Claude Skill · Conversion

CRO Page Teardown

Opinions about button colours are worthless. This produces hypotheses tied to a specific friction, with a stated mechanism and a way to measure it — and it won't recommend a test your traffic can't support.

Language · Markdown License · MIT Type · Claude Skill

View on GitHub north_east Fork it fork_right

01 · The Problem

Why This Exists

Most page feedback is a list of preferences dressed as expertise. "Make the CTA bigger" has no mechanism, so it can't be right or wrong — it just ships and nothing changes.

Worse, most teams A/B test at volumes that cannot resolve the difference. Under roughly a thousand conversions a month, testing is theatre; sequential changes and five user interviews will teach you more.

And frequently the page isn't the problem. If the real constraint is the offer or the traffic quality, this skill says that first, before anything else. Most "CRO problems" are targeting problems.

02 · Method

How It Works

01

Read the page as a visitor does

The five-second test from above the fold alone: what is this, who is it for, why this one, what do I do next. If any is unanswerable that's finding #1 and it outranks everything else on the page.

02

Map friction to funnel stage

Every finding is classified attention, motivation, anxiety, or friction. These have different fixes, and anxiety and friction problems are usually far cheaper wins than motivation problems.

03

Write hypotheses with mechanisms

"Because [evidence], [segment] is [stage problem]. Changing X to Y should increase [metric] because [mechanism]." Any hypothesis without a stateable mechanism is rejected, not softened.

04

Split the queue three ways

Just do it (high confidence, low effort, no test needed), test it (real uncertainty and enough traffic, with MDE and runtime), and research it — naming the method and the exact question to ask.

03 · Run It

Two Minutes To First Output

# 1. Clone the repo
git clone https://github.com/seosdm/cro-page-teardown.git

# 2. Copy SKILL.md into a Claude Project, then prompt:
Run cro-page-teardown on [url].
Conversion action: [x]. Traffic source: [y]. ~[n] conversions/month.

# 3. Try it on your own homepage first. It will not be gentle.

No API keys, no setup. The skill is a SKILL.md file plus a worked example — it runs anywhere Claude does.

04 · Proof

Sample Output

Findings by Funnel Stage Example · anonymized

4 Attention

2 Motivation

6 Anxiety

3 Friction

Verdict: the page fails the five-second test on "why this one" — four unfalsifiable claims above the fold and no proof adjacent to the ask. At 340 conversions a month there isn't traffic to A/B test this, so the queue is sequential: fix the six anxiety findings first, then run five interviews before touching the offer.

See the full worked example in /examples.

If The Tool Gets You 60% There

I'll Teach You The Rest

Free skill, real judgment behind it. Want it run on your business, or your team taught to run it themselves?

Bring Me In On Growth north_east