Most CRO advice is variance noise dressed up as wisdom. "Move the CTA up." "Test the button colour." "Add social proof." A few percentage points either way, indistinguishable from noise, presented as a science.
This isn't that.
If you ship landing pages and want to actually move conversion, three things matter. They cover maybe 85% of the gap between a page that works and one that doesn't. Most of the rest is statistical noise. Interesting on a dashboard, invisible to revenue.
What "conversion rate" actually measures
Before the checklist, a thing worth saying out loud: conversion rate is a ratio. Two numerators, one denominator. Most "CRO wins" are denominator tricks.
If you swap your homepage hero for a stronger one and your conversion rate doubles, you might have improved persuasion. Or you might have filtered out the casual traffic that was never going to convert anyway. The absolute number of conversions barely moved. The rate looks great in a screenshot.
Watch absolute conversions per channel, not rate. The rate is a vanity metric until you can hold traffic mix constant.
The three things that move the needle
1. The first 200 words
The headline, subhead, and first paragraph carry 90% of the work on most landing pages. They decide whether the visitor scrolls.
What "carry" means here is specific:
- The headline names the person, not the product. "Stop losing your weekends to broken builds" beats "Continuous integration platform" by a wide margin for engineering tools.
- The subhead adds the offer in concrete terms. Not "powerful platform." A specific outcome the visitor can picture.
- The first paragraph closes the credibility gap. Why should they believe this site? Names, numbers, a screenshot. Something verifiable in under five seconds.
I've watched pages double click-through on this section alone, with no other changes. Everything else is in service of these 200 words. If you only have time for one improvement, this is it.
2. The decision path
Most landing pages list features. The page reads like a brochure that's been chopped into bullets.
The conversion-optimised version is different. It walks the visitor through the questions they're already asking, in the order they're asking them, and resolves each one before moving on.
The order isn't "what's the product, how does it work, who uses it." It's closer to:
- Is this for me?
- Will this work for what I need?
- What's the catch?
- How much, and how do I start?
If your page answers these in this order, the visitor finishes it oriented. If it answers them in any other order, half drop off before they get the information they actually wanted.
A concrete test: read your page top to bottom and after each section ask "Did this just answer the question I'd be asking right now?" Most pages fail this on the second or third section. They jump to features before establishing fit.
3. The actual offer
The CTA is the easy part. The hard part is what comes after the click.
"Get started" goes to a sign-up form. The form has eight fields. Three are required. The submit button says "Submit" because someone forgot to relabel it. The next page is an empty dashboard.
This is where landing pages quietly leak. Not at the CTA. At step two.
If your CTA is honest about what happens next ("Start a free trial: no credit card, 30-second sign-up"), and the next step matches, you keep the visitor. If there's friction or surprise between the click and the value, you lose them. Conversion rate on the landing page looks fine. Activation rate is the bleed.
Test the whole flow as a stranger. Fresh browser, no cookies. From "I just heard about this" to "I'm using it." Count the seconds and the clicks. Under 30 seconds and 3 clicks is the bar.
What I stopped doing
Things I used to recommend that I now skip:
A/B testing button colour. Stops mattering past a few thousand visitors per variant per week. Below that, you're reading noise as signal.
Social proof above the fold. Sometimes helps, sometimes hurts. Depends on the customer base. Test it last, not first.
Exit-intent popups. Recover a fraction of a percent in conversions and annoy the people who would have come back anyway. Net negative on goodwill once you factor in brand perception.
Heatmaps without a hypothesis. They tell you where people clicked. They don't tell you why they didn't convert. Use them to confirm hypotheses, not generate them.
Long-form testimonials. Three short ones with a face, role, and a specific result beat one TED-talk paragraph every time.
What I tested and dropped
A few things that sounded right but didn't survive contact with real traffic:
- "Move the CTA above the fold." Doesn't help if the visitor isn't convinced yet. Sometimes hurts: people see the CTA, realise they don't know what they're committing to, and bounce.
- "Use video for the hero." Sometimes works for consumer products. For B2B, usually a slow loading penalty for no measurable lift.
- "Add a chatbot." Adds friction more often than it removes it. The traffic ready to chat is also the traffic ready to fill out a form.
- "Personalise by traffic source." Works at enterprise scale and breaks below it. You need volume to test variants without going insane.
This is what I've watched fail to move conversion across projects we've shipped. Your mileage may vary, but variance is the point.
A checklist worth running
Print this. Run it before any landing page launch:
- Headline names the person, not the product.
- Subhead adds a specific outcome.
- First paragraph has a verifiable credibility signal.
- Page answers visitor questions in the order they ask them.
- CTA is honest about what happens after the click.
- The post-click flow is under 30 seconds and 3 clicks to first value.
- Page loads in under 1 second on a mid-range mobile device.
- No broken images. No broken links. No console errors.
If you can tick all eight, you've done the 85% that matters. The other 15% is noise dressed as science. Skip it and ship.
If you want a landing page built with this checklist baked in from the start, see how we work on landing pages. One page. One job. Designed to beat the templated alternative, not just exist next to it.