I let Claude run a dropshipping store for 30 days: here’s exactly how the setup works

Imagine launching a brand new Shopify store, handing it over to an AI assistant, and coming back 30 days later to see over $12,000 in sales. That’s exactly what this setup does by turning Claude into a fully autonomous dropshipping operator that researches products, builds your store, creates video ads, and runs Facebook campaigns.

What this Claude-powered store actually achieved

The automated store was a fresh Shopify build that had only been live for about a week before Claude took over. Once the AI was in charge, it switched products, rebuilt the funnel around the new winner, and ran the ads for 30 days.

Over that period (Sept 21 to Oct 21), the store generated around $17,000 in gross sales and about $12,500 in sales over time, with a conversion rate of roughly 3.7%. The key point: after the initial setup and rules, Claude handled the day-to-day work.

The core idea: hire Claude as your dropshipping operator

The entire system treats Claude like a full-time hire whose job is to run a Shopify dropshipping business. Instead of giving it one-off prompts, you:

• Define its role, responsibilities, and boundaries
• Connect it to the tools it needs (Shopify, product research, ad platforms, creative tools)
• Train it with reusable skills and prompts
• Let it operate inside your desktop and browser so it can actually click, type, and build

Once this is done properly, you can launch new stores in a repeatable way and let Claude handle most of the execution.

The tools you need for full automation

To replicate this setup, you’ll need a specific stack of tools that Claude can control directly or via connectors:

• Claude desktop app (Windows or Mac) – gives you co-work and code modes, plus deeper system access.
• Claude Co-work – for high-level instructions, planning, and workflow orchestration.
• Claude Code – for more technical tasks, file handling, and using your local machine (terminal, folders, etc.).
• Shopify – your ecommerce platform, connected to Claude via the built-in Shopify connector.
• Winning Hunter – a product research tool Claude uses to find and validate winning products.
• Meta Ads (Facebook Ads) – where Claude launches and manages your campaigns.
• An AI creative studio (e.g., Higgsfield, TopView, OpenArt) – for auto-generating product images and video assets.
• USA Drop – as the primary dropshipping supplier and fulfillment partner.
• CapCut (web) – for video editing that Claude can control via your browser.
• Claude Chrome extension – so Claude can remotely control your browser tabs.

This is the same kind of agent setup you’d use if you were building a 1-person AI business with Claude. If you want a deeper dive into that mindset, check out how to start a 1-person AI business with Claude in 30 days.

Step 1: Install Claude and set up the right model

Always use the desktop app. The web version is fine for chat, but you need the desktop app for full automation, file access, and remote control.

Inside the app:

• Use Claude Co-work for high-level prompts and workflows.
• Use Claude Code when you want it to manipulate files, folders, and your local environment.
• Select the Opus 5.5 model on the High setting for this kind of heavy workflow.

Step 2: Define Claude’s role and rules

You don’t just say “help me with dropshipping.” You explicitly “hire” Claude into a role, using a long, detailed prompt plus a rules file:

• Role prompt: tells Claude it is your Shopify dropshipping operator, responsible for product research, store setup, creatives, and ads.
• Rules (skill file): a Markdown file (e.g., Claude_Rules.md) that defines how it should behave, what to prioritize, how to communicate, and what constraints to follow.

Upload the rules as a Skill inside Claude:

• Go to Skills → Manage skills → Add / Upload skill
• Upload the Markdown file or paste the content

This gives your Claude agent a persistent “operating manual” it can reuse across projects.

Step 3: Set hard limits for ad spend and testing

Before you let an AI touch your ad account, you must define spend limits and kill rules:

• Daily budget caps (e.g., $50–$100/day per test)
• Conditions to pause or kill ad sets (e.g., no sale after X spend, bad CTR, high CPC)
• Rules for scaling winners (when to raise budgets, when to duplicate, etc.)

You can embed these rules in the same Markdown file or a separate “Facebook ads rules” skill. Claude will then know exactly when it’s allowed to spend, scale, or stop.

Step 4: Connect Shopify and your key tools

Next, you give Claude access to the tools it needs via connectors and the browser.

Shopify connector

Inside Claude:

• Click the + icon → Connectors → Manage connectors
• Go to Discover and search for Shopify
• Sign in and authorize access

If you’re creating new stores frequently, you can reuse the same process for each Shopify account.

Custom connectors for other tools

For tools that don’t have native connectors (like Higgsfield, Winning Hunter, or Meta Ads), you use custom connectors:

• Go to Connectors → Manage connectors → Add custom connector
• Paste the custom URL the tool provides
• Authenticate by logging into your account

Repeat this for:

• Winning Hunter – product research
• Higgsfield (or alternative) – creative generation
• Meta Ads – for Facebook/Instagram campaigns

Think of Claude as a worker on a construction site. These connectors are its tools. Without them, it can’t build anything useful.

Step 5: Enable browser control with the Claude Chrome extension

Some sites and actions won’t be available through connectors alone. That’s where the Chrome extension comes in.

Install the Claude Chrome extension, then in your Claude settings:

• Go to Settings → Claude in Chrome
• Turn on Enable Claude in Chrome
• Set it to Allow all sites

When your tabs turn purple, it means Claude can now click around, type, and perform tasks in your browser. This is how it can operate tools like CapCut web or Facebook Business Manager even if there’s no direct API path.

Step 6: Install product research skills and find a winning product

Now you teach Claude how to hunt for products using a dedicated skill.

Upload the product research skill (for example, a “Proven Problem Product Hunt” skill) via the Skills menu. Then, in Claude Co-work, you call that skill with a specific prompt that tells it to:

• Use Winning Hunter data (Facebook ads, store performance, etc.)
• Score products on criteria like demand, margins, competition, and ad saturation
• Return a shortlist of 5–10 products with a ranked scorecard

Claude will output:

• A ranked list of candidates
• Target selling price and estimated profit margin
• Ad count and performance data from competitors
• AliExpress or supplier links

For the harness example, it chose a “bully-proof” dog harness designed for strong breeds, with clear differentiation from generic harnesses.

In your first 1–2 runs, manually review the winning product against your own criteria before trusting it fully. Once you’re happy, you can allow Claude to auto-approve within your rules.

Step 7: Find audiences competitors are ignoring

Instead of just copying existing ads, the workflow includes a skill to find “people nobody is talking to” for your product.

Upload the Avatar Competitor Gap skill, then run a prompt that tells Claude to:

• Scrape real customer language from places like Reddit, Quora, and reviews
• Identify distinct audience segments and their pain points
• Map which needs competitors are addressing—and which they’re ignoring
• Extract the top hooks and angles for each neglected segment

The output typically includes:

• Audience personas with real quotes
• Top 10 hooks per persona
• Competitor breakdowns (what they emphasize and what they miss)
• A “gap delivered” section listing the messaging opportunities your ads can own

This document becomes the backbone for your ad scripts, landing page copy, and creative angles.

Step 8: Connect your supplier and import the product

Before Claude builds the store layout, you want the product itself imported and linked to a supplier.

The workflow uses USA Drop as the main supplier:

• Create a free USA Drop account
• Install the USA Drop app from the Shopify App Store
• Connect your Shopify store to USA Drop

Then:

• Search for your product in USA Drop’s catalog and import it directly, or
• Submit a sourcing request with the AliExpress link and product images if it’s not listed

USA Drop offers faster shipping lines, in-house stock, and support for scaling and private labeling once you hit around 5 orders per day. That stability matters when an AI is driving volume—bad suppliers will quickly turn into refunds and chargebacks.

Step 9: Let Claude build your Shopify store

With the product imported and research ready, it’s time for Claude to build the actual storefront.

Use Claude Code with local access

Open Claude Code and:

• Create a new folder on your computer (e.g., Shopify_Store)
• Give Claude access to that folder so it can save assets (images, copy, sections)
• Allow access to the terminal so it can create theme sections or run scripts if needed

Claude can also use its built-in browser to log into your Shopify admin and configure the theme directly.

Use a pre-built theme for speed

The workflow assumes you upload a pre-built theme file (ZIP) that already contains optimized sections for high-converting product pages. Once uploaded to Shopify, you only need to customize:

• Brand colors and fonts
• Product images and lifestyle shots
• Headlines, benefits, FAQ, and reviews

Using your earlier research, you instruct Claude to:

• Generate a product name and brand story
• Write the full landing page copy (hero, benefits, social proof, FAQ, guarantees)
• Adapt everything to the audiences and gaps identified in the competitor research

Do this carefully the first couple of times, then tell Claude to save the structure and tone to memory so it can reuse the same style for future products.

If you want a focused walkthrough of using Claude Code specifically for Shopify, you may also like how to use Claude Code agents to run your Shopify store.

Step 10: Generate ad scripts from real-world winners

Next, Claude needs to create video ad scripts based on what’s already working in the market.

Using the Winning Hunter connector:

• Claude finds the top 3–5 best-performing ads for your main competitor
• Analyzes their structure, hooks, pacing, and CTAs
• Cross-references this with your avatar gap research

Then you give Claude a script-writing prompt that tells it to:

• Write a new script inspired by the structure of those winning ads
• Target the underserved audience segments it discovered earlier
• Keep the pacing, length, and style optimized for Facebook and Instagram

Review the script manually at first and remind Claude to explicitly reference the competitor gap document if needed.

Step 11: Design realistic AI characters and environments

Most AI ads fail because the people and environments look obviously fake. This workflow fixes that by using detailed image-based planning.

Create a character sheet

Using an image model (e.g., GPT-4o’s image model, Higgsfield, or similar), you have Claude generate a character sheet for your main on-screen persona:

• Consistent facial features and hairstyle
• Typical outfits and colors
• Angles and expressions

This ensures your AI-generated actor looks like the same person across all scenes and ads.

Create a location sheet

Next, you do the same for the environment:

• Living room, kitchen, park, or street where the product is used
• Lighting, decor, props, and overall mood
• Multiple angles of the same space

By defining the environment in advance, you stop the video model from constantly changing backgrounds or introducing weird glitches.

Step 12: Plan the video as 12 image frames

Instead of asking a video model to “just make a full ad,” you break the video into around 12 key frames.

Claude uses your script, character sheet, and location sheet to generate:

• 12 detailed prompts (one per frame)
• 12 still images showing each major moment in the ad (intro, demo, close-up, testimonial, CTA, etc.)

These frames act as a storyboard. Then a video model (e.g., Sora-style systems or similar) stitches them into a smooth video, filling in motion, expression, and lip-sync while staying faithful to the frames.

Again, for your first products, you’ll want to:

• Have Claude generate the 12 prompts
• Review and tweak them
• Tell Claude to learn from the final, approved version

Step 13: Turn frames into full video ads

Once the frames are ready, you instruct Claude (via your creative tool connector) to:

• Generate 1–3 versions of the video from the 12 frames
• Use the approved script for voiceover and timing
• Maintain character and environment consistency

The result: AI-made ads that look like real UGC-style videos, not obvious deepfakes.

Step 14: Let Claude edit the videos in CapCut

Raw AI clips still need editing—music, cuts, captions, and pacing. Instead of doing this yourself, you let Claude handle it inside your browser.

Workflow:

• Open CapCut web and upload the raw video assets
• Enable the Claude Chrome extension and make sure the tab is controllable
• Give Claude a visual reference (e.g., screenshot of a desired timeline layout)
• Prompt it to edit the video: trim, add music, add captions, add effects, and export

Once you’re happy with how it edits the first few videos, tell Claude to treat that as the default editing style for future ads.

Step 15: Launch and manage Facebook ads with Claude

With finished creatives ready, Claude can now launch your campaigns.

Two ways to do it:

• Via Meta Ads connector: Best if your account works smoothly with Claude’s MCP tools.
• Via browser control: Open Business Manager in Chrome and let Claude click through the campaign creation flow.

Claude can:

• Create campaigns, ad sets, and ads
• Set budgets and bid strategies based on your rules
• Upload the edited videos and write ad copy
• Monitor performance and pause or scale according to your limits

If you already have a favorite Facebook strategy (e.g., from a tutorial), you can have Claude transcribe that strategy and then follow it step by step when building your campaigns.

Step 16: Use remote control to run your store from anywhere

One powerful feature of this setup is remote control inside Claude Code. As long as your computer is on, you can:

• Use the Claude app on your phone
• Send instructions that control your desktop Claude session
• Let it use your local hardware (GPU, CPU, RAM) for heavy tasks

This makes your machine a kind of personal AI server that can keep building stores, generating creatives, and managing ads while you’re away.

Train first, then automate

The most important mindset shift is this: you don’t get full autopilot on day one. You first spend a couple of cycles teaching Claude how you want things done:

• Manually review product picks against your criteria
• Check store layouts, copy, and branding
• Refine character sheets, environment sheets, and frame prompts
• Inspect video edits and ad setups

Only after Claude has produced work you’re consistently happy with should you tell it to save those workflows as reusable patterns and run them automatically for each new product or store.

Turning this into a repeatable AI dropshipping machine

Once the full pipeline is dialed in, you can:

• Launch new Shopify stores weekly or monthly
• Let Claude handle product research, store build, creatives, and ads
• Focus your time on strategy, new offers, and checking high-level performance

The 30-day, $12k+ case study shows that with the right rules, connectors, and skills, Claude can be more than a copywriting assistant—it can function as a real operator for your ecommerce business.

If you treat it like a hire, give it the right tools, and train it carefully, you can build a dropshipping system that largely runs itself while you work on the next opportunity.