Amazon A/B testing is a controlled comparison of two versions of a piece of listing content, such as the title or main image, in which Amazon randomly shows each customer one version for the entire test and then reports how likely it is that one of them sells better. The tool is called Manage Your Experiments, labelled Experimente verwalten in the German Seller Central interface, and it is available to brands enrolled in Amazon Brand Registry. Because both versions run at the same time, seasonality, prices and advertising affect both groups equally. We checked all rules in this article against Amazon's official sources on 23 October 2026.
Key Takeaways
- Testable: Title, main image and secondary images, bullet points, description, A+ Content and Brand Story (Source: Amazon, sell.amazon.com/tools/manage-your-experiments).
- Requirements: A brand in Brand Registry and an ASIN with enough traffic in recent weeks. Amazon does not name a fixed threshold (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G9FCELUZGHBD9C7H).
- Duration: For a set duration, Amazon recommends eight weeks. With the to significance option, a test can end after as little as four weeks (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ).
- Results: Amazon shows weekly how likely it is that one version performs better (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/GY6DCFA53HYDH79H).
- Publishing: Automatic from a 66 percent probability, the option is preselected at setup; if you turn it off, Version A stays live (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ).
How does Amazon A/B testing with Manage Your Experiments work?
Manage Your Experiments randomly splits the customers who see your content into two groups: one sees your current content as Version A, the other your new Version B. Assignment is tied to the customer account, and each customer sees the same version for the whole run, on every device. Visits without an identifiable customer account are not counted in the sample (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/GY6DCFA53HYDH79H).
Amazon does not state a fixed split such as 50/50. For tests across several ASINs, for example with A+ Content, Amazon aggregates the results, so they apply to the group, not to each individual ASIN. The analysis uses a Bayesian approach: you do not get a yes or no, but a probability that Amazon updates weekly during the test (same source).
What can you A/B test on Amazon?
According to Amazon, you can test product titles, the main image and secondary images, bullet points, the product description and A+ Content including Brand Story (Source: Amazon, sell.amazon.com/tools/manage-your-experiments).
| Element | What you can vary | Rule according to Amazon |
|---|---|---|
| Product title | Length, order, brand name | No more than 75 characters including spaces |
| Main image | Angle, crop, package contents | Same image guidelines as the listing, such as a white background where required |
| Secondary images | Order, infographics, lifestyle images | Only for ASINs with four or fewer child ASINs, not in multi-variate tests |
| Bullet points | Benefits versus features, order | Up to five, ideally under 1,000 characters in total |
| Product description | Structure, dimensions, use | Same guidelines as regular content |
| A+ Content and Brand Story | Modules, order, comparison chart | Both versions approved and applied to the same ASINs |
(Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ and G9FCELUZGHBD9C7H)
A multi-variate experiment changes several elements at once, for example title, image and A+ modules (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G2BNMJ8NFCGZQVFH). It only shows whether the package works, though, not which element made the difference.
Customers already see the title and main image in search results, which is why we usually start there. Which image types belong in a gallery is covered in our article on Amazon product images, and Creating Amazon A+ Content explains how to build A+ modules.
Which requirements apply to Manage Your Experiments?
You need a brand in Amazon Brand Registry and an ASIN of that brand with enough traffic. Amazon decides whether the ASIN is eligible.
- Marketplace: According to Amazon, the tool is available in 13 countries, including Germany (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G9FCELUZGHBD9C7H).
- Brand: You are enrolled in Brand Registry and responsible for selling the brand in the Amazon store (same source).
- Traffic: Amazon does not give a number. ASINs with low traffic may not appear in the selection at all, and for variations the system shows the eligibility status along with reasons (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ).
- Content: The normal guidelines apply, and for A+ and Brand Story, Version A must be approved. For child ASINs, you also submit content for the parent ASIN (same source).
- Cost: Amazon describes the tool as free (Source: Amazon, sell.amazon.com/blog/manage-your-experiments).
How does an experiment work step by step?
You create an experiment in five steps in the Manage Your Experiments dashboard, which you reach in Seller Central via the Brands menu:
- Write down a hypothesis: for example, as in Amazon's own example, that a descriptive rather than a vague title drives more sales.
- Create the experiment: Click “Create a new experiment”, choose the type and the reference ASIN, meaning the primary product of the test.
- Create Version B: You enter texts and images directly in the experiment, A+ and Brand Story via “Start by duplicating Version A” in the A+ Content Manager.
- Set the details: Name, hypothesis, duration and auto-publish. Drafts are saved for up to two weeks.
- Submit and check the status: Amazon validates the content, so the start can be several days later. If validation fails, correct the content and resubmit.
(Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ)
Once the test is running, only the hypothesis and duration can be changed. For new content, you cancel and start over (same source).
Quick setup for titles: If you change the title of an eligible ASIN on the inventory page, Amazon creates a title experiment by default according to its FAQ, and the winner is published automatically. You can opt out via a checkbox (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G9FCELUZGHBD9C7H and YZYA77YZTWTF7YZ). So check the dashboard for tests you did not set up on purpose.
How long should an Amazon A/B test run?
For experiments with a set duration, Amazon recommends eight weeks, and eight to ten weeks are typical. With to significance, a test ends automatically as soon as the results are conclusive, in some cases after just four weeks (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ).
| Option | When it fits | What Amazon says |
|---|---|---|
| To significance | Titles, images or bullet points, required duration unclear | Duration depends on the page views of the ASIN |
| Set duration | A+ Content, Brand Story or a test that should end before a sales event | Eight weeks recommended, can be changed at any time |
(Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ and G9FCELUZGHBD9C7H)
- Do not stop early: According to Amazon, stopping because of promising interim results risks overestimating the effect or choosing the wrong version (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G2BNMJ8NFCGZQVFH).
- Plan around sales events: If Prime Day or Black Friday fall within the run, event behaviour mixes into the result. Schedule the start so the test ends beforehand.
How do you evaluate the results of an Amazon experiment?
What matters is the probability that one version is better and the worst case of the one-year projection. Amazon explains the probability like this: a 75 percent probability for Version A means that 75 percent of the possible impacts it calculated point to a positive sales effect from publishing Version A (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/GY6DCFA53HYDH79H).
| Metric | Meaning according to Amazon |
|---|---|
| Probability | Share of the calculated impacts that show more sales for this version |
| Units and sales | Units sold and sales per version |
| Conversion rate | Share of visitors in the experiment who made a purchase |
| Units per unique visitor | Units divided by the sample size |
| Sample size | Number of unique visitors per version |
| Projected one-year impact | Average daily sales increase of the winning version times 365, only for completed experiments |
(Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/GY6DCFA53HYDH79H)
The one-year projection shows the median in the Likely column and the bounds of the 95 percent interval in the best case and worst case columns. Amazon does not factor in seasonality or price changes (same source). Use it as an order of magnitude. If the worst case is negative, the losing version may still turn out better over time.
This is how we decide:
- Low-cost changes such as titles or bullet points: Amazon's auto-publish threshold of at least 66 percent probability is enough for us (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/YZYA77YZTWTF7YZ).
- Costly follow-up decisions such as a new photo shoot for an entire product line: we require a clearly higher probability and a worst case close to zero, otherwise we repeat the test on a second ASIN.
Without auto-publish, you publish the winner yourself: texts and images through the standard listing tools, A+ Content through the A+ Content Manager (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/GY6DCFA53HYDH79H).
No clear winner? Amazon lists possible reasons: the change was too small, there was not enough traffic, both versions work similarly well, or the change concerns something customers do not care about when buying (same source). That is a result too. Next, test a bigger difference.
Which hypotheses are good for getting started?
Good hypotheses change one element clearly and concern something customers need for their purchase decision. Among other things, Amazon suggests these approaches in its tips (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G2BNMJ8NFCGZQVFH):
| Element | Version B | Underlying assumption |
|---|---|---|
| Title | Add the brand name or make the title much more descriptive | Customers see faster whether the product fits |
| Main image | High-resolution image that stands out from competitors | The product gets noticed in search results |
| Secondary images | Only lifestyle images versus only product images | Customers understand the use faster |
| Bullet points | Put a different set of features first | Customers find the buying argument that matters to them |
| A+ Content | Comparison chart or new modules | Customers pick the right variation more easily |
Phrase every hypothesis as an if-then sentence with a reason, for example: if the pack size appears at the front of the title, units per unique visitor increase because customers compare pack sizes. What belongs in your title and bullet points before you test is covered in our listing optimization guide.
Which mistakes make an Amazon A/B test worthless?
Typical mistakes are versions that are too similar, an unstable environment and insights transferred without checking.
- Versions too similar: Amazon warns you during setup if titles or images are too similar, because a conclusive result becomes less likely (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G9FCELUZGHBD9C7H).
- Unstable environment: Price changes, coupons and new ad budgets affect both groups, but they shift the context in which the result applies. Keep them stable or log them with a date. If the item is out of stock, you collect hardly any usable data.
- Different ASINs for A+: Both versions must be applied to exactly the same ASINs, otherwise the test cannot be created (Source: Amazon Seller Central Help, sellercentral.amazon.de/help/hub/reference/external/G2BNMJ8NFCGZQVFH).
- Transferring results unchecked: What works for one product need not work for another. Confirm it there with a second test.
We are The Marketplace Guys, an Amazon Ads Partner from Bielefeld, founded in 2022 by Alexander Schnelle and Bhavesh Tailor. We have scaled more than 15 consumer brands and manage over 45 million euros in marketplace revenue. We plan listing tests as an Amazon SEO agency together with the listing copy, and we build new A+ versions as an Amazon A+ Content agency. If you want to set up your first test together, get in touch.
Frequently Asked Questions
What is Amazon A/B testing with Manage Your Experiments?
Manage Your Experiments is Amazon's tool for A/B testing listing content in Seller Central. Amazon randomly shows each customer either Version A or Version B for the entire run and reports at the end how likely it is that one version sells better. It is available to brands enrolled in Amazon Brand Registry.
What can you A/B test on Amazon?
According to Amazon, you can test product titles, the main image and secondary images, bullet points, the product description and A+ Content including Brand Story. A multi-variate experiment changes several elements but only tells you whether the package works. For clear insights, change one element per test and change it clearly, such as a much shorter title.
How long should an Amazon A/B test run?
For experiments with a set duration, Amazon's help pages recommend eight weeks, and typical runs are eight to ten. With the to significance option, a title, image or bullet point test ends automatically once the data is conclusive, sometimes after just four weeks. Do not stop early because of promising interim results, as early results can mislead.
What are the requirements for Amazon experiments?
You need the Professional selling plan (according to sell.amazon.com), your brand must be enrolled in Amazon Brand Registry and you must be responsible for selling it in the Amazon store. The ASIN must belong to your brand and have received enough traffic in recent weeks. Amazon does not publish a fixed threshold. The Manage Your Experiments dashboard shows you whether an ASIN is eligible.
When does Amazon publish the winning version automatically?
Yes, if auto-publish is active (Amazon preselects it at setup, you can turn it off at setup or during the test) and the winning version has at least a 66 percent probability of performing better, according to Amazon's help pages. The feature is available for titles, images and bullet points. Without it, all customers see Version A again after the test until you publish the winner yourself.