A/B testing
Two headlines. One sentence that says which won.
Pick a form, write a second headline and a second button label, choose how much of your traffic sees it. PopzIQ keeps each visitor on one side, counts views and sign-ups per side and writes one plain sentence about what happened.
No p values, no lift percentages and no chart to interpret. The reader is choosing between two headlines, not licensing a drug.
On every plan, including the free one
How it works
Six things that are true of every test
Each one is a rule in the code rather than a habit, so a test on your workspace behaves the way this page describes it or a named test fails.
Two versions of the words
A test changes the headline, the button label or both. It is one form with two sets of words, so there is no second form to keep in step and no way for the two sides to drift apart while a test runs.
A visitor sees one side, always
Each browser mints its own random key the first time it meets a form running a test and keeps it in its own storage. The key never leaves that browser, it is not a cookie and we never see it, so a returning visitor gets the same version without us knowing who they are.
The split is arithmetic, not a coin
The key and the test id are hashed into one of a hundred buckets. The buckets under your split go to B. Nothing random happens at display time, so the same visitor on the same form lands in the same bucket every time.
Views and sign-ups are events
Every view and every submission is an ordinary row in the event stream carrying the test id and the letter. The conversion rate is counted from the same log automations and webhooks read, so there is no second analytics store to disagree with it.
A hundred visitors a side before we speak
Under a hundred views on either side the verdict is "Not enough visitors yet to say." A two in a hundred difference on forty visitors is one person having a good day and we will not dress it up as a result.
Ninety percent, never a hundred
Below ninety percent confidence the sentence says the two are too close and tells you to keep both running. Above it, the number is capped at ninety-nine, because no sample supports being certain.
What it says
Three states, three sentences
A test is in one of three states and each one has its own sentence. These are the exact strings the product writes, for the exact counts beside them.
What the form's test page says
Early on
A 40 views, 3 sign-ups. B 38 views, 5 sign-ups.
Not enough visitors yet to say.
A real difference
A 1204 views, 61 sign-ups. B 1198 views, 88 sign-ups.
B is converting better, 7 in 100 against 5 in 100, and we are 98% sure that is not chance.
No real difference
A 900 views, 72 sign-ups. B 910 views, 76 sign-ups.
Too close to call so far. A is converting at 8 in 100 and B at 8 in 100. Keep both running.
The numbers
Every floor, written down
A tool that decides for you should say where its own limits are. These are the constants the product runs on.
| What | Value | Why |
|---|---|---|
| Versions per test | 2 | A and B. A third version would need three times the traffic to separate and most small businesses do not have it. |
| What a version can change | Headline and button label | A closed list of two. Everything else on the form stays identical, so a difference in the numbers is a difference in the words. |
| Buckets | 100 | Your split is a whole percent. Fifty means half, ten means one visitor in ten sees B. |
| Views per side before a verdict | 100 | Under it the sentence refuses to guess rather than printing a rate nobody should act on. |
| Confidence before a winner is named | 90% | Below it the sentence says the two are too close and to keep both running. |
| Highest confidence we will print | 99% | Rounding to a hundred would claim a certainty no sample supports. |
| How far back a test reads | 90 days | Views are deleted after ninety days and sign-ups are kept, so a longer window would count sign-ups whose visits are gone and print a rate that is not merely uncertain but wrong. |
Honestly
What a test on PopzIQ will not do
Every one of these is a decision rather than a gap we are quiet about. If you need one of them, this is the page that tells you before you sign up rather than after.
Where this stops
It will not run three versions. Two, by design. The letters are A and B in the schema and the embed refuses a test that carries any other number of sides.
It will not test two different forms. A test splits the words on one form. Two forms with different questions produce two numbers you cannot compare, because you changed more than one thing.
It will not move your traffic for you. The split you set is the split that runs. Nothing promotes the winner, stops the loser or edges traffic across while the test is live. You read the sentence and you decide.
It will not test a hosted link. A test needs a page it can measure a view on. A form published as a link on its own refuses to carry one rather than reporting halves of a number.
It will not follow a person across sites. The key that keeps a visitor on one side is random, first-party and per form. It is not an identity, we never receive it and a visitor clearing their browser storage is simply a new visitor.
Write the second headline this afternoon
Publish a form, add a second version of its words and watch the two rates separate on the form's own page. It is on the free plan like everything else that captures.
No card to start. No trial clock. The free plan runs the whole loop.