- You need a Professional account, a Brand Representative role in Brand Registry, and an ASIN with enough recent traffic.
- As of November 2026, Amazon lists images, titles, bullets, descriptions and A+ Content including Brand Story; secondary image testing is unconfirmed.
- Start with the main image on your busiest eligible ASIN, then the title.
- Amazon recommends 8 to 10 weeks for a custom duration, and to significance can sometimes finish in about four weeks.
- A 90% result means a 90% chance Version B is better, and the size of the gain sits in the conversion and units columns.
- After an inconclusive test, keep Version A and find out whether the cause was too small a change, too little traffic or a true tie.
What will you accomplish with Manage Your Experiments?
By the end you will have one live A/B test on your busiest eligible ASIN, a written hypothesis behind it, and a rule agreed in advance for acting on the result. That rule is what stops a 70% score from becoming a permanent listing change that nobody questioned.
You will need a Seller Central login with the right role, a Version B that differs clearly from the current content, and a rough daily visitor count for estimating test length.
Before spending eight weeks on a test, run the listing through the free listing checker and clear the obvious faults. Our listing mistakes guide covers repairs that need no experiment.
Amazon's tool page says optimized content can lift sales by up to 20%, citing internal data, and its seller blog says up to 25%. Both are ceilings, so leave them out of your forecast.
Who is eligible for Amazon Manage Your Experiments?
Amazon's tool page sets three conditions as of November 2026: a Professional selling account, a Brand Representative role for a brand enrolled in Brand Registry, and a product from that brand with enough recent traffic to give valid results. Brand Story and A+ tests also need that content published first.
Amazon publishes no traffic number. Moderator and staff posts on its Indian and European seller forums, both about three years old, say a high-traffic product might see several dozen orders a week or more, varying by category. Let the tool settle it: start a new experiment, pick a type, and see which ASINs appear.
If your best product is missing, the Indian forum post suggests sending it more traffic through advertising. Buying clicks purely to qualify for a test is hard to justify on a thin margin, so run the numbers in the ACoS and TACoS calculator first.
Which marketplaces and account types are covered?
Coverage is less clear than eligibility. ecomEngine's October 2024 guide ties access to the US marketplace and Helium 10 warns that some regions may lack it. Amazon's own 2023 best practice guide talks about running tests across multiple marketplaces, and Amazon staff have promoted the tool on UK and German forums.
We found no official list by country, so sellers in the UK and India should look under the Brands menu in their own account. ecomEngine and Gorilla ROI both say vendors have access too. Amazon's public page only describes Seller Central, so that stays unconfirmed.
What can you test in Manage Your Experiments right now?
As of November 2026, Amazon's tool page names images, titles, bullets, descriptions and A+ Content, with Brand Story folded into A+. Count Brand Story separately and you get six elements; count it inside A+ and you get five. Guides that still say three look out of date against that page.
| Source | What it lists | Count |
|---|---|---|
| Amazon tool page (read November 2026) | Product images, titles, bullets, descriptions, A+ including Brand Story | 5 or 6 |
| Amazon moderator, UK forum (about two years old) | Main images, titles, bullets, description, A+; later mentions supporting images and Brand Story | "Five types" |
| ecomEngine (October 2024) | Titles, bullets, description, images, A+, Brand Story | 6 |
| Gorilla ROI (updated February 2026) and Helium 10 | Titles, images, A+ | 3 |
Supporting images are the open question. The tool page says "product images" without saying which ones. Amazon's 2023 guide says main images, Helium 10 says secondary images cannot be tested, and the UK moderator post names supporting images in one sentence after listing main images in another. We could not read a current Seller Central help page on this, so treat secondary image testing as unconfirmed until you see the option in your account.
Amazon also offers multi-attribute experiments, which put title, images and bullets into a single test. An older European forum notice included A+ Content in this kind of test, so check your account. The Indian forum post says an ASIN can run only one experiment at a time.
We agree with ecomEngine: save multi-attribute for a full listing refresh, when you only want to know whether the whole package beats the old one.
How do you set up an experiment, step by step?
Setup runs from the Brands menu in Seller Central and takes eight steps, from confirming your role to reading the results page. Most of the thinking comes before you open the tool, because the hypothesis and Version B decide whether the test can teach you anything.
- Check that you have a Professional selling plan and a Brand Representative role on a brand in Brand Registry. The Rights Owner or an Administrator can assign the role.
- In Seller Central, go to Brands, then Manage Experiments, select Create a New Experiment and pick a type. The tool lists the ASINs that qualify for that type.
- Write down what you will change and the result you expect, with a number attached.
- Prepare the alternate title, copy, image or A+ layout you want to test.
- Choose the experiment type and primary product, then enter the experiment name and your hypothesis.
- Open View Settings and decide on the start date, the duration and whether auto-publish stays on.
- Select Schedule Experiment and avoid other edits to that ASIN until the test ends.
- When the test ends, select View Details and apply the decision rule you set in advance.
How do you write a useful hypothesis?
Amazon's guide suggests a simple form: changing X by adding or removing Y will produce result Z. Here is one for a main image: "Changing our main image from the boxed product to the product out of the box, filling more of the frame, will lift units per visitor by 8%."
How do you build Version B?
Version A is whatever is live now. Copy goes straight into the tool, and images go in through an upload and the file picker. For A+ and Brand Story, you duplicate Version A and edit it in the A+ Content manager. Amazon's guide says both A+ versions must sit on the same set of ASINs or setup will fail.
Which default settings should you change?
By default the test starts once validation finishes, runs to significance, and publishes the winner automatically. We usually turn auto-publish off, because of where the bar sits. Amazon's 2023 guide says a test concludes at 66% or better, and a forum notice from about four years ago applies the same 66% to auto-publish. The notice words it as "at least 66% better", but the guide's FAQ makes clear these percentages describe probability and say nothing about size. Amazon's current public pages do not restate the figure, so it may have changed.
What should you test first, and why?
Start with the main image on your busiest eligible ASIN, then the title, then A+ Content or bullets. That order is our judgment, and ecomEngine's guide also names the main image as a high-impact place to begin. Traffic decides how fast you get an answer, and shoppers meet these elements first.
| Order | Element | Version B ideas | Watch out for |
|---|---|---|---|
| 1 | Main image | Product out of the box, tighter crop, a different angle | Version B must still pass the main image requirements |
| 2 | Title | Move or remove the brand name, add a benefit, cut to under 100 characters (all from Amazon's tips) | Keep to the rules in our title guide |
| 3 | A+ Content | Swap plain image modules for a comparison chart or image and text pairs | Read our A+ Content strategy first, since each version costs design time |
| 4 | Bullets and description | Reorder by benefit, or lead with size and fit details | Fewer shoppers may read this far, so the effect can be harder to detect |
Pick the ASIN on margin as well as traffic. A 5% gain on a product earning $12 a unit is worth six times the same gain on a product earning $2 at equal volume.
One limit on image tests: Amazon's published list of result metrics has no click-through rate, so the score rests on purchases. Gorilla ROI says both versions appear in search results, while Amazon's page only describes shoppers viewing the detail page. We could not confirm how search placements are handled.
If you need a side-by-side comparison to spot the difference between A and B, a shopper scrolling on a phone probably will not notice it. Pick something bigger.
How long should you run an experiment?
Amazon recommends 8 to 10 weeks if you set your own duration. On the default to significance setting, results can sometimes arrive in about four weeks. Either way, let the test finish, because results refresh only weekly and Amazon warns that ending early can mean picking the wrong version.
Our choice depends on what the change cost. For text you can undo in minutes, to significance is fine. For a new photo shoot or an A+ redesign, we would set a fixed 10 weeks, since a test that can stop at 66% may leave you with thin evidence for that spend. Amazon's guide favors to significance, so this is our opinion.
Traffic sets the limit. This illustration uses plain normal approximation arithmetic, simpler than Amazon's Bayesian model, so read it for scale only. Take an ASIN with 200 unique visitors a day (100 per version) converting at 11%, and ask how long it takes to reach about 90% probability.
| True lift in conversion | Visitors needed per version | Time at 100 per version a day |
|---|---|---|
| 1 point (11% to 12%) | About 3,300 | About 5 weeks |
| 0.5 points (11% to 11.5%) | About 13,100 | About 19 weeks |
Halve the lift and you need roughly four times the visitors, which is why small tweaks so often end without a winner. Your count may run below your session figures, since Amazon's guide excludes visitors it cannot identify and counts one customer once a day.
Season matters too. Amazon's guide wants a test to span weekdays, weekends, and high and low periods. An 8 to 10 week test started in early November runs through Black Friday and into January, so read the promotion section below first. We found no published maximum length.
How do you read the probability and impact results?
The probability is the chance that one version is better, and it tells you nothing about how much better. Amazon's 2023 guide says a 90% score means a 90% chance that Version B beats Version A. Use conversion and units per visitor for size, and treat projected one-year impact as a rough estimate.
Amazon uses a Bayesian method, so a 75% score is the share of its modeled outcomes that show a lift. The other numbers have plain meanings:
- Sample size counts the unique shoppers shown each version.
- Conversion rate is the share of those shoppers who bought.
- Units per unique visitor divides units by that sample, so it also picks up multi-unit orders.
- Projected one-year impact appears only on completed tests. Amazon's guide describes it as the winner's average daily sales increase times 365, with no adjustment for season or price changes.
ecomEngine also reports an evidence strength label such as "Weak", plus units and sales from search, and notes that low-confidence winners can show a negative worst case. Amazon's public pages do not describe those fields, so check your own screen.
Amazon's bar is 66%. Seller Labs advises applying a winner only at 95% or when the test ends on its own. At 66%, about one time in three the other version is the better one, so we use a sliding rule based on what it costs to be wrong. The 80% and 95% cutoffs are our own.
| Probability | How we read it | What we do |
|---|---|---|
| Below 66% | Amazon's bar not met | Treat as inconclusive and keep Version A |
| 66% to 80% | Leaning, with real risk of a wrong call | Publish only if it is cheap to reverse and the worst case is tolerable |
| 80% to 95% | Fairly strong | Publish, then watch conversion for several weeks |
| 95% and up | Strong | Publish, then test the same idea on similar ASINs |
What does reading a result look like in practice?
This is an invented illustration of an eight-week main image test on a $30 product, and no brand's real data appears in it. Version B wins at about 94% probability with about 9% more units, so we would publish it while treating the one-year figure with caution.
| Metric | Version A | Version B |
|---|---|---|
| Sample size | 5,000 | 5,000 |
| Conversion rate | 11.0% | 12.0% |
| Units sold | 605 | 660 |
| Units per unique visitor | 0.121 | 0.132 |
| Sales | $18,150 | $19,800 |
Version B sold 55 more units and $1,650 more, a 9.1% gain, while seeing only half the traffic. The 94% comes from our simple approximation, and Amazon's model would produce its own figure.
For the one-year number, scale up to everyone seeing Version B. That doubles the gain to 110 units and $3,300 over 56 days, or $58.93 a day. Multiply by 365 and you get about $21,500 in sales, or about 717 units. If the product earns an assumed $8 a unit, that is about $5,700 in profit, which is the number to weigh against what the new image cost. Amazon does not publish how it scales for the split, so your screen may differ.
Now look at the range. The 95% interval on that one-point conversion gap runs from about minus 0.25 points to plus 2.25 points. Using the same scaling, one-year sales run from about minus $5,400 to plus $48,400. The downside is small against the upside, which supports publishing.
Suppose Version B had converted at 11.35%. Probability falls to about 71%, which clears Amazon's 66% bar, so auto-publish would likely push it live. The central estimate is about $7,500 a year, but the range runs from about minus $19,100 to plus $34,200. For an easy title change we might accept that. For a costly reshoot we would run a bolder Version B.
What should you do with an inconclusive result?
Keep Version A unless Version B has some other advantage, such as better brand fit or a lower production cost, then work out why the test could not separate them. Amazon's 2023 guide gives three possible causes: too small a change, too little traffic, or two versions that perform about the same.
| What you see | Likely cause | Next move |
|---|---|---|
| Thin sample, wide range, version gap could be large | Too little traffic | Rerun in a busier period, or with more ad traffic if the margin allows |
| Solid sample, and you changed one word or a slight crop | Change too small | Test a bolder Version B on the same element |
| Solid sample, a clearly different Version B, and almost no gap | Versions perform alike | Stop spending on this element and move to the next one |
On that last row, Amazon's guide points out that a flat result can show a type of content is not worth more investment, so two tied A+ layouts may save you a redesign budget.
Skip an identical rerun unless traffic was the problem. Amazon does suggest revisiting a hypothesis later, because shoppers' view of a brand changes as it grows. Keep a simple log of dates, hypothesis, sample size, probability and any promotions that ran, so nobody repeats a test by accident.
What mistakes ruin Manage Your Experiments results?
In our view, most wasted tests come from a few habits: changing too little, changing too many things, ending early, editing the listing mid-test, and running a short deal inside the window without planning for it. Each one either weakens the evidence or gives you a result that fails to hold in normal selling weeks.
Testing too small a change
Amazon's blog says the tool warns you when two versions look too similar. ecomEngine's example of a change too small is swapping one keyword in a bullet. As the traffic table above shows, a subtle lift can need months of visitors.
Bundling changes you want to learn from separately
Amazon's guide uses an apparel title as its example: add sizes, audience and a feature in one go, and you will not know which one mattered.
Ending early or editing during the test
Amazon warns that stopping early can overstate the effect or crown the wrong version. Seller Labs also advises against editing the title, images, bullets or description outside the experiment while it runs. We would add price to that list, since Amazon's projection takes no account of price changes.
Running a promotion inside the test
Amazon's 2023 guide recommends pairing an experiment with a promotion, Sponsored Ads or a launch to bring more traffic and faster results. Seller Labs says to avoid major promotions and Lightning Deals during the window, because price-driven spikes distort conversion. Amazon's guide dates from 2023, and no newer Amazon page we read reverses it.
Our view sits between them. Random assignment shows both versions the same deal, so the comparison stays fair, but a shopper buying at a steep discount is reacting largely to price. A winner chosen during a spike may not hold at full price, and the 365-day projection carries that spike forward. Steady ad spend across the whole test is fine. For a Lightning Deal or Black Friday, either wait until it has passed or run long enough that the event is a small share of the sample, and record it in your log.
Reading 90% as a 90% lift
Amazon's FAQ addresses this directly. In the illustration above, 94% probability went with about a 9% gain.
What should you check before and after each experiment?
Use this checklist each time you schedule a test and again when it ends. It covers access, the quality of Version B, timing around sales events and your decision rule. Agree that rule before launch, since a result is easier to read honestly when the threshold was already written down.
Run these before you schedule, then the second list once the test ends.
- Professional plan and Brand Representative role confirmed.
- ASIN appears as eligible for this experiment type.
- Hypothesis names the change and an expected number.
- Version B is plainly different and meets Amazon's content rules.
- Calendar checked for deals, price changes and stock gaps.
- Auto-publish decision made, and your probability threshold written down.
- Probability read as a chance of being better, with size read separately.
- One-year projection recalculated as profit, using your own margin in the FBA profit calculator.
- Conversion checked again a few weeks after publishing.
- Next hypothesis written while the result is fresh.
Sources and data notes
Facts in this article were checked against the sources below on November 6, 2026. Amazon and agencies change programs and prices often, so confirm details before you act.
- Amazon: Manage Your Experiments tool page
- Amazon: Optimize product content and increase sales with A/B testing
- Amazon: Manage Your Experiments Best Practice Guide (October 2023)
- Amazon Seller Forums: Improve Conversions with New A/B Testing Features
- Amazon Seller Forums UK: Increase Conversion, Manage Your Experiments
- Amazon Seller Forums Europe: Save time by running simultaneous experiments
- Amazon Seller Forums India: Manage Your Experiments moderator post
- ecomEngine: Manage Your Experiments on Amazon
- Seller Labs: Manage Your Experiments, A/B Testing Listings the Right Way
- Helium 10: Know your audience with Amazon's Manage Your Experiments tool
- Gorilla ROI: Amazon AB testing with Manage Your Experiments
Frequently Asked Questions
Is Amazon Manage Your Experiments free?
Yes. Amazon's seller blog describes it as a free tool, so there is no separate charge for running a test. You do need a Professional selling plan, which Amazon lists at $39.99 a month plus selling fees, and a brand enrolled in Brand Registry. Your other costs are the design time for Version B and any ads you run to bring in traffic.
How much traffic does an ASIN need for Manage Your Experiments?
Amazon publishes no fixed number. Older Amazon forum posts say high-traffic products may see several dozen orders a week or more, depending on category. The dependable check is inside the tool: select Create a New Experiment, choose a type, and see which ASINs are listed. If yours is missing, more traffic through ads or other channels may help it qualify later.
Can you test secondary images in Manage Your Experiments?
We could not confirm this. Amazon's tool page says product images without saying which ones, its 2023 guide says main images, and one Amazon moderator post mentions supporting images. Helium 10 says secondary images cannot be tested. The fastest answer is to open an image experiment in your own account and see which image slots the tool lets you change.
What does 66% mean in Manage Your Experiments?
Amazon's 2023 guide says a test concludes when one version reaches 66% or better, and an older forum notice applies the same bar to auto-publish. It is a probability, so it means about a two in three chance that the version is better. Amazon's current public pages do not restate the figure, so it may have changed since then.
Can you run two experiments on the same ASIN at once?
An Amazon moderator post says only one experiment can run on an ASIN at a time. To change several elements together, Amazon offers multi-attribute experiments covering title, images and bullets in a single test. The drawback is that you learn whether the package won, and you will not know which element caused the difference between the two versions.
Does Manage Your Experiments work in the UK and India?
We could not find an official list of supported countries. ecomEngine's 2024 guide ties access to the US marketplace, while Amazon's 2023 guide mentions multiple marketplaces and Amazon staff have posted about the tool on its UK and Indian forums. Sellers there should look under the Brands menu in Seller Central for Manage Experiments, which will show whether their account has it.
Can you stop an experiment early?
Gorilla ROI says you may end a test at any point, but Amazon advises against it. Its 2023 guide warns that stopping early can overstate the effect or lead you to choose the wrong version. Results update weekly, so a promising first or second week tells you very little. If something urgent comes up, such as a compliance problem, fix that first and restart later.
Want a second opinion on your next test?
Run your listing through our free listing checker first. If you would like help choosing what to test or reading a result, SellerVine can review the ASIN with you.
Get a free account audit
