Start free
Guides · AI for your store

AI A/B testing that lets real orders choose your best page

Dropify splits your visitors between versions of the same page on one link, counts the orders each version brings and keeps the winner. AI can write the new version for you, from a new headline to a full new page, and run the next test when the last one ends.

Start freeSee pricing

Updated: 2026-11-21

What an A/B test does for a cash-on-delivery seller

You pay for every visitor your ads bring. If one version of your page turns 3 visitors in 100 into orders and another turns 4, the second one gives you a third more orders for the same ad budget. You only know which one it is by testing both on real traffic at the same time.

On Dropify you create an A/B test on a landing page, add up to 6 versions and share one public link. Every visitor who opens it is sent to one version according to the traffic shares you set, and keeps seeing that same version. Point all your Meta, TikTok, Snapchat and Google ads at that single link: you never have to duplicate campaigns to compare pages.

Start from a ready test template

  • Classic 50 / 50: split traffic evenly between two versions.
  • Champion / Challenger: keep 80% of visitors on your proven page and test the new one on 20%.
  • By country: visitors from the countries you pick always see a dedicated page, for example a Gulf version and a Morocco version.
  • By language: serve a different page per browser language, such as Arabic and French.
  • Mobile / Desktop: show mobile visitors a page built for small screens.
  • Blank canvas: build the test yourself, with audience rules on country, language or device for any version.

Pick the metric that pays you, not just clicks

Each test is decided on one metric you choose: leads (submitted orders, the fastest signal), confirmed orders, delivered orders (the honest COD signal, slower but it is the money you bank) or revenue when versions sell at different prices.

For automatic decisions Dropify counts clean leads and lead value. Duplicate orders, phones on your blacklist and the same phone ordering the same product again within 24 hours never count, so a version cannot win on junk orders. Orders paid online count once they are paid, and refunds bring the lead value down.

Let AI write the challenger

You do not need a designer to try a new idea. In a test, choose Create variant with AI and describe in one sentence what the version should try, for example "a darker, urgency-driven version focused on the discount, with customer reviews near the top". The AI copies your current page and rebuilds it from your brief, with new images and video if you want them. The new page stays unpublished until you review it, and you can change anything in the drag-and-drop builder before launch.

AI work is paid in coins (100 coins = $1). The cost is shown before the build starts: coins are reserved, you are charged what the build really costs and the rest goes back to your balance.

Optimize a whole page, one test after another

Instead of planning tests yourself, you can ask Dropify to optimize a page. It lines up a queue of tests and runs them one by one, keeping each winner before the next starts. Before you start, you see how many tests, how many coins and how many days to expect.

  • Commercial tests: price, compare-at price, bundle offer, bundle tier prices and upsell.
  • AI content tests: a new headline, a new marketing angle, a new main image, a new angle with its image, or a complete new page.
  • Each finished step shows its result in plain words, such as "Version B won" with its lift in clean leads, or "No clear winner: original kept".

How the winner is decided

Dropify computes each version's chance to be the best on your metric and shows how many days a reliable verdict should take at your current traffic. When a version reaches about 90% chance to win, you get a notification to review it and promote it.

Automatic decisions wait for the minimums you set. By default a test runs at least 7 days, with at least 200 visitors and 20 clean leads per version. A challenger with a 95% chance to be best and more clean leads per visitor is promoted; one with 5% or less chance to beat the original is rolled back; after 28 days without a clear answer the test ends and the original page stays. You can change every one of these numbers.

  1. Create the test, pick a template and the deciding metric.
  2. Add versions: duplicate your page and edit it, pick an existing page, or generate one with AI.
  3. Run the launch checks: at least 2 versions, one control, every page published, and a warning if versions differ in more than one way.
  4. Launch, point your ads at the public link and follow visitors, add to cart, leads, confirmed, delivered and revenue per version, with segments by country and device.
  5. Promote the winner: the link now serves it permanently, the other pages stay in your library and the report is kept.

Guardrails that protect your margin

  • Price tests stay within a maximum price change, 15% by default, and never go below your product cost.
  • A challenger gets at most 50% of traffic by default, so most visitors keep seeing your proven page.
  • Bundle tiers must stay below their compare-at price; pages whose options have their own prices are not price-tested.
  • A promoted price, bundle or upsell can be undone within 7 days, back to the exact previous values.
  • Pause any test: while paused, visitors see the original or the winner. Every decision is kept in the Lab history.

Suggest mode or automatic mode

In suggest mode, Dropify recommends and you decide. In automatic mode, AI plans and launches the next tests on the pages you choose (3 by default), inside a monthly coin cap you must set. You decide what needs your approval: by default new text and images launch on their own, and full new pages wait for you ("Ask me first"). Notifications arrive in the dashboard, by email and, if you connect it, on Telegram.

Example: a Casablanca beauty store sends all its TikTok traffic to one serum page. It lets AI test a new marketing angle on 20% of visitors, decided on delivered orders, while the owner keeps the 15% price limit. When the test ends, the next step starts by itself and the owner only reads the result.

FAQ

What is AI A/B testing for a landing page?

It splits visitors between versions of the same page on one link, and AI writes the new versions and helps decide the winner from real orders.

How many versions can I test at once?

Up to 6 versions per A/B test, all served on the same public link.

Can I test my product price?

Yes. Price, compare-at price, bundle offers, bundle prices and upsells can be tested, within a maximum price change you set (15% by default) and never below cost.

How long does a test take?

By default at least 7 days and enough visitors and clean leads per version; the report shows an estimate of the days left at your current traffic. Tests end after 28 days without a clear winner.

Does AI change my page without asking?

Only if you allow it. In suggest mode you decide everything; in automatic mode you choose which changes launch on their own and which wait for your approval.

What does AI A/B testing cost?

Tests you build yourself, including price and offer tests, use no coins. AI-written versions are paid in coins, with the cost shown before each build, and automatic mode never spends more than your monthly coin cap.

Can I split test by country or device?

Yes. Any version can be shown to visitors by country, browser language or device, and the report breaks results down by segment.

Let your next page win on real orders

Build and preview your store for free, start selling with a $10 top-up, then launch your first A/B test and let AI write the challenger.

Start testing with AI