# Growth check prompts. Metagon Growth Starter Kit · file 1 of 7 You know your business too well to see it like a stranger does. An AI assistant sees it from outside. It also knows nothing about you, so its answer can only be as good as the question and the material you give it. These twelve prompts check the five places where growth usually leaks: the offer, the page, the ads, the counting and the follow-up. Copy one, paste it into any AI chat assistant, replace the `[PASTE: ...]` part with your real material, and compare the answer with "A good answer" underneath. ## Before you start. - **Remove private details.** No customer names, emails, phone numbers, health details, logins or passwords. A summary with those taken out works just as well. - **Paste the real thing.** The real page text, the real ad, the real numbers. Not your memory of them. - **One prompt per chat.** Answers get vague when one chat holds everything. - **Check what it tells you.** An assistant can be wrong and still sound sure. Check every number it gives you against your own. - **Write down what you decide.** Use the decision log in file 6, the weekly improvement loop. | Area | Prompts | | --- | --- | | The offer | 1 · 2 · 3 | | The page | 4 · 5 · 6 | | The ads | 7 · 8 · 9 | | The counting | 10 · 11 | | The follow-up | 12 | --- ## 1 · The stranger test. Checks: can someone who has never heard of you say what you sell, and to whom? Paste in: the headline and first paragraph of your main page, or the two sentences you use to describe the business. ```text Act as a busy potential customer who has never heard of my business. Read the text below once, quickly, the way you would on a phone. Then answer: 1. In one sentence: what do we sell, and who is it for? 2. What would you still need to know before you'd consider buying? 3. Which words are vague or sound like every other business? Quote them. 4. For each vague word, suggest a concrete alternative a customer would use. Be direct. Don't rewrite the whole text. [PASTE: your headline and first paragraph] ``` **A good answer:** its one sentence matches what you really sell. If it doesn't, the text is the problem, not the assistant. It quotes your exact words and offers concrete nouns: "bike repairs done in two days" instead of "premium mobility solutions". --- ## 2 · Your customers' words. Checks: the gap between how you describe the business and how your customers do. Paste in: 10 to 30 real snippets from reviews, emails, chat messages, call notes and the questions people asked before buying. Names removed. ```text Below are real words from my customers, with names removed. Sort them into four groups and quote the exact words in each: 1. The problem, as they describe it. 2. What they want instead. 3. What almost stopped them from buying. 4. Why they chose us. Then list the 10 phrases that come up most often or say the most, exactly as written. Do not turn them into marketing language. Keep their words. [PASTE: customer snippets, one per line] ``` **A good answer:** quotes, not summaries. Phrases you could put straight into a headline. If every group reads like a brochure, ask again and add "use only their exact words". --- ## 3 · The offer and the price. Checks: whether your offer is easy to say yes to, and whether the price leaves room to pay for new customers. Paste in: what's included, the price, what one sale leaves you after direct costs, and how often customers buy again. ```text Here is my offer and its numbers. 1. Describe the offer back to me in two sentences, the way a customer would understand it. 2. What makes saying yes hard? Think about risk, effort, price, time and anything unclear. 3. What one change would make the offer easier to say yes to, without lowering the price? 4. Using my numbers: roughly how much can I afford to pay to win one customer and still make money? Show the maths step by step. Use only the numbers I gave you, and ask if one is missing. [PASTE: what's included, price, money left per sale after direct costs, how often customers buy again] ``` **A good answer:** shows the maths and states its assumptions. It asks when a number is missing instead of guessing. The changes it suggests are things you could actually deliver. --- ## 4 · The first screen. Checks: what a visitor understands on a phone before they scroll. Paste in: everything on the first phone screen of your page: headline, sub-line, button text and a description of the picture. A screenshot works if your assistant reads images. ```text This is what a visitor sees on my page on a phone, before scrolling. Answer as someone who just clicked an ad and gives it five seconds: 1. What is this? 2. Who is it for? 3. What should I do next, and what happens when I do? 4. What one thing would make me leave? Score questions 1 to 3 from 0 (no idea) to 2 (instantly clear) and explain each score in one line. [PASTE: first-screen text, button text and a description or screenshot of the picture] ``` **A good answer:** every score below 2 comes with a reason that points at a specific word or a missing piece. If it hands you three 2s and praise, ask it to argue against the page. --- ## 5 · The unanswered questions. Checks: which questions customers ask that your page doesn't answer. Paste in: the full text of the page, and the questions people ask you before buying (from emails, calls and chats). ```text Below are (A) the full text of my page and (B) questions customers ask before they buy. 1. For each question in B, say whether the page answers it: yes, partly or no. Quote where. 2. List the questions the page leaves open, most important first. 3. For the top three, suggest where on the page the answer belongs and draft it in under 40 words, in plain language. Don't invent facts about my business. Mark anything I need to confirm with [CONFIRM]. [PASTE A: page text] [PASTE B: customer questions] ``` **A good answer:** a clear yes, partly or no for each question, with quotes. Drafts that mark gaps with [CONFIRM] instead of making up delivery times, prices or promises. --- ## 6 · The path to yes. Checks: every step between "I want this" and "done". Paste in: each step of your form, booking or checkout, in order: the fields, buttons, pages and messages people see. Include what they receive afterwards. ```text Here is every step a customer goes through to book, buy or ask, in order. 1. Count the steps and the fields. Which could be removed or moved to later? 2. Where would a person on a phone most likely give up, and why? 3. Which messages are unclear, especially error messages and the confirmation? 4. What does the customer know, and not know, right after they finish? List the changes in order of how much effort they save the customer. [PASTE: the steps, fields, button texts and messages, in order] ``` **A good answer:** names specific fields and words, not "simplify the form". It notices what's missing after the yes, such as a confirmation that says what happens next and when. --- ## 7 · Does the ad match the page? Checks: whether the promise in the ad is the first thing people see after the click. Paste in: the text of one ad, a description of its picture or video, and the first screen of the page it leads to. ```text Below are (A) an ad and (B) the first screen of the page it links to. 1. What does the ad promise, in one line? 2. Does the page repeat that promise straight away, in similar words? Quote both. 3. Would someone who clicked feel they landed in the right place? Why or why not? 4. Suggest a page headline that continues the ad, using words from the ad. [PASTE A: ad text and a description of the image or video] [PASTE B: first-screen text of the page] ``` **A good answer:** quotes both sides and says plainly when they don't match. The headline it suggests borrows words from your ad, not new marketing phrases. --- ## 8 · What your ads are telling you. Checks: the patterns in what worked and what didn't. Paste in: your ads from the last one to three months, each with its message or hook, its format, what it cost and how many real results (sales, bookings, requests) it brought. Use results you trust, not only clicks. ```text Below is a list of my recent ads with their spend and real results. 1. Group the ads by message or hook, not by format. 2. Which groups brought results at a cost I can afford? Which didn't? 3. What do the stronger ads have in common that the weaker ones lack? 4. Which conclusions are solid, and which rest on too few results to trust? Be explicit. 5. Suggest what to stop, what to keep and one thing to test next. Don't suggest raising the budget. The most I can pay to win one customer: [PASTE: your number from prompt 3] [PASTE: ad list with message, format, spend and results] ``` **A good answer:** separates "this is clear" from "too early to say". Its test idea is one idea, specific enough to brief (file 5). If it crowns a winner after a handful of results, push back. --- ## 9 · New ad ideas from real words. Checks: whether your next ads start from what customers actually say. Paste in: the customer phrases from prompt 2 and the two-sentence offer from prompt 3. ```text Using only the customer phrases and the offer below, write 5 ad ideas. For each one, give: - The hook: the first line, or what happens on screen at the start, in a customer's own words. - The one message of the ad, in one sentence. - What to show: something real from the business, no stock imagery. - The proof it would need. If we don't have that proof, say so. Don't make claims I haven't given you. No invented numbers, reviews or results. [PASTE: customer phrases] [PASTE: two-sentence offer] ``` **A good answer:** hooks you recognise from your customers' mouths. Every idea names its proof or admits it has none. Pick one and fill in the ad creative brief (file 5). --- ## 10 · What you count. Checks: whether you count the moments that matter, or only clicks and visits. Paste in: the list of actions your website counts today (ask whoever set it up for that list), and how a customer buys, books or asks. ```text Here is how customers buy from us, and the list of actions our website counts today. 1. Which action is the real result for this business? Is it counted? 2. Which important steps are missing from the list? 3. Which counted actions don't matter and only add noise? 4. Could any action be counted twice, for example when someone reloads a page? 5. Write the ideal short list: 3 to 6 actions, each with one line on why it matters. Don't recommend specific tools. Use plain words. [PASTE: how customers buy, book or ask] [PASTE: list of counted actions] ``` **A good answer:** a short list with the real result (a sale, a booking, a request) at the top. It flags anything that could be counted twice. Then test the list with the first-party tracking checklist (file 4). --- ## 11 · Counted versus real. Checks: how far your counted results are from what really happened. Paste in: for the same weeks, the results your ad platforms report, the results in your own website statistics, and your real results from your shop, booking calendar or invoices. Totals only, no customer details. ```text Below are results for the same period from three places: (A) the ad platforms, (B) my website statistics, (C) my real sales or bookings. 1. Show the differences between A, B and C, as numbers and as a share of C. 2. List the usual reasons for each gap. For example: visitors who declined tracking, the same sale claimed by two platforms, no counting on the thank-you page, different time zones or date ranges. 3. Which gaps look normal, and which suggest something is broken? 4. Which three checks would find the cause fastest? Use only my numbers. Tell me if the dates or numbers look inconsistent. [PASTE: A, B and C, week by week] ``` **A good answer:** shows the arithmetic, treats C (your real sales) as the truth, and separates normal gaps from broken ones. The checks it suggests are ones you can run yourself. --- ## 12 · What happens after the first yes. Checks: what people hear from you after they ask, buy or go quiet. Paste in: what happens today after someone (a) sends a request, (b) buys or books, (c) asks but doesn't buy. Who replies, how fast, with which message, and what happens after that. ```text Here is what happens today after a customer asks, buys, or asks but doesn't buy. 1. Where do people wait too long or hear nothing at all? 2. Which messages are the same every time and could be sent automatically? Any automated message must say it is automated, and a person must be able to answer a reply. 3. Where should a person stay involved, because it needs judgement? 4. Which one follow-up would most likely bring back people who asked but didn't buy? 5. Draft that follow-up in under 80 words, friendly and plain. Mark facts I need to confirm with [CONFIRM]. [PASTE: what happens after (a) a request, (b) a purchase or booking, (c) a request without a purchase] ``` **A good answer:** points at specific silences ("nobody replies at the weekend"), separates repeat work from work that needs a person, and drafts a message you'd be glad to receive yourself. --- ## Then what. You'll have more ideas than time. Pick three. First the problems that cost you money now, then the ones that confuse customers, then everything else. A broken count goes first, because you can't judge an ad you can't measure. Put the first one into the weekly improvement loop (file 6) and change one thing this week. The AI skill in this kit (`growth-check-skill/SKILL.md`) runs the same check as a guided conversation. --- Metagon Growth Starter Kit · metagon.co/growth-program