Device Targeting

Device targeting limits an experiment to some device classes — desktop, mobile or tablet. It’s a per-experiment setting, not per-variant, so every variant in the experiment shares the same device filter.

How device detection works

The Split Test Pro script classifies the visitor’s device in their browser when the page loads, in this order:

• Mobile — the browser’s user agent identifies a phone (iPhone, iPod, an Android phone, Windows Phone, BlackBerry, Opera Mini and similar).
• Tablet — the user agent identifies a tablet (an iPad, an Android device without “Mobile” in its user agent, a Windows on ARM device), or the browser window is between 600 and 1024 pixels wide.
• Desktop — everything else.

Phones are recognised first, so a phone is always Mobile whatever its screen size. The window-width check means a desktop browser that happens to be 600–1024 pixels wide when the page loads counts as Tablet.

When to use device targeting

Use it when:

• The change you’re testing is only visible on one device class — a sticky mobile add-to-cart bar shouldn’t be tested on desktop, where it doesn’t render.
• The conversion behavior differs structurally by device — for example, mobile visitors go through a different checkout flow.
• You want clean results for one device class without mobile and desktop conversion rates being averaged together.

When not to use device targeting

Don’t use it when:

• The change is purely visual and fits in a @media query. Use the media query inside the variant instead.
• You want to see device breakdowns for an experiment. The Devices filter on Results shows them regardless of targeting — see Segmenting Results.
• You want different variants per device. Device targeting filters which devices run the experiment; it doesn’t show different variants per device. Run a separate experiment for each device class instead.

Setting it up

On the Targeting step of the experiment wizard, the second question is 02 Which devices? It shows three cards: Desktop, Mobile and Tablet. All three are on by default.

1. Tap a card to turn it off or on.
2. That’s it — each tap is saved straight away.

The cards combine with or: Mobile + Tablet runs on phones and tablets, but not desktops.

At least one card must stay on. If you turn all three off, nothing is saved and the launch check at least one device blocks Launch Test until you turn one back on. An experiment that never had its devices configured runs on every device, the same as all three cards on.

You can change devices while the test is a Draft or Paused; pause a running test to change targeting.

Device targeting + URL targeting

Both filters apply: a visitor must match the experiment’s page rules AND its devices to be assigned a variant. Within the page rules, the and / or joins between rows decide which pages qualify — see URL Targeting. Any audience rules must pass as well.

Page rule: Path starts with /products/  +  Devices: Mobile only
                                                  ↓
   = Visitor on /products/widget on a phone → assigned a variant
   = Visitor on /products/widget on desktop → not assigned
   = Visitor on /collections/featured on a phone → not assigned (page doesn't match)

Reading results by device

Even on an experiment that runs on every device, the Devices filter on the Results tab narrows the report to Desktop, Mobile or Tablet. Use it when:

• The experiment is broadly targeted and you want to see whether mobile and desktop visitors responded differently.
• You suspect a regression on one device class even though the overall numbers look fine.

See Segmenting Results for the full filter bar.

Common mistakes

• Targeting only mobile when the experiment doesn’t need it. You’ll collect data more slowly than necessary. Default to all devices and use the Devices filter on Results.
• Confusing device targeting with responsive design. Device targeting decides who runs the experiment. A @media query inside a CSS variant decides how the variant looks at different sizes. They’re complementary, not interchangeable.
• Expecting a narrow desktop window to count as desktop. A desktop browser 600–1024 pixels wide at page load is classified as Tablet.
• Trying to target by exact device model. Split Test Pro classifies into three buckets only. iPhone vs Pixel vs Galaxy is not a distinction it makes. Use a JS variant for finer-grained branching.

What if my visitor’s user agent is faked?

Bots, headless browsers and developer tools can set any user agent. The classifier uses what the browser reports; anything it can’t place as a phone or tablet counts as desktop. Split Test Pro doesn’t strip bot sessions from results, so use your analytics tool for rigorous bot filtering.

Next steps

• Combine device targeting with the right page rules: URL Targeting.
• Choose which visitors are eligible: Audience Targeting.
• See how a single experiment performed across devices: Segmenting Results.
• Test a mobile-only change without filtering at the experiment level: CSS Variants (use @media).