These leaders all win on the same thing: product data. Dataïads is the AI-native Product Data Activation Platform that structures, enriches and activates your catalog everywhere, from Shopping to AI agents. Explore the platform
TL;DR

Ecommerce A/B testing splits traffic at random between two versions of a page, an ad or a product feed, then keeps the version that sells more. The tests that pay off fastest often sit before the click (feed titles and images, Catalog Ads) and on the page that receives paid traffic.

You pay for every Shopping, Performance Max and Meta click. Before you change a feed title, a Catalog Ad template or the page that receives that traffic, you want proof the new version sells more. A/B testing gives you that proof: two versions, comparable traffic, one decision.

This guide lists 30 ecommerce A/B test ideas, grouped by where they sit in the buying journey: feed and ads, post-click landing page, product page, category page and checkout. Each idea comes with the metric to watch and the tool that runs it. The last sections cover test length and the mistakes that break a result.

Ecommerce A/B testing in 30 seconds

A/B testing (also called split testing) is a controlled experiment: traffic is split at random between version A, the control, and version B, the variant. Only one thing changes between them. In ecommerce, you pick the winner on a sales metric chosen before launch: conversion rate, revenue per user (RPU), average order value or ROAS.

Two close cousins exist. An A/B/n test compares more than two versions of the same element. A multivariate test changes several elements at once (title and image, for example) and needs far more traffic, since every combination becomes its own arm. For most catalogs, a series of simple A/B tests teaches you more, faster.

An A/A test pits two identical versions against each other. Use it to check your setup: if two identical pages show a significant gap, your split or your tracking is broken.

Where to start: the step of the journey that carries the most revenue

Start where the most traffic and revenue flow, and where you can roll out the winner without an IT project. For a retailer buying Shopping and Meta traffic, the usual order is the feed and the ads (they touch all paid traffic), the page that receives that traffic, then the product page, category pages and checkout.

StepWhat it reachesExample testPrimary metric
Product feed and adsEvery Shopping, PMax and Catalog Ads impressionEnriched title vs original title on one categoryCTR, CPC, ROAS
Post-click landing pageAll paid traffic landing on the siteDedicated page vs original product pageBounce rate, conversion, RPU
Product pagePaid and organic trafficShipping and returns next to the buy buttonAdd-to-cart rate
Category pages and navigationVisitors who browseDefault sort orderProduct clicks, conversion
Cart and checkoutVisitors who have decidedShipping costs shown earlierAbandonment, conversion

Checkout reaches fewer visitors, but every point you win there is worth a lot. Baymard Institute, which averages 50 studies, puts the cart abandonment rate at 70.22%, and estimates that a large ecommerce site can lift conversion by 35.26% by fixing checkout usability (Baymard, updated October 28, 2026).

A/B test ideas before the click: product feed, Shopping and Catalog Ads

Before the click, you test what Google and Meta show of your products: titles, attributes, images and Catalog Ads templates. Google Ads supports this natively through Demand Gen experiments, where the product feed is a testable variable. Performance Max asset tests measure what text, images and video add on top of the feed alone. Meta supports it through the A/B test tool in Ads Manager. Test one category before you roll out to the full catalog.

Most lists of A/B test ideas stop at the website. For a catalog with thousands of SKUs, these pre-click tests affect every paid impression.

  1. Enriched title vs original title. “Brand + model” against “brand + model + the attribute shoppers care about” (material, use case, compatibility). Metrics: CTR and CPC.
  2. Completed attributes. Real color, material, size or a rebuilt product_type on one segment, against the source feed. Metrics: impressions and CTR.
  3. Main image. White-background packshot against a lifestyle image, within Merchant Center image rules. Metric: CTR.
  4. Catalog Ads template. Raw catalog against a template that shows price and discount in large type. Metrics: CTR, CPC, ROI.
  5. Performance Max assets. A feed-only asset group against the same group with added text, images and video. Metrics: conversions and conversion value.
  6. Standard Shopping vs Performance Max. Useful before you migrate a Shopping campaign that works. Metrics: ROAS and cost per conversion.
  7. Product video in ads. Static assets against videos built from the catalog, on Meta or Demand Gen. Metrics: CTR and cost per result.
TestNative toolHow it worksMetric to read
Feed A vs feed B (titles, images)Google Ads, Demand Gen experiment”Product feed” variable, 2 to 10 arms, 50% split recommendedConversions, CPA
Enriched assets vs feed onlyGoogle Ads, Performance Max asset testA feed-only asset group as control against the same group, enrichedConversions, value
Shopping vs PMaxGoogle Ads, Performance Max experimentAn existing Shopping campaign compared with a PMax campaignROAS, cost per conversion
Catalog Ads templateMeta Ads Manager, A/B testAudiences split with no overlap, one variable per testCTR, cost per result

Before its sale period, Faguo compared its standard Meta catalog with a Smart Creative template showing price and discount as a headline. Meta’s algorithm gave 66% of the budget to that variant, which delivered 86% more impressions, a 7% higher CTR, a 5% lower CPC and a 6.4% higher ROI (Faguo Smart Creative case study). At Castorama, feed optimization with Feed Enrich, our product feed optimization tool, was first tested on the Garden & Outdoor category, then rolled out to all 63,000 products: CPC down 4%, CTR up 10% (Castorama case study). Feu Vert Spain gained 32% ROAS on Google Shopping with Feed Enrich, measured on a 50/50 catalog A/B test (Feu Vert case study).

To produce the image and video variants you want to test, Smart Asset generates versions from your product images, which your team approves before they go live.

A/B test ideas for the Shopping and Meta landing page

Paid traffic usually lands on a product page built for a visitor who has already decided. Test that page, as the control, against a page built for paid traffic, as the variant: the same product at the top, more choice around it. Split traffic between the two, measure the gap in bounce rate, conversion and revenue per user, then send all traffic to the winner.

  1. Original product page vs dedicated landing page. The clicked product first, followed by similar products, variants and filters. Metrics: bounce rate and RPU.
  2. Ad promise repeated on the page. Same price, same discount, same wording as the ad. Metric: bounce rate.
  3. Merchandising. Similar products against category best sellers. Metrics: product clicks and average order value.
  4. Mobile layout. A lighter mobile template against an adapted desktop template. Metrics: mobile conversion and perceived speed.
  5. Meta page vs Shopping page. A visitor from a Catalog Ad is discovering the brand; a Shopping visitor is comparing. Test a separate page per channel. Metric: RPU by channel.

This test needs infrastructure that serves both versions without pulling in your IT team. In Smart Landing Page, traffic is split between a control and a challenger, the ROAS lift is measured, then 100% of traffic moves to the winning version. At Fnac, Smart Landing Pages tested against standard product pages in the Photo category raised revenue per user by 23.5%, conversion rate by 16.9% and add-to-cart by 9.2% (Fnac case study). Norauto Spain cut its bounce rate by 50% and raised conversion by 18% on a catalog of more than 100,000 products (Norauto case study). Faguo validated the approach through a progressive A/B test before rolling it out, with 6 mobile and desktop templates: bounce rate down 17%, revenue per user up 10.8% (Faguo case study).

For more on how to design these pages, read our guide to ecommerce landing page optimization.

Product page A/B test ideas

On the product page, test what helps a visitor decide: image order and type, title, review placement, price display and shipping information next to the buy button. Your primary metric is the add-to-cart rate. Check conversion as well, so you don’t keep a variant that only fills abandoned carts.

  1. Image order. Packshot first against a lifestyle or on-model photo first. Metric: add-to-cart rate.
  2. Product video. Gallery with a short video against photos only. Metrics: time on page and add-to-cart rate.
  3. Use-case title. Technical title against a title that says what the product is for. Metric: add-to-cart rate.
  4. Customer reviews. Star rating and review count under the title against a review block at the bottom. Metric: conversion.
  5. Compare-at price and savings. Strikethrough price alone against strikethrough price plus the amount saved. Metric: conversion, with an eye on margin.
  6. Shipping and returns next to the button. Delivery time, cost and return terms beside the buy button against the same information in a tab. Metric: add-to-cart rate.
  7. Variant picker. Visible color and size swatches against a dropdown. Metric: add-to-cart rate, on mobile above all.

Category page and navigation A/B test ideas

On category pages and internal search results, test product order, filters and grid density. These pages serve visitors who compare. Your primary metric is the click-through rate to a product page, then the conversion rate of sessions that go through the category page.

  1. Default sort. Relevance against best sellers against new arrivals. Metrics: product clicks and conversion.
  2. Filters. Key filters as chips above the grid against a side panel. Metric: filter use, then conversion.
  3. Mobile grid density. Two products per row against one. Metrics: product clicks and scroll depth.
  4. Badges. “Sale”, “New” or average rating on the thumbnail against a plain thumbnail. Metric: product clicks.
  5. Second image on hover. Back view or on-model photo on hover against a single image. Metric: product clicks on desktop.

Cart and checkout A/B test ideas

In the cart and checkout, test anything that removes a reason to leave: costs shown early, guest checkout, payment reassurance and the number of steps. According to Baymard, extra costs that are too high (shipping, tax, fees) are the top reason shoppers abandon (40% of those who leave for a reason other than browsing), ahead of slow delivery and forced account creation.

  1. Shipping costs shown on the product page or in the cart. Against costs revealed at the last step. Metric: abandonment rate.
  2. Guest checkout. Guest option up front against account creation offered first. Metric: checkout conversion.
  3. Number of steps. One-page checkout against a multi-step checkout. Metric: checkout conversion.
  4. Payment methods. Buy now, pay later or wallets (Apple Pay, Google Pay) shown from the cart against at the end. Metrics: conversion and average order value.
  5. Free shipping threshold. A progress bar toward the threshold against a plain mention. Metric: average order value.
  6. Reassurance. Payment logos, guarantee and return policy next to the pay button against in the footer. Metric: checkout conversion.

How long should an ecommerce A/B test run?

Run an ecommerce A/B test for at least one full week, to cover the gap between weekdays and weekends, and often two. Set the duration and the conversion volume you need before launch, and don’t read the result before the end. Meta recommends at least 7 days, and Google asks for at least 50 conversions per arm in a conversion-optimized Demand Gen experiment.

The ad platforms publish useful benchmarks:

  • Meta: you can schedule an A/B test for 1 to 30 days, and tests shorter than 7 days may be inconclusive. If your customers take more than 7 days to convert after seeing an ad, Meta suggests a longer test (Meta, A/B test best practices).
  • Google Ads, Demand Gen: at least 50 conversions per arm for CPA or conversion-based bidding, and a confidence threshold of 70% (default), 80% or 95% (Google Ads Help).
  • Google Ads, all experiments: one hypothesis, one variable, one or two success metrics chosen before launch, and no changes to the base campaign during the test unless sync is on (Google Ads Help).

Avoid testing during a sale or Black Friday if you want to apply the result to the rest of the year, since shoppers behave differently then. With low traffic, test a bigger change (a different page layout rather than a button color) or pool several categories into one test.

Mistakes that break an A/B test

Most A/B tests go wrong for one of three reasons: you read the result too early, the traffic split doesn’t match the plan, or you changed several things at once. Check all three before you call a winner. Expect most ideas to lose, too: Microsoft reports that fewer than a third of ideas evaluated in controlled experiments improve the metric they target, and even fewer at Bing (Bing blog, 2013).

  • Stopping as soon as the result looks good (peeking). Check the numbers every day and you will eventually see a “significant” gap caused by chance. Fix the duration and the stopping rule before launch.
  • Ignoring an SRM. A sample ratio mismatch is a gap between the observed split and the planned one, say 54/46 instead of 50/50. Microsoft researchers call it one of the most useful signals of a data quality problem (Fabijan et al., KDD 2019). If you see an SRM, find the cause (tracking, redirects, bots) before you interpret anything.
  • Changing several things at once. If the variant changes title, image and price, you won’t know which one mattered.
  • Editing the campaign mid-test. A budget, bid or asset change during the test makes the comparison meaningless.
  • Turning campaigns on and off by hand. Meta advises against it: audiences overlap and delivery becomes inefficient (Meta Help). Use the A/B test tool.
  • Optimizing for clicks instead of sales. A higher CTR doesn’t guarantee a better ROAS. Check conversion and revenue too.
  • Lacking sensitivity. On modest traffic, CUPED reduces variance by using pre-experiment data, so you can detect a smaller lift with the same sample (Deng et al., WSDM 2013).

Before you launch, run through this checklist:

  1. A written hypothesis and one primary metric.
  2. One change between A and B.
  3. Duration, target conversion volume and stopping rule set in advance.
  4. An A/A test or an SRM check over the first days.
  5. No campaign or page changes during the test.
  6. The result and the next hypothesis logged in a shared test journal.

Tools: what replaced Google Optimize

Google Optimize and Optimize 360 stopped working on September 30, 2023 (Google Analytics Help). To test an ecommerce business in 2026, you choose among three families of tools: website experimentation platforms, the native experiment tools of ad platforms, and tools built for the product feed and the post-click landing page.

FamilyExamplesWhat it testsLimit
Website experimentation platformsAB Tasty, Kameleoon, Optimizely, VWO, Shopify appsSite pages, journeys, personalizationDoesn’t touch ads or the feed
Ad platform toolsGoogle Ads Experiments, Meta A/B testCampaigns, bidding, audiences, assets, feed (Demand Gen)Measures on the ad side, not the on-site experience
Feed and post-clickDataïads (Feed Enrich, Smart Creative, Smart Landing Page)Feed titles and attributes, Catalog Ads, landing page for paid trafficFocused on Shopping and Meta traffic

Pick the tool based on the step you want to test. Many teams pair a website platform for checkout with the ad platform tools for campaigns. Dataïads covers the part in between: the feed that powers your ads and the page that receives the click.

FAQ

What is A/B testing?

A/B testing compares two versions of the same element (a page, an ad, a feed title) by splitting traffic between them at random. Only one thing changes between version A and version B. You roll out the version that wins on the metric you set before the test, such as conversion rate or revenue per user.

How do you run an A/B test on an ecommerce site?

Pick the step of the journey that carries the most revenue, write a hypothesis and a primary metric, then build a single variant. Split traffic 50/50 with a dedicated tool, run the test for at least one full week and until you reach the planned conversion volume, check the split, then roll out the winner and log what you learned.

What is an example of A/B testing in ecommerce?

Fnac compared its standard product pages with Smart Landing Pages on Google Shopping and Performance Max traffic for its Photo category. The landing pages raised revenue per user by 23.5% and conversion rate by 16.9%. Another example: Faguo tested two Meta Catalog Ads templates, and the one with price and discount as a headline delivered a 7% higher CTR.

Can you A/B test Google Ads and Performance Max campaigns?

Yes. Google Ads offers Performance Max experiments: measure the lift of adding PMax to your existing campaigns, compare a standard Shopping campaign with PMax, or test an optimization. Asset tests compare a feed-only asset group with the same group enriched. Demand Gen experiments test creatives, audiences, bidding and the product feed.

How do you A/B test product feed titles and images?

Build two versions of the feed for the same product segment, such as one category: original titles and images as the control, enriched titles or new images as the variant. Test them with a Demand Gen experiment (product feed variable) or by comparing two equivalent categories. Read CTR and CPC, then conversion, and roll out only after the test.

How long should an A/B test run?

At least one full week, often two, and until you reach the conversion volume you set at the start. Meta recommends at least 7 days and caps a test at 30 days. Google asks for at least 50 conversions per arm in a conversion-optimized Demand Gen experiment. Run longer if your customers take several days to buy.

What mistakes should you avoid in an A/B test?

Stopping the test as soon as a gap appears, testing several changes at once, editing the campaign mid-test, and drawing conclusions despite a traffic split that doesn’t match the plan (SRM). Don’t judge a test on CTR alone either: an extra click only counts if it turns into a sale.

What tool replaced Google Optimize?

Google Optimize shut down on September 30, 2023. For the website, teams use platforms such as AB Tasty, Kameleoon, Optimizely or VWO. For campaigns, Google Ads Experiments and the Meta Ads Manager A/B test are free and built in. For the product feed and the paid traffic landing page, Dataïads builds A/B testing into Feed Enrich and Smart Landing Page.

Test the page that receives your paid traffic

If your Shopping and Meta traffic lands on standard product pages today, that is often the test that pays back fastest. See how Smart Landing Page splits traffic between your current page and a dedicated landing page, then moves it to the version that sells more.

← Blog Free feed audit