How to build a sales funnel with Claude Code (and the parts it gets wrong from scratch)

You can build a sales funnel with Claude Code, and it will do the pages and the copy very well. The parts it gets wrong when it starts from a blank folder are the ones that touch money: trusting prices from the browser, fulfilling an order twice, keeping downloads open after a refund, and one-click upsells that fail when the bank asks for 3-D Secure. The reliable way is to give Claude Code a funnel that's already built and tested, plus a playbook that tells it what to do, and let it spend its time on your offer instead of on payment plumbing.

This post covers both routes, what a funnel needs, and the exact session, with the prompts to paste.

What a sales funnel needs

Before any tool, the list. A working funnel has six parts:

1. A sales page that says what you sell, who it's for and what it costs.
2. A checkout on that page, with your product, your price and your brand.
3. An order bump: one small add-on right before payment.
4. A one-click upsell after payment, on the card they just used, and a lighter downsell if they say no. (How one-click upsells work.)
5. Delivery and emails: the download or course access, the order email, a free gift for people not ready to buy, a welcome series, a checkout reminder.
6. Numbers: where visitors came from, who clicked buy, who paid, which offers they took.

If any of the six is missing, it's a page with a button, not a funnel.

What Claude Code does well from scratch

Give Claude Code a blank folder and a good brief and it will:

• Write the sales page. Structure, headline options, the FAQ, the guarantee. It's genuinely good at this, especially when you give it real details about your buyer.
• Build a clean, fast page in plain HTML and CSS, on your brand, working on a phone.
• Wire a basic Stripe Checkout for one product: a button that opens Stripe's hosted page and sends people to a thank-you page.
• Write the emails in your voice.

For a single product with no add-ons and no delivery, that might be enough. Most people want more than that, and that's where it gets hard.

Where from-scratch builds break

None of this is a knock on Claude. These are the parts of a funnel that are easy to write and hard to get right, because the bugs only show up with real money, real banks and real refunds:

• Prices from the browser. If the price sits in the page and the server believes it, anyone can buy your $297 product for $1.
• Webhooks. Stripe resends events until your server answers. If fulfilment runs on every delivery, a buyer can get two accounts, two emails or two commissions.
• Refunds and chargebacks. Most first builds deliver on payment and never check again. Refund someone and their download link keeps working forever.
• Ungated downloads. The PDF sits in a public folder with a guessable name. Someone shares the link and it's free.
• One-click upsells. Charging a saved card needs the card saved for later use at checkout, and a fallback for when the bank wants the buyer to confirm (3-D Secure). Without it, the upsell fails for a slice of buyers in Europe and elsewhere, and you never see why.
• Payment plans. "Three payments of $109" is a subscription told to stop. Forget that and the fourth payment goes out.
• Email deliverability. Order emails need a verified domain with SPF, DKIM and DMARC, or they land in spam and buyers think they've been robbed.
• Analytics. Knowing that 6 people paid is easy. Knowing that 34 saw checkout and which source they came from takes event tracking that most first builds skip.

Each of these is fixable. The full list, with how to check each one, is in what AI-written Stripe code gets wrong. The point here is that you only find most of them after launch, from an angry email.

Two routes

Route 1: from scratch. You describe the funnel, Claude Code writes all of it, and you test it yourself. It works if you're technical, you enjoy the build, and you'll read every line of the payment code. Budget a few days to get the money parts right, and keep maintaining them.

Route 2: a tested framework plus a playbook. You give Claude Code a funnel that's already built, with a written playbook it follows step by step. It spends its time on your offer, your page and your copy, and the payment parts arrive tested. That's what Funnelr is: a folder with the engine, a test kit, a fake Stripe for local testing, and an AGENTS.md file that tells the agent what to ask you, what to build, what to test and when to stop and wait for your OK.

The rest of this post is route 2, because it's the one that gets a non-developer live with real money in an afternoon.

The session, step by step

1. The brief

Before you open anything, write down five lines: who it's for, what they get, the price, one add-on for the order bump, one next step for after payment. Add a free gift for people who aren't ready. Then open the Funnelr folder in Claude Code and paste:

Read AGENTS.md and help me launch my first funnel. I sell a 6-lesson video course on meal prep for shift workers for $79. Order bump: the printable meal planner, $19. One-click upsell: a 20-minute call, $97, with a $49 recorded Q&A as the downsell. Free gift: a one-page grocery list. Brand: warm, plain, cream and tomato red.

The playbook makes Claude ask what you haven't said (your domain, a support email, the business name for the legal pages) and suggest answers, so you can mostly say yes.

2. The build

Claude makes the site folder from a template and fills it in: the config (products, prices in whole cents, the bump, the upsell and downsell, the free gift), the sales page in a proven order, the offer page, the emails, the refund, terms and privacy pages. Your files go in a private folder that's only served to people who paid.

It also does a design pass. The playbook has it pick a direction for your brand, take screenshots of every screen on desktop and phone, critique them and fix what's off. An automatic design check flags low contrast, tiny text and anything that overflows on a phone.

Read the copy. All of it. Then say what's wrong in plain words:

The headline is too clever. Make it say what they get. And the upsell page should say the call is on Zoom.

3. The test, with fake money

This is the step that saves you. Claude runs the site on your computer with a fake Stripe that pays instantly, then runs the test kit: checkout, bump, one-click upsell, decline then downsell, download, refund closing access, free gift, admin sign-in. It fixes anything that fails and runs it again.

Then you buy it yourself, on your laptop and your phone. Take the upsell once, decline it once. Open the download page. Open the admin and find your test order.

4. Connect your accounts

Three accounts, each with a free plan to start:

• Cloudflare hosts the site.
• Stripe takes the payments. Test mode first.
• Resend sends the emails from your domain.

Claude tells you exactly which key to make and which file to paste it into. Keys go in a file on your computer, never in the chat. You can paste this when you're ready:

I've made my Cloudflare, Stripe and Resend accounts. Walk me through the keys one at a time, then run a dry-run deploy so we know it'll work.

The dry run builds everything and has Cloudflare check it without uploading anything. Then the real deploy, then doctor, the setup checker, which confirms every key, the domain, email and the stats are working.

5. Live

Buy it with Stripe's test card on the live site. When it works, switch to your live key, buy it for real yourself, then refund it. That one purchase proves money in, delivery, and the refund closing access. Then tell people. The weekend launch plan covers what to post and how to read the first numbers.

Not on Claude Code? Same steps

Nothing here depends on one agent. The playbook is plain markdown, and the commands are plain node commands, so Cursor, Codex and Gemini CLI follow the same steps. The folder ships CLAUDE.md and GEMINI.md pointers to the same AGENTS.md, so whichever agent opens it finds the instructions.

What it costs

• Funnelr: once.
• Cloudflare: the free plan covers a small funnel; Workers Paid is $5 a month when traffic grows.
• Stripe: its card fees per sale, nothing monthly.
• Resend: free up to 3,000 emails a month.
• A domain: about $10 a year.

Compared with renting a funnel platform every month, the comparison page has the three-year numbers. If you're weighing a course platform specifically, see Claude Code vs Kajabi or Teachable.

Already have a page from another AI tool?

If Lovable, Bolt or v0 already made you a page you like, you don't have to start over. Your vibe-coded landing page can't take money yet covers how to keep it and add the funnel.

Not sure which route fits you? The two-minute quiz gives a straight answer, including "don't buy it yet" when that's the honest one. Or start with the free guide, which walks through the 30-minute version minute by minute.

FAQ

Can Claude Code build a sales funnel?

Yes. From scratch it builds the pages, copy and a basic Stripe checkout well. The money parts (server-side prices, webhook duplicates, refunds closing access, one-click upsells with 3-D Secure, payment plans) are where from-scratch builds usually have bugs. Giving it a tested funnel framework avoids most of them.

Do I need to know how to code?

No, if you use a framework with a playbook. You make the decisions, read the copy, test the purchase and create your accounts. Claude writes, tests and deploys the code. From scratch, you'll want to be able to read the payment code yourself.

How long does it take?

With a tested framework, about 30 minutes to a working funnel on your computer, and an afternoon to live payments, most of it creating accounts. From scratch, plan on days for the payment and delivery parts alone.

Does this work with Cursor, Codex or Gemini CLI?

Yes. The playbook is a markdown file any coding agent can read, and every step is a terminal command, so the session is the same in each.

#Claude Code#AI#sales funnels#Stripe#launching