Every seller has an opinion about which main image is better. The founder likes the angled shot, the designer prefers the straight-on view, and the agency swears by the pack shot with the box behind the product. Opinions are cheap, and on Amazon they can be expensive, because the wrong hero image quietly costs clicks on every search results page for months. The only reliable way to settle the argument is to let real shoppers decide.
Amazon gives brand owners a tool for exactly this: Manage Your Experiments, often shortened to MYE. This guide explains how to use it for images specifically, how to design a test that produces a clear answer, how long to run it, how to read the results without fooling yourself, and what to do if your listing is not eligible. If you want to understand the metrics that sit behind a test, our guide to how listing images drive click-through and conversion covers them in depth.
What Manage Your Experiments Actually Does
Manage Your Experiments is a split-testing tool inside Seller Central (under the Brands menu) that shows two versions of a piece of listing content to different shoppers at the same time. Amazon randomly assigns each shopper to version A or version B for the duration of the experiment, then compares how the two groups behave. Because both versions run simultaneously, seasonality, promotions and advertising changes affect both groups equally, which is what makes the comparison fair.
The tool supports several content types, and the selection has expanded over time. Main images, titles, A+ content and bullet points are among the options sellers commonly see. For image work, the main image experiment is the one that matters most, because the main image is the only picture most shoppers see before they decide whether to click.
At the end of an experiment, Amazon reports metrics such as units sold, sales, conversion rate and units sold per unique visitor for each version, along with a probability that one version is better than the other. It may also give an estimate of the projected impact over a year if you publish the winner.
Eligibility: Who Can Run an Image Experiment
Not every listing qualifies. The general requirements are:
- Brand Registry. You must be the brand owner, enrolled in Amazon Brand Registry, with the ASIN listed under that brand.
- Enough traffic. The ASIN must receive sufficient traffic for a test to reach a conclusion in a reasonable time. Amazon does not always publish the exact threshold, and it can change, so check the eligible ASIN list inside the tool.
- No conflicting experiment. You usually cannot run two experiments of the same content type on the same ASIN at once.
If an ASIN is missing from the eligible list, the most common reason is low traffic. Advertising can help raise sessions, but there is no guarantee an ASIN will become eligible. Requirements vary by marketplace and are updated, so always confirm the current rules in Seller Central rather than relying on a third-party summary.
Start With a Hypothesis, Not a Hunch
The biggest mistake sellers make with MYE is testing two images that are almost identical. If version B is version A with a slightly warmer colour grade, the test will very likely run its full length and end inconclusive. A good test starts with a written hypothesis that names one meaningful change and the reason you expect it to work.
A useful format is: "If we change X, shoppers will Y, because Z." For example:
- If we show the product at a three-quarter angle instead of straight on, more shoppers will click, because the depth of the drawer will be visible in the thumbnail.
- If we show the full 3-pack instead of a single unit, conversion will rise, because reviews say buyers were unsure how many they would receive.
- If we remove the retail box from the main image, click-through will rise, because the product will fill more of the frame at thumbnail size.
Writing the hypothesis down forces you to test something with a real chance of changing behaviour, and it gives you a lesson to keep regardless of the result. Our guide to creating a main image that wins the click is a good source of hypotheses worth testing.
What to Test First: A Priority List for Images
Test the changes most likely to move the needle before the subtle ones. A sensible order for main images is:
- Angle and orientation. Front, three-quarter, top-down or open versus closed. These change how the product reads at small sizes.
- Product versus product with packaging. Packaging can build trust for gifts and premium items, or it can shrink the product and waste space.
- Single unit versus multipack or full set. Especially useful when the listing sells a quantity that is not obvious.
- Colour variant shown. For parent listings, the default child colour on the main image can affect clicks.
- Fill and crop. How tightly the product fills the frame, within the rule that it should occupy roughly 85% of the image.
Both versions must comply with main image rules: a pure white background (RGB 255, 255, 255), no text, logos, badges or borders, and only the product being sold. MYE does not exempt a test image from policy, and a non-compliant version can be rejected or cause problems on the live listing.
Setting Up an Image Experiment Step by Step
- Open Manage Your Experiments from the Brands menu in Seller Central and choose to create a new experiment.
- Select the content type, such as main image, and pick the eligible ASIN.
- Name the experiment clearly, for example "Lunchbox main: straight vs 3/4 angle, Oct", so you can find it in your history later.
- Enter your hypothesis in the field provided. It is optional, but it becomes useful documentation.
- Version A is your current live image. Upload version B, which must meet all technical requirements, including at least 1000 pixels on the longest side so zoom works.
- Choose a duration or accept the duration Amazon recommends.
- Submit. The new image goes through a review, and the experiment starts once approved, usually on a scheduled date.
Do not edit the listing's other content in a way that could interfere while the experiment runs. Changing the title or price mid-test does affect both groups equally, but it can change the context in which the image is judged and make the lesson harder to apply.
How Long to Run an Image Test
Experiments typically run for several weeks. Amazon allows you to choose a set duration, commonly in a range of roughly four to ten weeks, and it can recommend a duration to reach a conclusive result based on the ASIN's traffic. The available options change from time to time, so check them when you set up the test.
Some practical guidance:
- Use the recommended duration where offered. It is based on your traffic and the effort needed to reach confidence.
- Cover full weeks. Shopping behaviour differs between weekdays and weekends, so durations in whole weeks avoid bias.
- Avoid ending early. Early results swing about. A version that looks ahead after five days can easily lose by week four.
- Be careful around big events. Prime Day and peak season attract different shoppers. A test that runs entirely during a deal event may not reflect normal behaviour.
Reading the Results Without Fooling Yourself
When the experiment ends, MYE shows each version's metrics and a probability that one version is better. Treat that probability with respect. A high probability, such as a strong majority in favour of one version, suggests a real winner. A result that sits close to an even split means the test could not separate the two versions, and the honest conclusion is that the change did not matter much to shoppers.
Look at more than one metric. A main image can raise clicks but lower conversion if it overpromises, for example by making a product look larger than it is. Check units sold per unique visitor and sales, not only one headline number. If one version wins on the measure you cared about and loses badly on another, dig deeper before publishing.
An inconclusive result is still useful. It tells you the variable you tested is not a lever for this product, so your next test should try something bolder. Keep a simple log of every experiment: date, ASIN, hypothesis, versions, duration, outcome and what you learned. Over a year this becomes a valuable record of what your customers respond to.
A Worked Example: Testing a Kitchen Storage Main Image
A brand sells a set of six stackable glass food containers with bamboo lids. The current main image shows all six containers stacked in a tower, straight on. Reviews mention that buyers were surprised the set includes three sizes, and the team suspects the tower hides that detail at thumbnail size.
The hypothesis: "If we show the six containers arranged side by side in a shallow arc, grouped by size, more shoppers will choose this listing, because the three sizes will be visible at a glance." The designer produces version B on pure white with the set filling the frame, no text, and the same lighting and colour grade as the original so only the arrangement changes.
The ASIN is eligible, and Amazon recommends a six-week duration. The team schedules the start for the beginning of a week, outside any major deal event, and agrees not to change price or coupons during the test. After six weeks, version B shows higher units sold per unique visitor with a high probability of being the better version. The team publishes version B, records the result, and plans the next test: whether adding one lid removed to show the seal improves conversion further.
Testing Images When You Cannot Use MYE
If your ASIN is not eligible, or you are not the brand owner, you can still learn, just with less certainty.
- Sequential testing. Run image A for a fixed period, then image B for the same length, and compare sessions, conversion rate and, if available, search click-through. This is vulnerable to seasonality and ad changes, so keep everything else steady and repeat the test if the result is close.
- Sponsored ads as a proxy. Click-through rate on Sponsored Products reflects the main image as well as price and title. A clear change in ad click-through after an image swap, with bids and targets unchanged, is a reasonable signal.
- Search Query Performance. Brand owners with access to this report can compare impression-to-click ratios before and after a change for important search terms.
- Consumer polling services. Paid panels of shoppers can vote on image options before you upload anything. They tell you about preference, not purchases, so treat them as a filter for ideas rather than a final answer.
Whatever method you use, change one thing at a time and write down the dates. Without a record of when the image changed, the data in Business Reports is very hard to interpret later.
Secondary Images and What Not to Over-Test
Main images get the most testing attention because they influence the click. Secondary images influence the purchase decision once a shopper is on the page, and Amazon may offer experiment types that cover them or related content. Where you cannot test them directly, focus on the order and content of the stack rather than tiny design tweaks, and judge changes by conversion over a fair period.
Do not test endlessly. Each experiment ties up an ASIN for weeks. Prioritise your best-selling and most-advertised ASINs, where a small improvement adds up to the most sales, and accept good enough on low-volume products. When results suggest the whole set is tired rather than just the hero shot, our guide on when and how to refresh your listing images explains a fuller approach.
Image Experiment Checklist
- Confirm Brand Registry and that the ASIN appears in the eligible list.
- Write a hypothesis that names one meaningful change and the reason.
- Make sure both images meet main image rules and technical specs.
- Keep lighting, colour and quality identical so only the tested variable differs.
- Use the recommended duration, in full weeks, away from major events.
- Freeze price, coupons and title changes during the test where you can.
- Read probability and several metrics before publishing a winner.
- Log every experiment and plan the next test from what you learned.
Desnora designs test-ready image variants built around clear hypotheses, so every experiment has a real chance of teaching you something. Explore our Amazon listing image design service or contact us to plan your next round of tests.
Frequently Asked Questions
Do I need Brand Registry to use Manage Your Experiments?
Yes. MYE is available to brand owners enrolled in Amazon Brand Registry, and only for eligible ASINs that receive enough traffic to produce a result.
Why is my ASIN not eligible for an experiment?
The most common reason is insufficient traffic. Other reasons include an experiment of the same type already running or the ASIN not being associated with your registered brand.
How long should an Amazon image experiment run?
Usually several weeks. Amazon lets you choose a duration and can recommend one based on your traffic. Use whole weeks and avoid stopping early.
Can I test a main image with text or a badge on it?
No. Both versions must follow main image rules, including a pure white background and no text, logos, badges or borders.
What does the probability figure in the results mean?
It is Amazon's estimate of how likely it is that one version performs better than the other. A result close to an even split means the test could not tell the versions apart.
What should I do with an inconclusive result?
Treat it as evidence that the change did not matter much to shoppers. Keep the current image and design a bolder test next time.
Does Amazon publish the winner automatically?
Depending on your settings, Amazon may offer to publish the winning version automatically, or you can choose it manually. Check the option when you set up the experiment.
Can I change my price during an image test?
You can, but it is better not to. Both groups see the price, yet a big change alters the context in which shoppers judge the image and makes the lesson less clear.
How can I test images without Manage Your Experiments?
Use sequential before-and-after periods, compare Sponsored Products click-through with bids unchanged, or use paid shopper polls to filter ideas before uploading.
Should I test two very similar images?
Usually not. Small differences rarely produce a clear result. Test one meaningful change, such as angle, packaging or quantity shown.
How many image experiments should I run?
Focus on your highest-volume ASINs first, where small gains matter most, and run one experiment at a time per content type on each ASIN.
Should I keep records of past experiments?
Yes. A simple log of hypotheses, dates, versions and outcomes builds a picture of what your customers respond to and stops you repeating tests.