How to write the one difference between two homepage variants
Two homepage variants are a test only when you can say what is different in one sentence. Here is how to write that sentence before you submit either page.
Picture Sam and Ana, an illustrative pair of cofounders, standing at a whiteboard in a Denver coworking space on a Thursday. The board holds two homepage sketches. Sam has changed the headline, the hero image, the button, and the first proof point. Ana has labeled the right-hand sketch "variant B" as if the label made it an experiment. They are about to submit both to LandingCrush and wait for a crush rate to pick a winner. The rates will arrive. The sentence that would make the rates mean something is still missing.
A variant is not a second design. A variant is a page that keeps the offer still and changes one thing a voter can notice. LandingCrush can hold both pages as separate homepage submissions, show them in the swipe feed after approval, and report a crush rate on each. It cannot untangle four changes after the fact. You do that work before you upload. This article is the worksheet. If you want the timed version of the same idea, read how to run a homepage duel after the sentence exists.
Start from the offer, not from the mock
Write the offer in one line that both variants must keep. The offer is who it is for, what they get, and what they are asked to do. "Finance leads at companies with a controller. A weekly close checklist. Book a demo." Both screenshots have to carry that offer. If variant B changes the buyer, you are not testing a homepage. You are testing a different product story, and a crush for B means people preferred a different promise. That can be a useful strategic argument. It is a bad thing to discover after you thought you were picking a typeface.
Put the offer line above both images in the doc. Anyone who edits either file has to leave that line true. This is the rule that stops a "quick visual pass" from becoming a second company.
You might be thinking real design does not change one atom at a time. You are right that a shipped page is a whole. You are wrong that a test can explain a whole. When everything moves, the vote still happens. You just lose the right to say why. If you need a broad visual direction first, use the Style Explorer as a way to look at directions, then freeze one direction and test a single difference inside it. Do not submit five aesthetics and call the favorite an A/B result.
Name the difference in a sentence a stranger can check
Use this shape: "A and B show the same offer. B changes only ___." Fill the blank with something a person can point at on the screenshot without your help.
Strong blanks are visible. The headline names a finance lead instead of "teams." The button says "Book a demo" instead of "Get started." The first proof is three customer logos instead of a paragraph. The mobile capture stacks the button above the paragraph instead of below it.
Weak blanks are intentions. "B feels clearer." "B is more premium." "B has better hierarchy." A voter cannot point at a feeling. You cannot implement a feeling the same way twice. If the only honest blank is a feeling, you do not have a variant yet. You have a preference. Sit with the screenshots until you can point.
Read the sentence to someone who has not been in the draft. Ask them to find the difference on the images. If they find a second difference you forgot, either remove it or add it to the sentence and accept that you are no longer running a one-change test. Hidden differences are how teams fool themselves. The illustration changed because the new headline was a different length. Say so. Then decide whether the illustration change is small enough to live with, or whether you should rebuild B so the picture stays put.
Cut the variant until the sentence is true
This is the unglamorous part. Open B and revert everything the sentence does not mention. Same navigation. Same logo. Same background treatment, unless the background is the difference. Same offer line. Same proof, unless the proof is the difference. The page can still look considered. It should look like A with one deliberate edit, not like a redesign that happened to share a logo.
If the change is a headline, keep the line length from forcing a new layout. A headline test that also moves the button has tested a stack, not a sentence. How to test a headline without changing the offer is the tighter version of this case. Use it when the words are the only argument.
If the change is the button label, do not also change the color, the size, and the destination story. "Book a demo" versus "Start free" is already two offers if one is a meeting and one is an account. Check the offer line. If the action changed, the offer changed. Either update the offer line and admit you are comparing promises, or put the action back.
Write the sentence under both screenshots in the file you will submit. Future you, reading comments in ten days, needs the sentence more than future you needs another adjective.
Submit the pair so the product can tell them apart
Each variant is its own homepage submission at /submit. Paste the URL or upload the screenshots that match the frozen files. Give them names a teammate can tell apart in the dashboard. "Close checklist, buyer named" and "Close checklist, buyer unnamed" are names. "Final" and "Final v2" are how you lose the thread.
The free plan includes one active homepage. Two live variants need Creator Pro, which is $19 a month or $190 a year and includes unlimited submissions. Wait for both to be approved before you compare anything. A pending page has no crush rate. Do not compare an approved page to a file that is still on your desktop.
When both are live, do not watch the rates hourly and rewrite. You decided the difference already. Let votes accumulate. The comparison is allowed when the counts are in a similar range. Twelve votes against eighty is not a result. The product's own caution, when it checks a panel prediction, waits for 50 or more community votes. Use that as a floor for calling a preference, not as a magic threshold that creates statistical certainty. LandingCrush is not a traffic splitter on your site. Say the count when you say the rate.
Comments matter more than the gap between two close rates. Read the comments that mention the thing in your sentence. If people talk about a part you thought was identical, your sentence was wrong and the test leaked. Note the leak. Do not pretend the headline won because the rate moved two points while the comments are about the picture.
Record the outcome in the same sentence you started with
Fill this in when you stop the comparison:
"We kept the offer: ___. We changed only ___. Page A had ___ votes and a crush rate of ___. Page B had ___ votes and a crush rate of ___. Comments about the change said ___. We will ship ___ because ___."
If the because-clause mentions conversion, delete it and write what you actually know. You know which submitted page received more crushes, if the counts support that claim. You will know about conversion after the live page changes and your analytics have a before and after. How to ship the winner is that next step. Do not skip ahead inside this doc.
If the rates are close, the outcome is "no preference." Keep the version that is simpler to build and support. A one-difference test that finds no difference is a success. It saved you from a redesign.
The pair you submit next should be boring to anyone who wanted a reveal. Boring means the sentence was true. Which one line will you refuse to let variant B rewrite?
Common questions
Q: Can I test a full redesign as variant B?
A: You can submit it. Call it a new direction, not a one-difference test. You will get a crush rate. You will not get a clean reason.
Q: Do both variants need their own submission?
A: Yes. LandingCrush scores the homepage you submitted. A second image that never goes through approval does not collect a crush rate.
Q: What if the rates are almost the same?
A: Record no preference and keep the simpler page. A small gap on a modest count is not a mandate to rebuild.