — Prompting as Professional Communication Due: End of Week 4 Submit: A link to one shared folder or PDF portfolio (Google Drive, OneDrive, or a single exported PDF) containing your three emails, the prompt behind each, your comparison write-up, and your screenshots. Non-technical submission — no GitHub, no code. See Appendix A for account setup and submission mechanics. Allowed tools: BoodleBox — the AI platform Concordia licenses (sign in with your Concordia account at box.boodle.ai); run the whole project in one BoodleBox chat, and use PromptBot and Coach Mode to draft and refine your prompts (see Appendix A). Fallback: any one free consumer assistant reachable from a browser or phone — ChatGPT, Claude, or Gemini. A word processor for the final documents. No coding required at any step. AI policy (Phase 1): AI is used and evaluated, not blindly trusted. The assistant drafts every email; you decide the audience, verify every fact, correct what it invents, and choose what ships. Every deliverable carries an honest AI usage note. Never paste confidential or personal data into a public tool — invent realistic details instead. The human owns the verdict.

The Setup

You are Maria Delgado, the office manager at Riverbend Family Dental, a small five-chair practice in a mid-size town. There is no marketing department and no communications team. There is you, a front desk, three dentists, two hygienists, and about two thousand patients who trust this office with their Saturday-morning toothaches.

Next month the practice goes live on a new online patient portal — patients will book appointments, see reminders, and pay bills online instead of by phone. It is a genuine improvement, and it is also exactly the kind of change that makes a 70-year-old patient nervous and a busy front-desk teammate anxious about what breaks on day one. The same news has to reach three very different people, and if you send all three the same email, you will alarm the patients, under-prepare the staff, and bury the one thing the practice owner actually needs to decide.

So you are going to write one message, three ways — the same underlying news, fitted to three readers:

1. The patients — warm, reassuring, plain-English; they need to feel safe and know the one thing they must do.
2. The front-desk staff — direct and operational; they need to know what changes for them, what to say to patients, and when training happens.
3. The practice owner / lead dentist — a tight executive brief; they need the timeline, the risks, and the one decision you need from them.

Your job is not to let an AI write three emails and call it done. Your job is to communicate — to fit each word to its hearer — and to prove, in writing, that you know why each version lands.

Setup (the starter)

This chapter ships two starter artifacts in code/. You will lean on both:

• code/prompt-skeleton.txt — the reusable Role · Objective · Context · Examples · Format template. Every one of your three prompts is built on this. Copy it, fill the five blanks three times.
• code/prompt-before-after.txt — a worked example of a vague prompt rebuilt into a strong one, for this exact dental-practice scenario. Your model for “what a good prompt looks like” before you write your own.

Copy these into your own working document. Do not submit them unchanged — make them yours.

In BoodleBox — Do all three emails in a single chat: sign in at box.boodle.ai with your Concordia account, start a chat, and paste the skeleton in as your scaffold. When you build each R-O-C-E-F prompt, @-mention PromptBot (type @ to open the bot picker) to co-draft it, or switch on Coach Mode to be nudged toward the part you left out — then you write the final brief and send it. Keeping all three emails in one chat means the assistant remembers the news, so each follow-up is cheaper. Fallback: if you can’t reach BoodleBox, any free assistant (ChatGPT, Claude, Gemini) runs the same project.

Learning Targets

You will demonstrate that you can:

• Turn one piece of news into three genuinely audience-fitted messages, not three lightly-reworded copies of the same email.
• Build each prompt on the R-O-C-E-F skeleton and show the prompt that produced each result.
• Iterate a draft with deliberate follow-ups, and record what each follow-up bought you.
• Verify AI output — catch any fact the assistant invented and prove no fabrication shipped.
• Explain, in writing, why each version fits its reader — the judgment that no assistant can make for you.

Normal Tier

Goal: Produce three audience-fitted versions of the portal announcement, each with the prompt that made it, plus a short comparison and an honest AI-usage note.

Required features

1. Three emails — one each for patients, front-desk staff, and the practice owner. Same underlying news (new portal, go-live date, what it does), three genuinely different treatments of tone, length, detail, and call-to-action. Invent any specific facts you need (exact dates, the practice phone number) — do not use a real business’s private information.
2. The prompt behind each email — for all three, include the actual prompt you used, built on the R-O-C-E-F skeleton from code/prompt-skeleton.txt. Label each part (Role / Objective / Context / Examples / Format) so the grader can see the structure.
3. One documented iteration per email — for each of the three, show at least one follow-up you sent to improve the draft (“warmer,” “half as long,” “you invented a discount — remove it”) and one sentence on what it changed. Keep the before and after.
4. A verification pass — read each final email with a cold eye and confirm no invented facts survived (a fake discount, a wrong date, a promise you never made). Note anything the assistant fabricated and how you caught it. If it invented nothing, say so — and say how you checked.
5. A comparison write-up (250–400 words) — put the three side by side and explain what changed between them and why: which word choices, which details you added or cut for which reader, and which single version you are proudest of. This is the heart of the project.
6. An AI usage note — one honest paragraph: which tool/model you used (e.g., “BoodleBox running Claude Opus 5, mid-2026” — note which bot you @-picked), what it did well, what it got wrong or invented, and what you verified.

Normal-tier rubric (out of 100)

Medium Tier (+up to 25% extra credit)

Push from “wrote three emails” to “built a small communication system.”

1. Subject-line A/B test. For the patient email, generate three subject-line options, predict which will get the most opens before you decide, then pick one and defend the choice in two sentences. (Prediction-before-measurement is the habit.)
2. A reusable template from scratch. Turn your best prompt into a blanks-and-variables template ({{AUDIENCE}}, {{NEWS}}, {{ONE ACTION}}, {{TONE}}, {{FORMAT}}) saved in a my-prompts.txt, and prove it works by filling it for a fourth, unrelated message (e.g., a holiday-closure notice). This is the first entry in the prompt library you build fully in Week 12.
3. A “context off” comparison. Re-run the patient email prompt with the Context section deleted, and paste both outputs. In two sentences, name exactly what the missing context cost you.

Hard Tier (+up to 25% additional extra credit)

This is the judgment the machine cannot make for you, and it is graded as such.

Write DECISION-MEMO.pdf (one page, addressed to the practice owner) that makes and defends a single recommendation: given a real constraint — the go-live got moved up two weeks and you now have time to send only one of the three emails before launch — which audience do you send to first, and why?

You must:

• Name the audience you would prioritize and the specific reader-need that makes them first (a nervous patient who no-shows costs a chair; an unprepared front desk breaks on day one; an uninformed owner can’t defend the change).
• State plainly what you are accepting as a cost of not reaching the other two audiences yet, and your fallback for them.
• Tie the recommendation to the chapter’s spine — a word fitly spoken is fitted to its hearer, and you choose which hearer most needs the word first — and to the spine rule: you choose the tool, you own the verdict. The AI drafted the emails; the sequencing decision is yours to own and defend.

A memo that says “send all three” or “it doesn’t matter” fails. The grade is in the discrimination: showing you can weigh three real readers against a real constraint and defend the line. No assistant can make this call for you, because it does not know your practice, your patients, or what is truly at stake on go-live morning.

Submission

Put everything in one shared folder or a single exported PDF and submit the link per Appendix A:

P4-one-email-three-ways/
  email-1-patients.pdf        (or .docx)
  email-2-staff.pdf
  email-3-owner.pdf
  prompts-and-iterations.pdf  (the R-O-C-E-F prompt + follow-ups for each)
  comparison.pdf              (the 250–400-word write-up + AI usage note)
  screenshots/                (a few images of your actual AI conversation)
  DECISION-MEMO.pdf           (Hard tier only)
  my-prompts.txt              (Medium tier only)

Deliverables are ordinary documents — .docx, .pdf, .txt, and screenshots. No code, no repository, no command line. If your folder link is private, set it to “anyone with the link can view” before you submit. Never include a real patient’s, customer’s, or employee’s private information — invent realistic stand-ins.

Hints

• Decide the audience before you prompt. The whole project fails if the three emails are the same email in three fonts. Write down, for each reader, one sentence: “After reading this, they should feel ___, know ___, and do ___.” Three different answers → three different emails.
• Fill the skeleton three times, don’t freewrite. code/prompt-skeleton.txt exists so you don’t stare at a blank box. The Context block is where the three prompts diverge most — that is the point.
• Make it invent something, then catch it. If you give the assistant thin context, it will fill gaps — a discount, a “limited time,” a wrong date. That is not a disaster; it is the lesson. Let it happen once, catch it in your verification pass, and write it up. A caught fabrication is worth more in your report than a clean draft.
• The follow-up is the tool. Do not try to write the perfect prompt in one shot. Get a draft, then steer: “warmer,” “shorter,” “add the deadline.” Screenshot the steering — that is your iteration evidence.
• The comparison write-up is the grade. Anyone can generate three emails. The points are in explaining why the patient version is shorter and warmer and the owner version leads with the decision. That “why” is you thinking like a communicator.

What Mastery Looks Like (Beyond the Rubric)

A mastered submission is one where I can read your three emails without labels and instantly know which is for the patient, which for the staff, and which for the owner — because the words themselves are fitted to each reader, not just the salutation. Your comparison write-up shows you understand that the differences are choices you made for real people, and your verification note proves you read every line with a cold eye and caught what the assistant made up. The three emails are table stakes. The judgment about who needs which word, and the honesty about where the AI drifted — that is the project.

Coach’s Note — The temptation this week is to let the assistant’s first three drafts stand because they “read fine.” Resist it. “Reads fine” is where fabricated dates and misjudged tone hide. The professional move is to slow down at the exact moment the draft looks done and ask: is this word truly fit for this reader, and is every fact in it actually true? That two-minute pause is the entire difference between using AI and using it well — and it is the part with your name on it.

When You’re Done (a short checklist)

• Three emails that a stranger could sort by audience without the labels.
• Each email’s prompt shown, built on R-O-C-E-F with the five parts labeled.
• At least one documented follow-up per email, with what it changed.
• A verification pass: fabrications caught (or their absence confirmed) and how.
• A 250–400-word comparison explaining why each version fits its reader.
• An honest AI usage note; no real private data anywhere in the work.
• (Hard) DECISION-MEMO.pdf picks one audience to reach first and defends it.
• Everything in one shareable, view-enabled folder or PDF.

A theological footnote. “A word fitly spoken is like apples of gold in a setting of silver” (Proverbs 25:11, ESV). This project is a small, deliberate rehearsal of that verse. The gold is the true news — the portal, the date, the one action. The setting of silver is the fitting: the same true thing shaped warmly for a frightened patient, plainly for a busy teammate, briefly for a deciding owner. An assistant can polish the gold in seconds. Only you know the three people it is for, and only you can judge whether the word truly fits each one. That fitting is not decoration; Luther’s explanation of the Eighth Commandment presses us to “explain everything in the kindest way” — clarity offered as a small act of love to the reader. The assistant drafts. You fit the word to the hearer. You put your name on what goes out. That is a word fitly spoken.

See you next week.