Here's the thing nobody tells you when you open a Shopify store: the product was never the hard part. The hard part is that you are the copywriter, the media buyer, the support rep, the supplier auditor, and the analyst, all at once, all unpaid, all a little badly, because no human is good at six jobs in the same evening.
That's the job most people quit. Not the product. The six jobs around it.
I didn't hire anyone. I handed the six jobs to Claude and kept the one job that actually matters: deciding what to do with what it gives back.
This isn't "AI writes your captions." It's using one $20 subscription as the operator that does everything you'd otherwise pay a small team to do. Below are the prompts that make up the entire store. Copy them -> they work.
But first, the part the gurus skip.
The math, honestly
Build a store the "real" way and here's what you're signing up for before a single person clicks Buy:
- Copywriter: ~$150 per product
- Support VA: ~$600–800/month
- Media buyer: ~$1,000–1,500/month
That's well over $2,000/month in overhead before you even know the product sells. It's why most stores die in month three, the costs eat the margin before the store finds its footing.
My version:
- Claude: $20/month
- Shopify: ~$1/month for the first three months, then ~$39
Same jobs. No salaries, no briefs, no managing anyone.
Now the honest disclaimer the income-screenshot crowd leaves out: this is not passive and it's not free money. I spend six to ten hours a week on this store. The prompts don't make decisions -> they hand me better decisions, faster. You still make the calls. Anyone telling you a store runs itself while you sleep is selling you a course, not a method.
Here's the shape of a recent month so you can see what "working" actually looks like:
Revenue:
$19,100
Product cost (~33%):
-$6,303
Shopify + fees:
-$310
Ad spend:
-$1,950
Claude:
-$20
Shopify plan:
-$39
Profit:
$10,478
Not retirement money. But real and it sharpens every month, because the store doesn't change, the prompts just get better at their jobs. Month one basically broke even. That's normal. Keep reading and I'll tell you why that's the expensive-but-necessary part.
A quick note on the prompts below: they're long on purpose. Every line either stops Claude from inventing numbers or pins down the exact shape of the answer. A short prompt gives you a pretty, useless reply. A long one gives you something you can actually make decisions on.
The store is six prompts
- research — finds the opening
- supplier — kills the disasters before they happen
- angle — finds the position nobody else owns
- page — writes copy that doesn't read like a product page
- ad — stops the scroll
- sunday — tells you the truth about your numbers
Each one does a job a human used to charge for. Here's all six.
1. research -> finds the opening
Most people pick whatever's trending on TikTok. That's why they're always late and always competing on price. By the time a product is "trending," twenty stores already own it.
I don't look for trends. I look for an annoying, recurring problem nobody has clearly solved yet. And this matters the prompt is built so Claude can't pass off a guess about margin or competition as a fact. A dropshipper who trusts invented numbers loses money.
1You are a dropshipping product strategist who has launched and killed 200+2products. You're paid to find me money, not to make me feel good.34Goal: 5 product opportunities I could launch on Shopify in the next 30 days,5ranked best to worst.67If you have web search, use it and cite a source for any number (price,8demand, competition). If you do NOT have web search, label every factual9claim "[ASSUMPTION — verify]" so I never confuse your guess with a fact.10Do not fabricate precise statistics.1112Hard filters — reject anything failing even one, and say which filter killed13the rejects:14- Recurring, annoying problem (not a buy-once novelty)15- Sells $29–$7916- No single brand owns the search results yet17- Ships from AliExpress / CJ in 14 days or less18- Before/after obvious in 5 seconds of silent video1920For each survivor:211. The product, one plain sentence222. The exact phrase a buyer types into Google/TikTok the moment they want this233. Who they are and the specific moment the pain hits244. Rough cost vs price and gross margin % (label the assumption if unverified)255. The angle every current seller is missing266. Why this could flop — the single most likely reason it's a trap2728Then rank all 5 and tell me which ONE you'd put $200 behind first, and why.2930End with a 5-point checklist of what I must personally verify before spending31a dollar.3233No categories, only specific products. If you can't name the exact buyer and34the exact moment, drop the idea.
Five options in under a minute. Pick the one whose problem you've personally had -> you'll write about it better than someone guessing. And run that final checklist yourself before you spend a dollar.
2. supplier -> kills the disasters
My first store didn't die from a bad product. It died from a bad supplier. The product was fine; the shipping wasn't, disputes piled up, and the payment processor froze the account. Gone.
Now no supplier gets an order until Claude has read their reviews like a detective with quotes, and an honest "I don't know" when the data's thin.
1I'm about to trust an AliExpress / CJ supplier with real orders and real2money. Be the skeptic. If the data isn't enough to judge, say UNKNOWN —3don't guess to sound helpful.45Data:6[paste: store name, rating, total orders, years active, advertised shipping7window, 10 most recent reviews in full, 5 worst recent reviews in full]89Check, and for every red flag QUOTE the exact review line that triggered it:10- Reviews clustered in a suspiciously tight window (bought ratings)11- Gap between advertised shipping time and what real buyers report — give12 the rough real average13- Whether complaints cluster on product / packaging / wrong item14- Frequency of "refund," "broken," "never arrived"15- Signs the listing's product was swapped (old reviews describe a different item)16- Weight the last 30–60 days more heavily than old reviews1718Output:19VERDICT: GREEN / YELLOW / RED / UNKNOWN20CONFIDENCE: high / medium / low (and why)21Three sentences of reasoning, each tied to the evidence above.22If YELLOW / RED / UNKNOWN: two search terms to find a better supplier for23the same product.
GREEN — order. YELLOW — buy a $20 sample first, always worth it.
RED — walk away no matter how good the price looks. This ten-minute step has saved me from at least three suppliers I'd have trusted on a quick skim.
3. angle -> finds your position
This one is two free research tools in one. Your competitors' best listings show you the angle everyone's already crowded into. Their worst reviews show you the exact thing customers are angry about which is the exact thing you put on your page before anyone has to be angry about it.
1You are a sharp e-commerce strategist. I'm entering a market that already2has sellers. I want the gap, not a louder copy of what's there. Be blunt;3if there's no real gap, tell me that instead of inventing one.45Competitors:6[paste the top 3 listings, full text]78Their unhappy customers:9[paste 20–30 of their 1–3 star reviews]10111. The one angle all three lean on (one line each) — and whether they're all12 making the same mistake132. The top 3 complaints by frequency. For each: roughly how often it appears,14 a real quoted snippet, and whether it's a product / shipping / expectations15 problem163. The position none of them occupy that I could own — and the honest reason17 it's empty (real opportunity, or empty because buyers don't actually care?)184. Two specific lines I should put on my page to defuse the #1 complaint19 before it happens205. One thing in these reviews customers LOVE that I should make louder than21 they do2223"Add reviews" and "better photos" are banned answers.
If the complaint is "took 25 days," your page says exactly when it ships, in bold, near the top. Answering the objection before it's raised converts better than pretending it doesn't exist.
4. page -> writes copy that converts
The description AliExpress hands you reads like it was translated off the side of the box. It kills the sale before the visitor finishes the first line.
1Rewrite this into a product description that sounds like a person, not a2catalog. Give me TWO versions so I can test.34Product: [name]5AliExpress text: [paste]6Real reviews: [paste 3–4, include one mixed one]7Buyer: [exact customer + their problem]8Price: $[X]910Voice: a friend recommending something they actually use. Direct, a little11understated, never salesy.1213Each version:14- First line = the one problem it solves. Nothing else. It must work on its15 own (most people only read this on mobile).16- Three benefits written as outcomes, not features.17 Not "high-density memory foam" → "you stop shifting around at 2am."18- One short paragraph on why people keep it — no invented reviews.19- A closing line that earns the click. No timers, no CAPS.2021Under 180 words each. No emojis.22Make Version A plain and calm, Version B with a bit more edge, so I can23A/B them.24After both: flag any sentence that could appear on ANY product page for ANY25product, and rewrite it.
The instruction doing the heavy lifting is "a friend recommending something they actually use." Without it, you get catalog voice. With it, you get something a human reads to the end. And two versions hand you something to A/B test instead of one guess.
5. ad -> stops the scroll
The best ads in 2026 look like an accident. Lo-fi, vertical, filmed in a real room. Polish reads as "ad" and people scroll. A clip that looks like a friend's recommendation gets watched.
The trap: "casual" is hard to write. Claude defaults to ad-voice unless you fight it. This prompt fights it and hands you a shotlist you can actually film from.
1Write a 30-second vertical video script for TikTok / Reels that feels like2a real person filmed it, not a brand.34Product: [name] — [the one benefit that matters]5Viewer: [exact person, exact situation]6Setting: phone, real room, real clutter on screen.78Give me:9- The script as a 4-row shotlist: timestamp | what's on screen | exact words10 spoken (verbatim, the way someone actually talks)11 0–3s: problem shown not said — "that's literally me" before anyone speaks12 4–14s: the problem, like venting to a friend, specific13 15–24s: product, one function, one unbroken shot14 25–30s: one real result, then a soft "link's in bio"15- 4 alternative first-3-seconds, each a physical action I can actually film16 (not a line of dialogue)17- For each opening, one line on why it might stop the scroll1819Banned: "game changer," "you need this," "are you tired of," "POV," "this20changed my life," "I was today years old." If any appear, scrap and restart.21Everything must be filmable in a normal apartment, on a phone, with no help.
Film three versions in one evening, changing only the first three seconds. Run each at ~$15/day for 48 hours. Kill the two with weak hooks. Put the budget behind the survivor. Most people lose three weeks overthinking this instead of doing it in one night.
6. sunday -> tells you the truth
The hardest part of running a store alone is that nobody tells you when you're wrong. You read your own numbers the way you'd read your own diary -> gently. This prompt doesn't. And it calculates your break-even ROAS the number that tells you whether you're making money or burning it.
1You are an e-commerce analyst with no incentive to be nice. One week of my2store's numbers below. Tell me what's broken.34Margin info (so you can do real math):5Selling price: $[X] | Product + shipping cost: $[X]67This week:8Visitors: [X] | Conversion: [X%] | AOV: $[X] | Cart abandonment: [X%]9Returning visitors: [X%]10Top 3 products (name, revenue, units): [...]11Highest-return product (name, %): [...]12Traffic sources (% split): [...]13Ad spend: $[X] | Revenue from ads: $[X] | ROAS: [X]1415First, calculate and state:16- My break-even ROAS (the ROAS below which I lose money) and whether I'm17 above or below it18- Roughly what my conversion rate SHOULD be for this price/category, and how19 far off I am2021Then six questions, one paragraph each, no lists:221. Biggest leak in the funnel right now?232. Which product do I scale this week, and why that one?243. Which product do I cut, and why?254. Is my ad money buying buyers or zombies?265. What does my returning-visitor rate say about real product-market fit?276. If I touch nothing for two weeks, what breaks first?2829End with: the single most important number on this page, and the ONE thing30to fix this week.31No encouragement. No "normal for this stage." If something's broken, say it32in the first sentence.
Twenty minutes every Sunday. Paste the numbers, read what's broken, pick one thing to fix that week. That's the whole optimization loop. The stores that plateau are the ones where the owner stopped looking.
Once it's working -> three more
The six prompts above build the store. The next three decide whether it stalls or grows. You add them once you have sales and almost nobody does.
7. email -> the flow that runs itself
Everyone obsesses over the first sale and goes silent right after it. But someone who already bought from you is far easier to sell to than a cold stranger you paid to reach. Set this up once in Klaviyo and it runs forever.
1Write a 5-email post-purchase flow for a Shopify store. Living-person voice,2short sentences, zero corporate filler ("we value your business" is banned).34Product: [name + one line] | Buyer: [who] | Shipping: [X] business days56Email 1 (right after purchase): confirm in human language, honest shipping7 expectation, one line on what to expect. No upsell.8Email 2 (day 3): one genuinely useful tip they wouldn't know. <80 words.9Email 3 (day after delivery): check in, one question — how is it? No links.10 <40 words.11Email 4 (day 14): one product that truly pairs with theirs + why. No discount.12Email 5 (day 30): ask for a review, name the exact thing to mention. One CTA.1314For each: TWO subject line options + body. <100 words each. No emojis, no15caps, no fake urgency.
For me, this flow quietly brings in around a fifth of revenue every month, with no extra ad spend. It's the closest thing to free money in the whole system.
8. offer -> raises your average order value
Most people, the moment sales start, go chasing more sales more traffic, more ad spend. But the cheapest dollar is the one from a customer already standing at checkout. Raising your average order value is usually faster and cheaper than raising traffic.
1You are a conversion strategist focused on one thing: raising my average2order value without discounting myself to death.34My store:5Main product: [name, $X, cost $Y]6Other products I have: [list with prices]7Current AOV: $[X] | Buyer: [who they are]89Give me:101. One bundle (2–3 items) priced so it feels like a deal but my margin goes11 UP, not down. Show the math.122. One post-purchase upsell (shown right after checkout) — what to offer and13 at what price, and why someone who just bought would say yes.143. One "buy 2, [benefit]" reason to add a second unit that is NOT just a discount.154. The exact one-line offer text for each, in plain human language.1617For each idea, tell me what it does to my margin per order, and which ONE18to test first.19No "free shipping over $X" unless you show it actually raises profit, not20just AOV.
One bundle or upsell that works can lift profit per order by 20–40% on the same traffic, with the same ad spend.
9. iterate -> learns from your ad results
The ad prompt makes creatives. But making ads isn't the skill. Reading the results is. Most people run a batch, see one did better, and... run more random ads. They never extract why. This prompt turns your numbers into a pattern and tells you what to shoot next, so each batch is smarter than the last instead of random.
1You're a paid-social strategist. I ran a batch of video ads. Help me read2the results and decide the next batch — not "test more," real decisions.34Per ad I tested:5[paste: hook/opening line, spend, impressions, CTR, cost per click, add-to-carts6or purchases]78Tell me:91. Which ad won, and the most likely REASON it won — hook, problem framing,10 offer, or just variance? Be honest if the sample is too small to know.112. Which ads to kill now and which to give more time.123. The pattern across winners — what do the good hooks have in common that13 the bad ones don't?144. 3 new hooks for the next batch that build on what won (not random new angles).155. Whether my problem is the creative, the offer, or the product — and how16 to tell which.1718If the data is too thin to conclude anything, say so and tell me the minimum19spend/clicks I need before trusting it.
After a few rounds you stop guessing hooks and start deriving them. That's the difference between someone burning budget on random clips and someone whose ads get cheaper with every batch.
The point
The stores that die aren't the ones with bad products. They're the ones where the owner is the bottleneck on every decision -> copy, ads, support, analysis, until the work buries them and they quit.
Everyone has the same $20 access to Claude. The person making nothing and the person making a few thousand a month are using the identical tool. The only difference is that one of them handed the work to prompts and kept the part that matters.
Six prompts build the store. Three more keep it from stalling. But it was never about the number of prompts
-> it's that you finally stop being the single bottleneck in your own business.
