SKILL.md
Conversion Rate Optimization
Diagnose in order of impact, not in order of what you noticed first. A perfect button on a page with a vague value proposition converts nothing.
Setup
Read .agents/product-marketing.md or .claude/product-marketing.md if either exists. Then establish three facts before analyzing: page type (homepage, landing, pricing, feature, blog), the primary conversion goal (signup, demo, purchase, download), and where traffic comes from (organic, paid, email, social). Ask for current conversion rate, prior experiments, and any heatmaps or recordings; analyze without them if unavailable, but say so.
Analysis Framework
Work these seven dimensions top to bottom. Higher rows move the number more.
| # | Dimension | Core check |
|---|---|---|
| 1 | Value proposition | Can a visitor say what this is and why they care within 5 seconds? Benefit-led, specific, in customer language? |
| 2 | Headline | Communicates the value prop, specific, matches the traffic source's message? |
| 3 | CTA | One clear primary action, visible without scrolling, copy states value ("Start Free Trial", not "Submit")? Logical primary vs secondary hierarchy, repeated at decision points? |
| 4 | Visual hierarchy | Does a scanner get the message? Important elements prominent, white space, images supporting not decorating? |
| 5 | Trust signals | Logos, attributed testimonials, real numbers, review scores. Placed near CTAs and after claims? |
| 6 | Objection handling | Price, "will it work for me", implementation effort, "what if it fails" each answered via FAQ, guarantee, or comparison? |
| 7 | Friction | Form fields beyond necessity, unclear next steps, confusing nav, mobile issues, slow load? |
Page-type specifics: landing pages need message match with the ad and a single CTA (kill the nav). Pricing pages need a recommended plan and a "which plan is right for me" answer. Feature pages need feature-to-benefit chains and a path to try. Blog posts need contextual CTAs at natural stopping points.
For experiment ideas per page type and the full form optimization playbook (field reduction, multi-step, error handling): read references/experiments.md when you reach recommendations or when the page is a form.
Output Format
Deliver findings in four buckets:
- Quick wins. Easy changes, likely immediate impact.
- High-impact changes. Bigger lifts worth prioritizing.
- Test ideas. Hypotheses to A/B test, not assume. State the hypothesis and the metric.
- Copy alternatives. 2-3 options with rationale for the headline and primary CTA.
Verification
Do the 5-second test on the current hero: read only the headline, subheadline, and CTA, then state what the product is, who it is for, and what to do next. Expect all three answerable. If not, the top recommendation must be value proposition clarity, whatever else you found.
Then trace the conversion path: click the primary CTA and follow it to completion. Expect no surprise steps, no fields that are not needed. If the path breaks or bloats, friction findings outrank cosmetic ones.
Good vs Bad
Judgment call: what to recommend first on a weak page.
Bad: "Change the button color to orange and add urgency copy." Cosmetic tweaks on top of an unclear offer.
Good: "Visitors can't tell what this does within 5 seconds. Rewrite the headline outcome-first ('Close the books in 3 days, not 3 weeks'), move the logo bar under it, then test button copy." Impact order respected.
Footguns
- Recommending tests when the sample is too small. Low-traffic pages cannot resolve an A/B test in reasonable time. Fix: for low traffic, recommend the best-judgment change directly and say why testing is skipped.
- Optimizing the page when the traffic is wrong. No page converts mismatched intent. Fix: check message match with the traffic source before blaming the page.
- Ignoring what happens after the click. A great page feeding a broken signup flow wastes the win. Fix: always trace the full path.
- Everything-is-important reports. Twenty flat recommendations get zero implemented. Fix: four buckets, ranked, with a clear first move.
Completion Checklist
- [ ] Page type, goal, and traffic source established
- [ ] All seven dimensions assessed in impact order
- [ ] Findings bucketed: quick wins, high-impact, tests, copy alternatives
- [ ] 5-second test and path trace run and reported
- [ ] First recommended move is unambiguous
Any box unchecked: not done. Fix or say so.