# The Prompt Library

**Every prompt in *The 4-Hour Side Hustle*, copy-paste ready.**

Upload this whole file to the Claude Project or ChatGPT Project you built in Chapter 3.3 — the one holding your Business Brief, five writing samples, and raw customer language. Then call a prompt by name when you need it.

Run without that context, these produce competent generic output. Run with it, they produce something recognizably yours. The context is doing most of the work; the prompts just point it.

---

## The pattern worth understanding first

Most of these give the model an explicit way to **refuse rather than invent**:

| Token | Prompt | Means |
|---|---|---|
| `WEAK CLIP` | Caption | The transcript is too fragmentary to make a point |
| `[NEEDS AUTHOR INPUT: …]` | Restructure | A passage is unclear — flag it, don't fix it |
| `NO ANSWER IN DOCS` | Support | The answer isn't in the knowledge base |
| `NEEDS DECISION` | Support | This needs a human to decide, not draft |
| `NOT A THREAD` | Repurposing | The piece is descriptive, not argumentative |

A model with no way to say "I can't" will always produce something. Something is worse than nothing when it's going out under your name. Every prompt below that touches facts has an escape hatch, and you should add one to any prompt you write yourself.

---

## P1 — Skill Extraction *(Ch 2.2)*

Run against a transcript of you describing your day job.

> You are a business analyst who specializes in productizing professional expertise. Below is a transcript of me describing my day job. I have been doing this work for years and I can no longer see which parts of it are valuable to other people.
>
> Your task:
>
> 1. Identify every piece of specialized knowledge in this transcript — including things I mention in passing, complain about, or clearly assume everyone knows. Pay particular attention to anything I describe as obvious.
> 2. For each one, name who else would pay for it, and roughly what they'd pay.
> 3. Propose 10 specific digital products. Each must be deliverable as a file, a template, or a short course — nothing requiring my ongoing time per customer.
> 4. For each of the 10, note the single biggest reason it would fail.
>
> Rules: Do not invent expertise I did not describe. Do not propose generic products (courses on productivity, general consulting, "coaching"). Prefer boring and specific over exciting and broad. If something in the transcript suggests a product I'd be uniquely credible selling, say so explicitly and explain why.
>
> Transcript follows.

**Then, as a second message — do not skip this:**

> Now argue against your own top three. For each, tell me why a busy professional with fifteen hours a week should NOT build it.

The second pass is where the useful answer lives. The first pass is a model being agreeable.

---

## P2 — Caption *(Ch 5.2)*

One platform per run. See `caption-prompts.md` for all four platform blocks as a standalone Project.

> You are writing a social caption for a short video clip from my long-form content. My Business Brief, writing samples, and buyer vocabulary are in this project — use them. Write as me, not about me.
>
> **Input:** the clip transcript is below, along with the topic of the source video.
>
> **Hard rules:**
> - Use only claims, numbers, and examples that appear in the transcript. Invent nothing.
> - No emoji. No "In today's world." No rhetorical question as the opening line.
> - Do not describe the video ("In this clip I discuss…"). Say the thing itself.
> - If the transcript is too fragmentary to make a coherent point, reply only with `WEAK CLIP` and stop.
>
> **Platform — LinkedIn:**
> - First line is a complete, specific claim that stands alone. It is the only line most people will read.
> - Then a line break, then 3–5 short paragraphs of one to two sentences.
> - Close with an observation, not a call to action. No "what do you think?"
> - 120–200 words.
> - 3 hashtags, lowercase, industry-specific, no generic tags like #business.
>
> **Transcript:**

---

## P3 — Restructure *(Ch 6.2)*

Turns a spoken transcript into an article draft. Also available standalone as `restructure-prompt.md`.

> Below is a transcript of me speaking about a topic in my field. Turn it into a first draft of a written article.
>
> **Your job is to restructure, not to rewrite.**
>
> - Preserve my specific examples, numbers, names, and dates **verbatim**. Do not smooth, generalize, or paraphrase them.
> - Preserve my opinions, including ones stated bluntly. Do not hedge them.
> - Remove only: filler, false starts, repetition, and things I clearly abandoned mid-sentence.
> - Reorder for a reader rather than a listener. Spoken tangents that work aloud usually need to move or go.
> - Add subheadings that describe what the section actually says, not generic labels.
>
> **Hard rules:**
> - Invent nothing. Add no statistics, studies, examples, or claims not present in the transcript.
> - If a passage is unclear or a claim seems incomplete, do not fix it — mark it `[NEEDS AUTHOR INPUT: what's unclear]` and move on.
> - Do not add an introduction that summarizes what the article will cover. Start with the substance.
> - Do not add a conclusion that restates the article.
>
> Target 1,000–1,400 words. Transcript follows.

---

## P4 — Support Draft *(Ch 9.2)*

> You are drafting a support reply on my behalf. My product documentation, FAQ, and past replies are in your knowledge base. I will review and send every draft — you are never speaking directly to the customer.
>
> **Rules:**
> - Answer only from the documentation. If the answer isn't there, reply with `NO ANSWER IN DOCS` and a one-line summary of what they're asking. Do not construct a plausible answer.
> - Match my tone: direct, warm, no exclamation marks, no "Great question!"
> - Lead with the answer. Context after, if needed at all.
> - If they sound frustrated, say so plainly and don't apologize more than once. Draft it short — I'll add anything personal.
> - If the question reveals something confusing about the product, add a line at the end starting `FIX:` describing what to change. This goes to me, not the customer.
> - Never offer a discount, an extension, or an exception. Flag those with `NEEDS DECISION`.
>
> Draft the reply below.

---

## P5 — Sales Page Block *(Ch 8.1)*

**Run separately for each of the nine blocks. Never generate the whole page in one pass.** The example below is block 3.

> I'm writing a sales page for a digital product. Below are twenty verbatim sentences from real prospects and customers describing their problem in their own words.
>
> Draft **block 3 only** — "the problem, stated back." 80–120 words.
>
> **Rules:**
> - Use their vocabulary, not marketing vocabulary, and not mine. Where they used a specific phrase repeatedly, use that phrase.
> - Do not mention my product, its name, or its features anywhere in this block.
> - Describe their Tuesday, not their aspiration.
> - Include one specific detail that shows I've actually seen this situation.
> - No rhetorical questions. No "sound familiar?"
>
> Customer sentences follow.

---

## P6 — Repurposing *(Ch 6.3)*

Four separate prompts run against a published article. Never one "repurpose this."

> **LinkedIn post.** Take the single most interesting claim in the article below and develop it into a standalone post. Do not summarize the article. Do not mention that an article exists. One idea, one supporting example from the piece, 150–250 words. Open with the claim.

> **X thread.** Compress the argument of the article below into 5–7 posts. Only proceed if the piece contains an actual argument — if it's descriptive rather than argumentative, reply `NOT A THREAD` and stop. No hashtags. First post must stand alone.

> **Newsletter section.** Write 200 words framing why this matters to my reader *this week*, ending with a link. Written as me talking to one person who already subscribes, not as a promotion.

> **Carousel slides.** Five slides: hook, three points, close. Under 20 words per slide. Output as CSV with columns `slide1,slide2,slide3,slide4,slide5` for Canva Bulk Create. Text only.

---

## P7 — Agent Job Description *(Ch 15.1)*

Not a prompt — the document you write **before** building any agent.

> **AGENT:** [name]
>
> **Purpose:** [one sentence — what problem it removes]
>
> **Inputs:** [exactly what it can see. Be narrower than feels necessary]
>
> **Allowed actions:** [an explicit list. Not "manage X" — the specific actions]
>
> **Forbidden actions:** [never send, never delete, never spend, never contact a customer, plus anything domain-specific]
>
> **Escalation:** [what it does when unsure. Always: stop and tell me]
>
> **Review:** [what I check, how often, and which categories I spot-check because errors there are expensive]

---

## Troubleshooting

Five failures you'll hit, and what actually fixes them.

**"It sounds like a press release."** Not enough writing samples, or samples that are too formal. Add emails you wrote to a colleague you like. The register you use with a friendly peer is almost always the register your buyers respond to.

**"It keeps inventing statistics."** Your prompt lacks an explicit prohibition and an escape hatch. Add both: *invent nothing*, plus a refusal token.

**"The output is fine but generic."** The context is thin, not the prompt. Generic output means the model is writing for the average reader of your industry because that's all it knows. Add the raw customer language from Chapter 2.3 — the specific, slightly wrong phrasing is what makes output land as yours.

**"It ignores half my instructions."** Too many rules in one prompt. Split it. The caption prompt works because it does one thing for one platform.

**"It was good last month and it's worse now."** Almost always the project knowledge has drifted — you've added documents, or the samples now include AI-assisted writing that's teaching it to sound like itself. Prune back to your five best human-written samples and re-test.

**The meta-rule:** when output degrades, look at the context before the prompt. The prompt is the small half.

---

## Checklist

- [ ] This file uploaded to your Claude or ChatGPT Project
- [ ] Business Brief in the same Project
- [ ] Five writing samples in the same Project
- [ ] Raw customer language from Chapter 2.3, unedited, in the same Project
- [ ] Skill Extraction run in two passes — the argue-against pass is not optional
- [ ] Caption prompt run one platform at a time, never four in one call
- [ ] Sales page run one block at a time, never nine in one call
- [ ] Every prompt you write yourself has a refusal token
- [ ] Project knowledge pruned when output starts drifting