# The Business Brief **From Chapter 3.3 of *The 4-Hour Side Hustle*.** This is the 500-word document that makes every AI chain in Part II work. Without it, the model writes for an average reader on behalf of an average business, because that is genuinely all the information it has. With it, the model has a specific buyer, a specific price, and a specific voice to pattern-match against. Write it yourself, in plain language. Do not have AI write it — the entire point is that it contains things only you know. Aim for 500 words. If you're at 900 you're explaining; if you're at 200 you're being vague. Neither helps. --- ## How to use this file 1. Copy everything between the two rules below into a new document. 2. Answer each section. Delete the italic guidance as you go. 3. Upload it to a Claude Project or ChatGPT Project named after your business. 4. Add five writing samples and your raw customer language beside it. That's the whole setup. Re-read it once a quarter and after any price change. --- ## THE TEMPLATE — copy from here ### 1. Who I serve *Be specific to the point of discomfort. "Small business owners" is not an answer — it describes forty million people with nothing in common. Name the job title, the company size, the industry, the situation they're in. If your answer would fit on a billboard, it's too broad.* *Test: could you name three real people who fit this description? If not, narrow it.* **I serve:** --- ### 2. What I sell, and what it costs *The product, the format, the price. Include the price even in early drafts — it changes how the model writes about value, and a brief without a price produces copy that dances around the subject.* *If you have more than one thing, list them in the order you'd want someone to buy them.* **I sell:** **Price:** --- ### 3. The problem it solves — in my buyer's words *Not your words. Theirs.* *This is where the vocabulary from your 2.3 demand test goes: the exact phrases from comments, replies, and questions, unedited and un-tidied. If you haven't run the demand test yet, use support emails, sales call notes, or the last five questions someone asked you about this.* *The difference between a page that converts and one that doesn't is usually just this. Marcus's buyers said "covenant tracking" and "compliance monitoring" — he'd been writing "regulatory workflow optimization" for a year.* **They describe the problem as:** **Phrases they actually use (verbatim, at least five):** - - - - - --- ### 4. What I refuse to do *Products you won't build. Claims you won't make. Tones you won't use. Clients you won't take.* *This section does more work than it looks like it should. A model with no boundaries will happily write you a testimonial you don't have, a statistic you can't source, and an urgency deadline that isn't real. Writing the refusals down is what stops that at the draft stage instead of the publish stage.* **I don't build:** **I don't claim:** **I won't work with:** --- ### 5. How I sound *Three adjectives, then — far more usefully — three things you never say.* *Adjectives are a low-bandwidth way to describe a voice. "Conversational" means twenty different things and the model will pick the average one. "I never say 'unlock'" is unambiguous and immediately actionable.* **Three adjectives:** **Three things I never say:** 1. 2. 3. --- ## Copy to here — end of template --- ## A worked fragment So you can see the level of specificity that actually works. This is section 1 and section 5 from a brief for a fictional compliance-workbook business: > **I serve:** compliance officers at community banks under $2B in assets, usually a department of one or two, usually reporting to a CFO who doesn't have time to read what they produce. They've been in the role three to fifteen years. They're not looking to change jobs; they're looking to stop working Sundays. > **Three adjectives:** direct, practical, unhurried. > > **Three things I never say:** "game-changer," "in today's regulatory landscape," and anything that begins "as a compliance professional, you know that…" — they do know, and telling them they know is filler. Notice that the second one is more useful than the first, and took less thought. --- ## Testing it Ask the Project to write a 100-word customer email about something ordinary — an order confirmation, a reply to a question you get weekly. **If it doesn't sound like you, add material, not adjectives.** This is the single most important rule of the whole setup. The instinct is to add instructions — "be more conversational," "less corporate." That almost never moves the output, because you're competing against everything the model learned from the rest of the internet, and a sixth adjective doesn't outweigh that. A sixth writing sample does. Ten pieces of real context beat a hundred words of clever instruction, reliably. --- ## What this is not for This setup makes the model good at sounding like you. It does not make it you. It doesn't know what you decided last week, it doesn't know which customer is about to churn, and it has no taste — it has your patterns, which is a different and much shallower thing. Use it to draft, restructure, and produce first versions you'll edit. Don't use it to decide. --- ## Checklist - [ ] Written by you, not by AI, in plain language - [ ] Around 500 words — not 200, not 900 - [ ] Section 1 is specific enough that you could name three real people who fit - [ ] Section 2 includes the actual price - [ ] Section 3 contains at least five verbatim buyer phrases, un-tidied - [ ] Section 4 lists refusals, not just intentions - [ ] Section 5 has three things you never say - [ ] Uploaded to a Claude Project or ChatGPT Project named after the business - [ ] Five writing samples uploaded alongside it - [ ] Raw customer language uploaded alongside it, unedited - [ ] Tested with a 100-word customer email - [ ] When output was wrong, you added samples rather than adjectives - [ ] Diarised to re-read once a quarter, and after any price change