90 天 5.3 万利润：Claude 成了我整个 Shopify 团队

90天盈利53,000美元：Claude成了我的整个Shopify团队 2026年5月25日 · 12分钟阅读 · 查看原文 ↗ Claude Advertising Shopify Side Hustle  迈阿密，凌晨3:08 Ethan 正在熟睡 他的手机亮起一条 Shopify 通知： 新订单 — 柏林 — $118.40 五分钟后：又一条 然后又一条  到他醒来时： - 供应商已经收到每一笔订单 - 物流单号已自动发送 - 客服已回复六位客户 - 三条新的广告钩子已准备好测试 - 昨天的数据分析已经为他准备好一份拆解报告 Ethan 没有亲自处理其中任何一件 他的店铺上个月营收达 $91,340 扣除产品成本、费用和广告后，他留下了 $53,180 工具成本： Claude Pro — $20/月 Shopify — $39/月 仅此而已 - 没有代运营公司 - 没有虚拟助理 - 没有文案写手 - 没有媒介采购 只有 Shopify 和 Claude 但这里有个没人看到的部分： 这是 Ethan 的第三家店 前两家彻底失败了 ## 坟场 2025年10月 第一家店 便携桌面小工具这个细分市场 五周烧掉 $1,700 产品在视频里看起来很酷，但解决不了任何重要问题 转化率：0.6% 几乎没有回头客  2025年12月 第二家店 宠物这个细分市场 这家店其实开始有起色了 广告有转化 客户下单了 评价都是正面的 然后供应商毁掉了一切 订单开始迟到 18–26 天 退款请求暴增 Stripe 冻结了 45 天的打款 产品不是问题 是物流毁掉了这家店 总损失：不到三个月内将近 $5,000 打水漂 大多数人在这个阶段就放弃了 Ethan 没有 他改变了一件事 ## 两家店都坏掉的那个环节 Claude 其实一直都在 1. 打开标签页 2. 随便问个问题 3. 得到一个随便的答案 4. 关掉标签页 这就是全部的问题所在 大多数人把 Claude 当成一个更聪明的 Google 搜索来用 每一次对话都从零开始 没有任何积累 在第三家店上，Ethan 不再把 Claude 当聊天机器人，而是把它当成他一次性雇来的员工  每个工作流都配备了： 1. 一个角色 2. 一套永久系统 3. 一个可复用的 prompt 从那以后，Claude 不再是「AI」 它成了这门生意的后端 这个转变改变了一切 ## 六个 prompt。两组 一次性决策（只跑一次） > product_find.prompt > > supplier_audit.prompt > > listing_builder.prompt 每天运行的系统 > ads.prompt > > support.prompt > > weekly_diagnostic.prompt 每一个负责店里不同的部分 它们组合在一起，替代了大部分运营工作 ## 1. product_find.prompt —— 在产品饱和之前找到产品 大多数新手在 TikTok 上搜索「爆款产品」 这必然导致： - 过度饱和的市场 - 被抄袭的广告 - 微薄的利润 - 无休止的竞争 这个 prompt 走的是相反的方向 ``` You are a product analyst with 7 years of experience in US and EU dropshipping. Find me 4 Shopify products with a real shot at working right now. Hard filters: - Solves a recurring everyday pain (not seasonal, not gift, not novelty) - Retail $24–$68 (impulse-buy threshold without long deliberation) - The pain shows up in Reddit complaints within the last 90 days - Top 3 Amazon listings are under different brands (no dominant player) - Ships from Spocket or a verified AliExpress US warehouse in under 14 days - The product's effect can be shown in one continuous camera shot For each product: 1. Specific product name (not "kitchen gadget", not "pet accessory") 2. Wholesale price vs retail price, exact margin % 3. Customer avatar: age, situation, the moment of the day they Google for a fix 4. Why this product works in 2026 specifically, not 2024 5. The angle no competitor is currently using 6. One specific Reddit post that proves the pain is real Generic categories are banned. "Pet accessories" is not an answer. If your answer could be applied to any other niche, start over. ``` 它不是追逐趋势，而是搜寻人们在网上反复抱怨的那些挫败 这个 prompt 迫使 Claude 找出： - 精确的客户 - 情感触发点 - 竞争对手为什么弱 - 为什么这个产品恰恰在当下行得通 - 为什么这个问题反复出现 关键之处在于： 1. 产品必须在 10 秒以内的视频里展现其价值 2. 仅这一点就能立即筛掉大部分坏产品 3. 一分钟内给出四个想法 一个新手研究员手工做同样的事要花上好几天 ## 2. supplier_audit.prompt —— 防止供应商毁掉你的店 大多数 Shopify 店铺不是因为广告失败而死掉的 它们死于供应商毁掉了客户信任 发货迟到 质量差 虚假的物流更新 退款混乱 这就是毁掉 Ethan 第二家店的原因 现在，每家供应商在上架前都会被审查一遍 ``` I'm vetting an AliExpress supplier before launching a store. Your job is to find everything that's off in their profile. What I'll give you: - Store profile: name, rating, years on platform, country - Shipping times listed in the product cards - 15 most recent reviews, full text - 6 reviews at 1-3 stars, full text - If I requested a sample: screenshot of the chat and response times - If I found the same supplier on Alibaba: link and their product range there Analyze for these signals: - Review velocity anomalies (10+ five-stars in 48 hours indicates bots) - Gap between listed shipping time and what real customers report - A repeating defect in negative reviews (bad stitching, broken seal, wrong color) - Signs of a rebrand (an older store name appearing in older reviews) - Mismatch between the AliExpress and Alibaba range (likely a reseller, not a factory) - DM response speed: under 12 hours is good, over 36 hours is bad Verdict: GO / SAMPLE / SKIP GO means take it now. SAMPLE means $15-30 on a test order before committing. SKIP means don't even start. Reasoning: 4 sentences max. If SAMPLE or SKIP, give me 2 alternatives with the search terms to find them. ``` Claude 会分析： - 预计发货时间 - 评价历史 - 负面投诉 - 响应速度 - 可疑的评分飙升 它能捕捉到人类通常会忽略的模式 一个糟糕的供应商就能毁掉好几个月的势头 这个 prompt 的存在，就是为了在钱被烧掉之前阻止这一切 ## 3. listing_builder.prompt —— 写出有人味儿的页面 大多数 Shopify 产品描述一眼就让人感觉很假 「高端品质」 「创新设计」 「革命性技术」 没人会那样说话 改变一切的那句话是： 「像一个用产品两周后给朋友发消息的人那样写」 仅仅这一条指令就彻底改变了语气 ``` Rewrite a product description for a Shopify product page. What I'll give you: - Name and the gist in one line - The raw AliExpress text (as-is, don't edit it) - 5 reviews: 3 enthusiastic, 2 with reservations - Customer avatar: age, situation, the specific moment they decide to buy - Retail price Voice: like a friend in a Telegram chat who bought this thing two weeks ago and is messaging you in DM about why it turned out better than expected. Not "introducing our exclusive new product." Not "premium quality." Structure: - Opening line: the one pain this product addresses. No fluff. - Context: one specific moment in the customer's day when that pain is sharpest. - 3 outcome bullets, not feature bullets. Wrong: "adjustable height, 8 settings" Right: "no more leaning toward the screen, back doesn't kill by 6 PM" - One paragraph: what this product does NOT do. What not to expect. This builds trust harder than any "100% guarantee" line. - Social proof: 1-2 real quotes from the reviews, no invention. - CTA: one sentence. No timer, no all-caps, no emojis. Word limit 220. If it reads like a product page, rewrite from scratch. ``` 现在，这些描述听起来： - 随意 - 可信 - 有人味儿 不再是一副官方腔调 最终转化率最高的部分竟是： 「这个产品做不到什么」 大多数店铺会隐藏产品的局限 这一部分直接把它们点出来 加上它之后，信任度几乎是立刻提升 ## 4. ads.prompt —— 写出不像广告的 TikTok 2026 年表现最好的广告看起来几乎一点都不专业 手机拍摄 凌乱的房间 自然光 真实的挫败感 广告看起来越精致，人们滑走的速度越快 ``` Write a 40-second vertical video script for TikTok, Reels, and Shorts. Given: - The product and its single core promise (1 line) - The target viewer in one phrase Style: UGC, home-shot. Phone camera. No tripod, no lighting setup, no scripted pause. Background should have stray household items (mug, phone, cables). This grounds the video in reality. Structure: 0–4s: Hook. Direct to camera. A specific pain-statement. By second 3, the viewer should think "that's me." 5–12s: Demonstrate the pain. No product in frame yet. Show the moment when the annoyance hits. Make it rough, real, uncomfortable. 13–22s: "I tried X and Y, neither worked." Name two real alternatives. This builds credibility. Don't lie about those alternatives. 23–32s: The product. One function only. Demo in one shot, no cuts. 33–40s: The outcome. One specific result. CTA in one sentence, no pressure. After the main script, give 5 alternative openers. Each 2-4 seconds. Each starts with a specific action or situation, not a question. Banned openers: - "Are you tired of..." - "Did you know that..." - "Introducing..." - "I found the secret to..." - "This changed my life..." - Any dramatic zoom-toward-camera with a whisper ``` Claude 会生成： - 钩子 - 情绪节奏 - 视觉动作 - 开场场景 - 供测试用的备选开场 然后 Ethan 用一个晚上拍出 4–5 个版本 不过度纠结 没有三周的剪辑流程 就是快速迭代 胜出的广告通常 48 小时内就显而易见了 ## 5. support.prompt —— 自动处理客服 店铺一规模化，客服量就爆炸式增长 40 条消息 70 条消息 有时一天超过 100 条 > 反复出现同样的问题： > > 我的订单到哪了？ > > 我能改地址吗？ > > 为什么发货延迟了？ > > 我能退款吗？ 现在，Claude 几乎自动起草每一条回复 ``` import Anthropic from "@anthropic-ai/sdk"; import express from "express"; const app = express(); app.use(express.json()); const client = new Anthropic(); const POLICY = ` You are customer support for [Store Name]. Tone: calm, concrete, human. Not saccharine. Hard store policies (do not improvise outside these): - Shipping: 9-15 business days from dispatch - Tracking number: sent within 36 hours of dispatch - Returns: 21 days from delivery, no questions asked - Refunds: processed within 4 business days of receiving the return - Address changes: only within the first 90 minutes after order Reply rules: - Maximum 70 words - Customer's first name once, at the very start - If delay is 16+ days: short apology, explain international customs in one sentence, offer code SORRY15 (15% off next order) - If refund request: confirm eligibility, give return warehouse address, state the refund timeline - Never use the phrases: "please be patient", "we apologize for the inconvenience", "sorry for the wait" - Never name competitors - If the customer is aggressive, threatening chargebacks, or hostile: reply with exactly one word, no period: HANDOFF If the request is not covered by the policies above: HANDOFF. Never promise anything outside these policies. `; app.post("/inbox-hook", async (req, res) => { const { from, name, message, orderId } = req.body; const response = await client.messages.create({ model: "claude-sonnet-4-20250514", max_tokens: 250, system: POLICY, messages: [{ role: "user", content: `Customer: ${name} Order: ${orderId} Message: ${message} Draft a reply. If outside policy, return exactly: HANDOFF` }] }); const draft = response.content[0].text; const needsAgent = draft.trim() === "HANDOFF"; res.json({ draft: needsAgent ? null : draft, needsAgent, from, orderId }); }); app.listen(3000); ``` 目前的通过率：约 82% 大多数回复发送前只需要微调 再也没有客户要等上半天了 仅这一点就大幅提升了信任 ## 6. weekly_diagnostic.prompt —— 每周日修复业务 每个周日，Ethan 会导出 Shopify 分析数据，把数字粘贴给 Claude 然后 Claude 会拆解： - 漏斗泄漏 - 表现疲软的产品 - 流量质量 - 转化瓶颈 - 放量机会 ``` You are a former CMO with 8 years in e-commerce. I'm handing you my budget for the week. Your job: find what I'm doing wrong and say it bluntly. Past 7 days of data: Unique visitors: [X] Conversion to checkout: [X%] Average order value: $[X] Cart abandonment: [X%] Returning visitors: [X%] Top 3 products by revenue: [name, revenue, orders, for each] Top 1 product by returns: [name, orders, return %] Traffic sources: [TikTok / Meta / organic / email, % breakdown] Ad spend this week: $[X] Revenue from paid traffic: $[X] ROAS: [X] Answer 7 questions. One full paragraph each. 1. The single biggest funnel bottleneck right now. Specific number, specific step. 2. Which product I scale this week and why exactly. 3. Which product I pause this week and why exactly. 4. I have 4 hours of changes I can make this week. What do I spend them on? 5. Paid traffic quality: am I paying for the right audience or burning money on the curious? 6. What does my returning visitor % say about product-market fit right now? 7. If I keep this same stack for the next 14 days, what 2 risks become real? Banned: complimenting, reassuring, saying "this is normal for early stage." If I'm doing something stupid, call it stupid. No bullet points. Full paragraphs only. ``` 最重要的一条指令： 「不要客气」 否则 AI 默认会输出一堆打鸡血的废话 一旦反馈变得毫不留情地直接，业务进步的速度就快多了 每周日花五分钟，替代了数小时的瞎猜 ## 真实的账 月度营收：$91,340 支出： 商品成本：−$29,400 广告支出：−$13,100 Shopify + 各项费用：−$4,660 净利润：约 $53,180 - 第1个月：勉强盈利。 - 第2个月：营收 $16K。 - 第3个月：营收 $41K。 - 第4个月：营收 $91K。 这家店并不是突然变得神奇了 是系统在复利积累 这才是真正的优势 ## 核心启示 四个月前，Ethan 拥有的是： - 两家失败的店 - 被冻结的打款 - 将近 $5K 的亏损 - 没有一套行之有效的系统 上个月：净利润 $53,180 同一间公寓 同一部手机拍摄 同一个人 只有一件事变了： 不再每天从零开始打开 Claude， 而是一次性搭建了六套永久系统 那六套系统，就成了这家店。 **如果你想了解更多关于 AI 系统、Shopify 以及现代在线商业策略的内容，欢迎关注我**  @insomnia_vip 标签：# X # Claude # Advertising # Shopify # Side Hustle # Side Hustle # Marketing # Growth # AI # Tiktok # Ecommerce # Sonnet 相关文章 Claude + Shopify = $10,478 利润。这是我的「桌面」长什么样： 关于下面这些 prompt 的一点说明：它们故意写得很长。每一行要么是为了阻止 Claude 编造数字，要么是为了锁定答案的确切形态。一个短 prompt 给你的只会是一个漂亮但没用的…… Advertising Claude Shopify Side Hustle 一家 Shopify 店月入 $91,340，靠的是把 Claude 当运营者，而不是写手。 Shopify + Claude 的重点从来不是产品描述。 Shopify Claude Advertising AI 如何靠浏览器小游戏赚 $15,000：Claude Fable 5 + Claude Code + Poki/CrazyGames 在 Poki 和 CrazyGames 上，你赚钱靠的不是卖游戏，而是玩家看到的广告。平台会在游戏周围和游戏内部投放广告并分享收入。在 Poki 上，当玩家……时是五五分成。 Games Advertising Claude Side Hustle