Chapter 4 — Reps
Ten reps to turn “prompting” from a hopeful guess into a repeatable skill. They move from upgrading a vague prompt, to adding context and examples, to iterating a draft, to saving your first reusable template. The keyboard is the gym. Do not read these — run them in a real AI tool.
Ground rules
- Type it yourself. Write each prompt out by hand, in your own words. Copy-pasting a prompt teaches your clipboard, not you. The skill is composing the brief, not owning the brief.
- Run everything. Every rep goes into a real assistant (ChatGPT, Claude, or Gemini — free tiers are plenty; see Appendix A). A rep you only thought about is a rep you did not do.
- Predict before you measure. Before you run a prompt, write one sentence: what do I expect back? Then compare the real output to your prediction. The gap is the lesson.
- AI usage note (every rep). Each rep that uses AI ends with one honest line: what you asked, what it got wrong or invented, and what you verified. This keeps you the author and the tool the assistant. The human owns the verdict.
- Never paste secrets. No real names, customer data, patient info, or confidential material into a public tool. Use “a patient around 65,” not a real record. See Appendix C.
- Keep a
reps.txt. One short written reflection per rep, in a plain document. This is graded thinking, not busywork.
Reps 1–3: From incantation to communication
Rep 1 — Run the vague prompt and name what’s missing
In your assistant, run the deliberately weak prompt exactly as written: “Write an email to my customers about the new system.” Read what comes back. Then list every question a competent colleague would have had to ask you before they could actually write that email well (who are you? what system? what deadline? how long?).
Reflection: How many questions did you list? Every one of them is a piece of context you were supposed to supply. Which single missing piece hurt the output the most?
Rep 2 — Upgrade it with the full skeleton
Open code/prompt-skeleton.txt. Rebuild that same email prompt using all five parts — Role, Objective, Context, Examples, Format — inventing a realistic scenario of your own (a gym, a library, a small shop; not your real employer’s confidential details). Run the upgraded prompt in the same assistant.
Reflection: Put the two outputs side by side. In one paragraph, name the three biggest differences and tie each one back to the specific part of the skeleton that caused it.
Rep 3 — One toggle at a time
Take your Rep 2 prompt and strip it back to just Objective + Format. Run it. Now add only Context and run again. Now add only a Role (instead of context) to the stripped version and run again. Three runs, one variable changed at a time.
Reflection: Which single addition — Context or Role — improved this task more? Predict-before-you-measure: did you guess right? Why do you think that ingredient mattered more here?
Reps 4–6: Context, examples, and role
Rep 4 — Fix a generic answer with one sentence of context
Ask the assistant a real everyday question from your work (“draft a reminder that our Friday meeting moved to Monday,” or similar) with no context. When the answer comes back generic, do not rewrite the whole prompt — add exactly one sentence of the right context (“the team is remote and half of them missed last week’s notice”) and rerun.
Reflection: Did one sentence of context outperform what a “cleverer” rewording would have done? What does that tell you about where good prompting actually lives?
Rep 5 — Match your voice with an example (few-shot)
Find a short message you actually wrote and liked (an email, a note — nothing confidential). Paste it in and say: “Match the voice of this. Now write me a [new short message] in the same voice.” Compare it to the same request made without the sample.
Reflection: Describe, in your own words, what “your voice” is — and whether the example carried it across better than adjectives like “warm” or “professional” could have. This is few-shot prompting; name why showing beat telling.
Rep 6 — Same news, four roles
Write one short piece of neutral news (“our office will be closed July 4”). Ask the assistant to write it four times, each under a different role: a warm office manager, a formal HR director, a punchy marketer, and a plain-language editor for seniors. Keep all four.
Reflection: Which role fit the news best, and why? Then the sharp question: did any role change a fact, or only the voice? What does that prove about what a role can and cannot do for you?
Reps 7–9: Reasoning, iteration, and reuse
Rep 7 — Make it show its work
Give the assistant a small decision with tradeoffs (“should a 12-person team use a shared calendar or a group chat for scheduling — pick one”). Run it once asking only for the answer. Run it again adding: “First list the factors, then weigh them, then recommend.”
Reflection: Did the step-by-step version reason better — or just look more thorough? Find one assumption the assistant revealed when it showed its work that you had never given it. Then say plainly: does visible reasoning make the answer checkable, true, or both?
Rep 8 — The follow-up is the tool
Take any first draft from an earlier rep. Improve it with exactly three follow-up messages, each changing one thing: (1) tone, (2) length, (3) a fact you correct or add (“you invented X — remove it” or “add the real deadline”). Do not rewrite the original prompt — use the conversation.
Reflection: Log what each of the three follow-ups bought you. Was the final version better than any single opening prompt you could have written? What does that say about treating the first answer as a draft?
Rep 9 — Save your first reusable template
Take your best prompt from this week and turn it into a template with blanks: replace the specifics with {{ROLE}}, {{AUDIENCE}}, {{FACTS}}, {{TONE}}, {{FORMAT}}. Start a real file called my-prompts.txt, paste it in with a one-line note about what it is for, and then use it by filling the blanks for a brand-new task.
Reflection: How long did the templated version take to reuse versus writing a fresh prompt from scratch? This file is the seed of your Week-12 prompt library — what is the first other template you already want to add to it?
Rep 10: Run the lab
Rep 10 — Drive the Prompt Upgrader
Use the Prompt Upgrader widget embedded below the chapter. Start from the vague prompt, and before flipping any toggle, predict which single toggle — Role, Objective, Context, Examples, or Format — will improve that prompt the most. Then flip them one at a time and watch the prompt and its likely output change.
Reflection: Was your prediction right? Which toggle produced the biggest jump, and which produced the smallest? What does the pattern tell you about where to spend your effort when a real answer disappoints you?
Done? One Last Thing.
A miniature of Project 4, end to end. Pick one short real piece of workplace news you could plausibly need to send (a schedule change, a new process, a closure). Then, in one sitting:
- Write it once for an internal teammate (direct, operational, what-changes-for-you).
- Rewrite the same news for an outside customer or client (warm, reassuring, benefit-first).
- In three sentences, say which version you are prouder of and why it fits its reader better — the fit word for that hearer.
Build both on the R-O-C-E-F skeleton, iterate each with at least one follow-up, and end with your AI usage note. If you can do this for one message tonight, you can do Project 4 — that is the whole project in miniature: the same truth, fitted three ways, and you deciding which one lands.
Up next: Project 4 — Project P4: One Email, Three Ways.