“LAUNCH” How I Used Claude To Find The Leaks In My Sales Page (And Build The A/B Test To Fix Them)
Claude will build you a $10,000 looking website in an afternoon. Cool. Most of them convert like trash. Not because they're ugly. Because nobody ever checked what visitors actually do on the page. I ran this exact process on a launch that did over $47,000 in six days. Here's all of it: the free tool, the setup, and every prompt, word for word.
Your Pretty Page Is A Guess
Every landing page is a pile of guesses. You guessed the headline. You guessed where the button goes. You guessed people would scroll far enough to see the offer.
Big companies pay conversion experts five figures to replace those guesses with data. The job has a fancy name, conversion rate optimization, but it's simple. Watch what real visitors do. Find where they quit. Change one thing. Measure if it worked. Repeat.
You're going to do that whole job with a free tool and three prompts. No analytics background needed. If you can paste, you can do this.
MY REAL NUMBERS
What A $47,000 Launch Looks Like Page By Page
This is my last launch. A $297 product, six days, doors closed September 13. Orders come from my payment system, visitors from my analytics.
2,726
visited the sales page
894
clicked through to the cart (32.8%)
146
bought (16.3% of cart visitors)
31
took the $197 upsell (21.2% of buyers)
Overall: 5.4% of sales page visitors bought.That was mostly warm traffic (my email list and my YouTube viewers), so don't expect cold ads to do the same.
Now look at where people left. 67% never clicked through to the cart. And 84% of people who reached the cart didn't buy. Those two spots are where the money is.
Without data you would never know which page to fix first. You'd just redesign the hero again. My own page had decisions I was guessing on, like hiding the buy button at the top for the first five minutes. Smart or expensive? A guess can't tell you. A test can.
STEP 1
Install Microsoft Clarity (Free Forever)
Clarity is Microsoft's free tool for watching what visitors do on your site. No traffic limits. No trial. It gives you three things:
Heatmaps. Where people click and how far they scroll.
Session recordings. Actual replays of real visitors using your page.
Frustration signals. Rage clicks, dead clicks, people who land and bounce right back.
Setup takes five minutes
Go to clarity.microsoft.com and sign up.
Create a project for your website.
Choose Install manually and copy the tracking code.
Paste the prompt below into Claude, in the project where your website lives.
Install prompt
Install this Microsoft Clarity tracking code on every page of my site, inside the <head>.
Rules:
1. Make it only fire on my live domain, so my local previews and my own test visits stay out of the data.
2. If my funnel uses more than one tool (for example a landing page here and a checkout page somewhere else), tell me exactly where to paste the same code on the other pages so the whole funnel is tracked in one Clarity project.
3. When you're done, tell me how to confirm it's working.
[PASTE YOUR CLARITY TRACKING CODE HERE]
Not using Claude Code?If your page lives in a page builder (GoHighLevel, Shopify, WordPress, Webflow, Wix), look for a setting called “head tracking code” or “custom code” and paste the Clarity code there. Ask Claude: “Where do I paste a head tracking code in [YOUR TOOL]?”
Then wait
Clarity needs real visitors before it can tell you anything. Give it a few days. Aim for at least 100 sessions per page before you trust what you see. Ten visitors is a story, not a pattern.
One thing to handle like an adult: Clarity records visitor sessions. It hides sensitive fields like card numbers by default, but mention it in your privacy policy, and if you sell to Europe, put it behind your cookie banner.
THE DECODER
What Clarity Is Actually Telling You
Claude reads all of this for you. But you should know what the words mean, so you can tell a good finding from a lazy one.
Scroll depth
How far down the page the average visitor gets.
What it usually means: If your offer sits at 80% down the page and the average visitor reaches 35%, most people never see what you sell. Move the important stuff up or give them a reason to keep going.
Dead clicks
Someone clicked something and nothing happened.
What it usually means: They thought it was a button or a link. Usually an image, a headline or a price box. Either make it clickable or make it stop looking clickable.
Rage clicks
Someone clicked the same spot over and over, fast.
What it usually means: Something is broken or slow, and they're annoyed. On a checkout page this is an emergency. Fix first, test later.
Quick backs
Someone landed on a page and went straight back.
What it usually means: The page wasn't what they expected. Usually the link or button that sent them there promised one thing and the page opened with another.
Excessive scrolling
Someone scrolled up and down way more than normal.
What it usually means: They're hunting for something and can't find it. Price, guarantee, what's included, how to buy. The recordings tell you which.
STEP 2
Turn Claude Into Your Conversion Optimization Expert
Claude needs a browser for this part, because heatmaps and recordings only live inside the Clarity dashboard. Use the Claude desktop app (it has a browser built in) or the Claude in Chrome extension.
Sign in to Clarity yourself first. Never paste passwords into a chat. Then paste this prompt and fill in the brackets.
Prompt 1: the Clarity audit
You are a senior conversion rate optimization expert. Your job is to audit my funnel using Microsoft Clarity and hand me A/B tests worth running.
MY FUNNEL
- Clarity project: [PROJECT NAME]
- Pages to audit, in funnel order: [SALES PAGE URL], [CART PAGE URL], [UPSELL PAGE URL]
- Date range: [FOR EXAMPLE: THE LAST 14 DAYS, OR MY LAUNCH WEEK]
- The goal of the funnel: [FOR EXAMPLE: SELL A $297 COURSE, THEN A $97/MO UPSELL]
- Where my traffic comes from: [FOR EXAMPLE: MY EMAIL LIST, YOUTUBE, PAID ADS]
I am already signed in to Clarity in the browser. Use the browser to do this research yourself. Do not ask me to look things up for you.
STEP 1: GET THE NUMBERS
Open the Clarity dashboard for the date range. For each page, record sessions, average scroll depth, engagement time, and the rate of rage clicks, dead clicks, quick backs and excessive scrolling. Split everything by mobile vs desktop. Find the page and device where the funnel leaks the most. That is where you spend most of your time.
STEP 2: READ THE HEATMAPS
For each page, open the heatmaps on both mobile and desktop.
- Scroll map: note the exact sections where big chunks of visitors drop off, and what percent reach each call to action button.
- Click map: note what gets clicked most, what gets clicked that is not clickable, and which buttons get ignored.
- Attention map, if it's available: note which sections people actually stop and read, and which they fly past.
Take a screenshot of anything important so you can show me.
STEP 3: WATCH RECORDINGS
Watch at least 10 session recordings per page, picked on purpose:
- Visitors who reached checkout and did not buy
- Sessions with rage clicks or dead clicks
- Sessions with quick backs
- A few buyers, so you know what a winning session looks like
- A majority on whichever device converts worse
For each one write one line: device, traffic source, where they hesitated, where they left, and your best read on why. Then tell me the patterns that repeat. One weird session is noise. The same behavior in four sessions is a finding.
STEP 4: RUN THE EXPERT CHECKLIST
Now look at each page the way a conversion expert would, and check it against what the data showed you:
- First screen: can a stranger tell what this is, who it's for, and what to do next, without scrolling?
- Message match: does the headline match what the ad, email or video promised before they clicked?
- The button: is the main call to action visible early, easy to tap on a phone, and does the text say what happens next?
- Proof: are testimonials and results placed BEFORE the points where the scroll map shows people leaving?
- Price and offer: when does the price show up, and is the value clear before it does?
- Friction: count every field, click and page between "I want this" and "I bought it." Flag anything that isn't needed.
- Mobile: anything cut off, overlapping, too small to tap, or slow to load?
- Distractions: links, menus or popups that pull people off the path to buying?
STEP 5: DIAGNOSE
List the top 5 conversion problems across the funnel, ranked by how much money they are likely costing me. For every problem, cite the evidence: the number, the heatmap, or the recordings that prove it. If you are guessing, say you are guessing. Do not give me generic best practices. If the data does not show it, leave it out.
STEP 6: A/B TEST PLAN
Turn the findings into 3 to 5 A/B tests. For each test give me:
- Hypothesis: "Because we saw [evidence], we believe [change] will [result]."
- Control vs variant: exactly what changes, with the new copy or layout written out word for word
- Page and device it targets
- Primary metric to judge the winner
- Expected impact (low / medium / high) and effort to build (low / medium / high)
Rank them by impact vs effort and tell me which ONE to run first and why. Each test changes one thing only, so I know what caused the result.
Also give me a separate short list of "just fix it" items: broken things, dead clicks on elements that look clickable, mobile layout bugs. Those do not need a test.
RULES
- Be blunt. I want the truth about my page, not compliments.
- Real data only. Never invent a number. If Clarity does not have enough sessions to support a conclusion, tell me.
- Stay read-only in Clarity. Do not change any settings, filters I have saved, or anything on my live pages.
- If you cannot access something (a heatmap will not load, recordings are missing), tell me what you could not see instead of working around it silently.
Finish with a one paragraph summary I could read out loud: the biggest leak, the evidence, and the first test to run. Then stop and ask me which suggestions I want to implement.
Why the prompt is built this way
Numbers first. Claude finds the leakiest page and device before it looks at anything else. That stops it from wasting an hour on a page that's fine.
Recordings picked on purpose. Random recordings are mostly boring. The prompt sends Claude to the sessions that teach you something: people who almost bought, people who got angry, and a few buyers for comparison.
The expert checklist. This is the part a paid consultant brings. First screen, message match, button, proof, price, friction, mobile, distractions. Claude checks each one against what the data showed.
Evidence or it doesn't count. Every problem has to point to a number, a heatmap or a recording. That kills the generic “add more social proof” advice.
What comes back
The top 5 problems costing you money, ranked, each one backed by evidence
3 to 5 A/B tests with the new copy written out word for word
The ONE test to run first
A “just fix it” list: broken stuff that doesn't need a test
The recordings step is slow. Claude is watching real sessions one by one. Go make a coffee.
STEP 3
Let Claude Build The A/B Test
An A/B test is simple. Half your visitors see your current page (Version A). Half see the new one (Version B). Whichever makes more money wins.
Pick the suggestions you like from the audit. Then paste the second prompt.
Prompt 2: build the A/B test
BUILD THE A/B TEST
Here are the suggestions I want to implement: [LIST YOUR PICKS FROM THE AUDIT]
My conversion goal is: [PURCHASES / OPT-INS / BOTH]
- Opt-in = [FOR EXAMPLE: THE FORM IS SUBMITTED, OR THE THANK YOU PAGE LOADS]
- Purchase = [FOR EXAMPLE: THE ORDER CONFIRMATION PAGE LOADS]
1. LOOK BEFORE YOU BUILD
Inspect how my landing page is built and hosted (the code folder I give you, or the page builder it lives in). Tell me in plain English how you plan to run the test on my setup before you touch anything. If my setup cannot support a clean test, say so and give me the closest option that works.
2. BUILD THE VARIANT
- Version A is my current page. Leave it exactly as it is.
- Version B is a copy of it with ONLY the changes I picked. Nothing else changes. No bonus "improvements."
- Show me Version B as a preview with desktop and mobile screenshots. Do not put anything live until I say go.
3. SPLIT THE TRAFFIC
- Send 50% of visitors to A and 50% to B, at random.
- A visitor who comes back must see the same version they saw the first time (save their version in a cookie).
- Both versions keep the same URL for ads and emails, so I do not have to change any links.
- Keep my existing tracking parameters (utm tags) intact through the split.
- Give me a preview link for each version so I can force A or B to check them. Those preview visits must not count in the results.
- Give me a one step kill switch that sends 100% of traffic back to Version A.
4. TRACK CONVERSIONS PER VERSION
Every conversion must be credited to the version that visitor saw. If checkout happens on a different domain or tool than the landing page, work out how to carry the version through (URL parameter, hidden form field, or tag on the contact) and tell me how you did it. Count each visitor once, even if they reload or come back.
Also tag the version inside Microsoft Clarity as a custom tag, so I can filter heatmaps and recordings by Version A vs Version B later.
5. BUILD THE SCOREBOARD
Create a simple results page I can open anytime that shows, for A and B side by side:
- Visitors
- Opt-ins and opt-in rate
- Purchases and purchase rate
- Revenue per visitor, if purchase amounts are available
- The lift of B over A as a percentage
- A confidence reading: how sure we can be that the difference is real and not luck
Rules for the scoreboard:
- It must say "NOT ENOUGH DATA YET" until each version has at least [100] conversions AND the test has run [14] full days. No declaring a winner early off a lucky weekend.
- It only calls a winner at 95% confidence or higher. Below that it says "no clear winner yet."
- Filter out bots, my own visits, and the preview links.
- Show results split by mobile and desktop too, since one version can win on one device and lose on the other.
6. PROVE IT WORKS BEFORE LAUNCH
Before going live, run a test visit through each version and show me that the visit and a test conversion both show up on the scoreboard under the right version. Then remove the test data.
7. HAND IT OVER
Tell me in a short list: where everything lives, how to read the scoreboard, how to hit the kill switch, and what to do when there is a winner (make the winner the main page, remove the losing version and the split code).
RULES
- Nothing goes live without my explicit "go."
- Never edit or delete my original page. If anything breaks, Version A must keep working on its own.
- Do not collect any personal data for the test beyond what my page already collects.
- If you need a login, an API key, or access you do not have, stop and tell me. Do not ask me to paste passwords or keys into the chat.
Claude builds Version B with only the changes you picked. It splits your traffic 50/50 on the same link, so you don't touch your ads or emails. It credits every opt-in and every purchase to the version that person saw. Then it builds you a scoreboard page showing which version is winning.
The part most DIY tests get wrong.If your checkout lives on a different tool than your landing page (mine does), the version has to be carried across that hop. If it isn't, your scoreboard counts clicks, not money. The prompt forces Claude to handle it.
This step works best in Claude Code or the Claude desktop app with your website's files open, because Claude needs to edit the page. If your page lives in a page builder, Claude will tell you the closest clean way to split traffic on your setup.
KEEP GOING
4 Follow-Up Prompts Worth Saving
The audit is a starting point. These get you more out of it.
Show me the proof
For problem #1 in your audit, open the 3 recordings that prove it best. Give me the link to each one and the exact timestamp where the problem happens, so I can watch it myself.
Mobile only
Redo the audit for mobile visitors only. Ignore desktop completely. I want the top 3 problems that only exist on a phone, with screenshots of each.
Rewrite from the evidence
Based on where the scroll map shows people leaving, rewrite my headline and the first screen 5 different ways. For each version, tell me which finding from the audit it is fixing. Keep my voice and do not invent any results, numbers or testimonials.
Weekly check-in
Open my A/B test scoreboard and my Clarity dashboard. Give me a 5 line update: visitors per version, conversions per version, whether we have enough data yet, anything broken, and whether I should do anything this week. If the answer is "wait," say wait.
READ THIS TWICE
Don't Fool Yourself
An A/B test can lie to you if you let it.
Change one thing per test. Change five and you'll never know what worked.
Don't call a winner early. One lucky weekend means nothing. The scoreboard says “not enough data yet” until each version has about 100 conversions AND two full weeks of data. Leave it alone until then.
Small traffic? Test opt-ins first. At a 2% to 3% purchase rate, 100 purchases per version takes thousands of visitors. Opt-ins pile up way faster, so you get answers in weeks, not months.
Test big swings. A new headline, a new offer angle, moving the button up. Button colors are a waste of your traffic.
Fix broken stuff without a test. A dead button doesn't need a hypothesis. It needs fixing today.
Then you run it again. Find the leak, test the fix, keep the winner, repeat. That loop is the whole job of a conversion expert. Now it's a prompt.
Quick Answers
Is Clarity really free?
Yes. Microsoft says free forever, with no traffic limits. You need a Microsoft, Google or Facebook login to sign up.
Do I need Claude Code?
For the audit, no. Any Claude that can use a browser works. For building the test, Claude needs to edit your page, so Claude Code or the desktop app with your site's files is the easy path.
My page gets 20 visitors a day. Is this worth it?
The audit is, today. Recordings of 50 real visitors will show you broken things no amount of staring at your own page will. The A/B test needs more traffic, so test opt-ins, test big changes, and be patient.
Can Claude see my visitors' private info?
Clarity hides sensitive fields by default, and the audit prompt keeps Claude read-only inside your dashboard. You stay signed in yourself. Claude never gets your password.
Checked on September 19, 2026. Tools change their plans and menus, so check their pages for current details. I have no affiliation with Microsoft Clarity. My launch numbers are my own results from warm traffic and are not a promise of what you'll get.
The Next Step
Guides show you what. AI Builders shows you how.
Inside the community, I walk through every build live, including the stuff that doesn't make it into guides. Regular people (not developers) figuring out AI together, shipping real projects, asking me anything. No fluff, no theory, just the actual work.