Experiments (Site-wide A/B Testing)
Experiments let you A/B test changes to your existing site — swap headline text, change a color, hide an element, or inject custom CSS — without editing your site's code. Turnjo splits traffic between variants, tracks a conversion goal, and can conclude the test automatically when a variant wins.
Experiments are different from Popup A/B tests, which test variants of a Turnjo widget. Experiments change your own page elements.
How it works
- Go to Experiments and click New experiment. Give it a name, choose a type (DOM changes to edit the page visually, or Split URL to redirect to different pages) and the page it runs on. Traffic, statistics and audience are under Advanced; the defaults suit most tests.
- Turnjo creates two variants: a Control (your site as-is) and Variant B, each at 50% weight.
- Make your changes to Variant B and add a goal. An experiment can't start without a goal, or (for DOM tests) without at least one change on a variant.
- Click Start at the top of the experiment's page. Settings save as you go.
Visitors are assigned a variant at random according to the weights, and the assignment is sticky — a visitor keeps the same variant across visits (stored in their browser). A small blocking snippet hides the page until the variant's changes are applied, so visitors never see the original content flash first.
Split URL tests
A Split URL experiment sends a share of visitors to entirely different pages instead of editing the current one — perfect for comparing two separately-built landing pages.
- The entry page (set under "Redirects visitors arriving on pages where") is the original page visitors arrive on (the Control).
- Each non-control variant has a Destination URL. A visitor assigned to that variant is redirected there (and stays there on future visits, since the assignment is sticky).
- Set a goal — usually a Page visit goal on a shared success URL (e.g.
/thank-you). Because the assignment follows the visitor, a conversion on the destination or a later page is still credited to their variant.
Add a destination to every variant before starting — Turnjo won't start a split-URL test with a missing destination.
Variant changes
(DOM-change experiments only.)
Each non-control variant holds a list of changes. Every change targets an element by CSS selector (except Custom CSS):
| Type | What it does |
|---|---|
| Text | Replaces the element's text |
| CSS | Sets a single CSS property (e.g. background-color) on the element |
| Visibility | Shows or hides the element |
| Image | Swaps an image's source (must be an https:// URL) |
| Custom CSS | Injects a block of CSS into the page |
A visual editor is available from the variant so you can point at elements on your own site.
Targeting and audience
- Runs on pages where — restrict the experiment to matching pages. Pick a match type — matches (exact), starts with, contains, ends with, matches pattern (
*wildcards), or matches regex — and enter the value. Leave the value blank to run on every page. For the homepage, choose matches and enter/. - Traffic — enrol only a percentage of eligible visitors (1–100%).
- Audience — layer on targeting rules: device type, visitor type, URL contains, time on page, scroll depth, visit count, exit intent, and location (country / region / ZIP radius). Rules of the same kind match if any one does; different kinds must all match. Audience targeting requires a plan that includes it.
Goals and conversions
Set a goal that defines a conversion. The first goal you add is the primary goal used to decide a winner. Goal types:
- Click — a visitor clicks an element matching a selector.
- Page — a visitor reaches a matching URL (e.g. a thank-you page), using the same match types as the experiment's page targeting.
- Custom event — your site dispatches a
turnjo:goalevent with a matching name.
Each visitor counts once per goal, so repeated clicks or page views aren't double-counted.
Automatic conclusion
Turnjo checks running experiments every 15 minutes and can conclude them for you:
- A winner is found — a variant beats the control with statistical significance at your chosen confidence level (80/90/95/99%), once minimum sample size is reached (and at least 100 visitors on each side) and the improvement clears your minimum improvement (default 20%).
- Time runs out — the experiment passes its maximum duration with no clear winner.
Because results are checked so often, the bar a variant must clear starts high and falls as the experiment nears its maximum duration, and it's higher for each variant when there are several. This keeps the chance of a false winner at your confidence level however often results are checked.
You can also pause or conclude an experiment manually at any time. When an experiment concludes, team members who've opted in are emailed.
Tuning the confidence level and minimum sample size requires a plan that includes advanced experiment settings; otherwise confidence defaults to 95%.
Results
The Results page shows each variant's enrollments, conversions, conversion rate, uplift versus control with a likely range (for example "+18% (likely −4% to +40%)"), the z-score and the z-score needed right now, and an estimated completion date. It warns you if the split between variants is further from their weights than chance would explain, which usually means something other than the experiment (a redirect, a cached page, bots) is deciding who's counted.
Plan limits
Your plan caps how many experiments can run at once and how many variants each can have. See Billing & Plans.
More docs
Getting Started
Popups
Surveys
Guided Tours
Experiments
Visitor Feedback
Integrations
Team & Account
Billing & Plans
Privacy & Compliance