A/B page tests
Show two versions of a page at the same address, each to half of your visitors, and see which one books more patients.
Ads → A/B tests compares two versions of one page of your website. Visitors to the page are split 50/50: half see your current page (version A), half see version B, at the same address and with no flicker. Smileline then counts which version leads to more bookings — not leads, because a cheap enquiry that never books isn't a win. You pick the winner; nothing changes on its own, and a test runs until you stop it.
A/B tests come with every plan, and work on websites built with Astro through the Smileline Astro integration. The online booking page and sites built with anything else can't be tested yet.
Who can do what
| To do this | You need | Who has it |
|---|---|---|
| See tests and their results | View ads, captured leads & results | Owners, managers, front desk and analysts |
| Start, stop and archive tests | Connect providers, manage lead forms & draft campaigns | Owners and managers |
Before you start
- Install the Astro integration with its server key. Until it has checked in, the page warns that The Smileline Astro integration hasn't checked in yet, and everyone sees version A.
- Build version B as another page of the same site, for example
/invisalign-bbeside/invisalign. It can differ in anything: headline, offer, photos or layout. - Make both pages render on demand: your project needs a server adapter, and each page needs
export const prerender = false. A prerendered page can't be split. - Don't cache either page at your CDN. The integration sends tested pages as private, uncached responses; a caching rule of your own would show every visitor the same version.
Start a test
Open Ads → A/B tests and click New test.
Give the test a name, such as "Invisalign page: new headline".
Enter the Page to test (version A) — the path visitors open today, such as /invisalign — and the Version B page, such as /invisalign-b. Use paths only: no domain, no ? or #, and nothing under /_sl/, which belongs to the integration.
Click Start test. Within about a minute, half of the page's visitors see version B at the address of version A.
A page can be in one test at a time: a path used by a test that isn't archived yet, as either version, can't start another.
How the split works
- Each visitor gets a version the first time they open the page, and keeps it through Smileline's visitor cookie on every later visit.
- A visitor whose browser hasn't stored the cookie — because of your consent setting, their choice, or blocked cookies — still sees one of the two versions, possibly a different one each time, but isn't counted.
- The address in the browser stays version A's. Search engines are pointed at version A, and version B's own address is marked not to be indexed if someone opens it directly.
Read the results
Each test is a card with its status (Running, Using B, Kept A, Stopped or Archived) and one line on where it stands:
- Too early to call, with how many counted visitors have booked so far out of the 10 needed before Smileline suggests a winner.
- Version B is ahead, but it's too close to call: enough bookings, but not yet a clear winner.
- Version B books more patients: one version has at least a 90% chance of being the better page, and the matching button is highlighted.
The table under it shows, for each version:
| Column | What it means |
|---|---|
| Visitors | Counted visitors who saw this version. |
| Booked | Of those, the visitors who booked (see below). |
| Booking rate | Booked ÷ visitors. |
| Chance to be best | How likely this version is the one with the higher booking rate, given the bookings so far. |
| Coverage | How many of this version's visits are counted. |
A booking counts for a version when the visitor, after first seeing it, made an enquiry from your website, and that enquiry led to a booking made within 30 days of the first view. Each visitor counts once, however many times they book. The counts cover the last 13 months of visitors, the time visitor detail is kept.
Coverage can differ between versions: someone who gives cookie consent after seeing a page they like is counted, someone who leaves isn't. If one version's coverage is far below the other's, read its booking rate with care.
Page A isn't splitting appears when views of the tested page arrive without a version: usually the page is prerendered, or the integration isn't installed with the server key.
Pick a winner
When you're ready, use one of the buttons on the card:
- Use B serves version B to everyone at version A's address, until you archive the test.
- Keep A shows everyone the original page again.
- Stop without a winner does the same, without recording a winner.
Each asks you to confirm. Results stay Provisional after a test stops, because bookings from visitors who saw a version before it stopped can still arrive. They become Final once all of these are true:
- every counted visitor's 30-day booking window has ended,
- the website events received before the stop have finished processing (allow about two weeks),
- nothing that could change the counts is still being retried automatically.
Until then the numbers can still change, usually only slightly. Enquiries waiting in your Review queue don't hold results back: if you approve one after the results are final, it doesn't change them.
Archive a test
Archive a stopped test once its result is in place, so both pages are free for new tests. Archived tests stay at the bottom of the page under Archived tests.
After Use B, archiving stops the integration serving version B. Put version B's content on version A's page first, then archive.
Session replays
Watch anonymous recordings of real visits to your website, page by page, with everything typed into a form hidden.
Ad campaigns
Build a Meta or Google Search campaign from a ready-made template in four steps, set it live deliberately, and manage campaigns imported from every connected ads platform.