How to Run a Gallery Coverage Audit in Five Minutes

Nine labels, one per image

You are not judging how good each photograph is. You are asking one narrow question of every image: which single job is it doing for the shopper? Write the label down and move on without going back to improve an answer. The value is in the finished list.

Each image gets one label, and it has to be the first label that fits. That rule is what makes the pass work. If you let an image carry two labels, every half decent lifestyle photo starts counting as a scale shot. The gap you were looking for dissolves into a maybe. Be strict: a hand holding the product in a kitchen is lifestyle use, not scale, unless the hand is there to show size.

Keep the names as written. The method depends on labels meaning the same thing on every product you sell and every time you run the pass. Invent variants like hero lifestyle or angled detail and you lose what the pass gives you. That is a list you can compare against last quarter's, and against the gallery of whoever is outranking you.

The nine jobs, row by row

Screenshot this table and keep it on your phone. The second column is the one you use while labelling, because it is the question in the shopper's head at that moment. The last column is the version most sellers already have, and it is why so many galleries look complete while leaving the same questions unanswered.

The weak versions are not bad photographs. They are photographs doing a job that was already done. A second front view is a fine image, and it is still not an alternate angle, because it adds nothing the hero showed. Judge each image by what it adds to the set. That distinction is the whole pass.

Run the pass on your phone

Do it on a phone. Open your own listing the way a shopper reaches it, logged out, in the app or on the mobile site. The gallery you see there is the gallery that matters, and it is often not the one you think you uploaded. An image that reads fine on a monitor can be unreadable at thumbnail width inside a swipe carousel.

Swipe through once without stopping, so you meet the gallery the way a shopper does. Swipe again and write a single label in a note for each image, numbered in gallery order. Label first and decide afterwards. Stopping to rewrite a bullet leaves you with a half labelled gallery and no list to read. The pattern only shows up once every image is on the page together.

Then run the same pass on the two listings ranking above you, and compare lists instead of images.

When two images share a label

Two images with the same label raise a question. They are not automatically a mistake. Ask what the second one shows that the first does not. If you can answer in one line, keep both. A jacket whose back differs from its front earns two alternate angles. If you cannot answer in one line, you are spending a slot on a repeat, and slots are the scarce thing here.

When three images share one label, there are two possible causes. First check whether two of them are the same photograph reprocessed: the same shot cropped tighter, rebalanced, or dropped onto a new background. That is a delete. Only then ask whether you have three different photographs all answering the same question.

If they are three different photographs doing one job, keep the strongest and reassign the other two. You reassign the slot and keep the file. Look at your missing labels, take the two nearest the top, and shoot or generate for them. You are moving a slot off a question that is already answered.

Nine jobs, seven slots

You will not fit nine labels into an Amazon gallery. Amazon staff have put the display limit at the main image plus six alternates, with a further two uploadable but not shown. That figure comes from a staff reply in the Seller Forums, posted roughly two years before this was written. So there is a ceiling: nine images up, seven on the page. Display also differs between desktop and mobile, so check your own listing on a phone.

The shortage is what makes the pass worth running. With unlimited slots you would add. With seven, every label you cover is a label you chose over another one, and a duplicate costs you a whole job. Rank your missing labels and take the top few. A gallery that answers seven questions well beats one that answers nine badly.

Walmart works differently. Walmart Canada's Listing Quality Dashboard guide was updated on 8 June 2026. It flags an image count defect when a listing has fewer images than the average for bestselling items in the same subcategory. There is no fixed minimum to hit. The target moves with your category, so count the images on the listings sitting above you.

Which gap to fill first

Fill scale and dimensions first. Scale is the one job no other image can quietly do for you. A shopper can infer material from a texture crop and function from a feature image. Size they cannot infer from anything. Angles and lifestyle scenes leave a product looking the same ambiguous size, which is why a missing scale label costs you more than any other.

The published research here is thin and old, so treat it as a hint. Baymard's product page study found that 28% of sites offered no in-scale image at all. It also found that 42% of users try to gauge a product's size from its images (Baymard Institute, 2017). That study looked at 60 large e-commerce sites, and it is nine years old now.

After scale, order the rest by one test: which unanswered question would most often stop a purchase for this product? For a supplement it is label or specification, because the panel decides the sale. For a multipack it is bundle contents, because the count in your title has to be visible in a picture. Answer the blocking question, then work down.

What this pass does not check

This pass reads what your images depict, which is a separate question from the one marketplaces ask at upload. Marketplaces check the file. Keep the two jobs apart in your head. A gallery can cover all nine labels and still carry a main image that gets suppressed for a reason no coverage pass would catch.

Amazon's main image rules are the example worth knowing. Its Clothing and Accessories category style guide says images must occupy at least 85% of the image area, on a pure white background, showing the product complete and unclipped. So Amazon does define the basis, and it defines it as area. It states it in a category style guide and not in the general image help, which is why almost nobody quotes it. The sibling post below works through what that means in pixels.

Walmart Canada's Listing Quality Dashboard names thresholds you can check yourself. A zoom-capable image needs at least one side of 1,000 pixels or greater. Any image needs at least one side of 500 pixels or greater. The product must make up at least 50% of the image. Those are Walmart Canada's numbers, and Walmart's United States guide is a different document with a different scope, so do not carry them across the border. Flipkart and Meesho publish far less, and the figures that circulate for both contradict each other, so take the spec from your own seller panel.

Filling a gap without creating a new one

When you shoot or generate the replacement, keep the hero rules in mind even for the alternates. If the new image is directed to a plain white studio background, check the result with a colour picker before you upload. AI renderers do not guarantee an exact RGB value, and a background that reads as white on screen can sit a few points off pure white in the file.

The two labels people most often fill with AI are lifestyle use and scale and dimensions, because both usually want a person in the frame. Since late July 2026, Amazon requires the IPTC keyword contains-synthetic-performer in the dc:subject XMP field for images and video containing photorealistic AI-generated people. Simple AI retouching is treated as ordinary photo editing. Amazon says it will add a customer-facing indicator.

During product analysis, Picmato reads the images already on your listing against these same nine jobs. It reports which are covered, which are weak and which are missing, then suggests which of its 45 listing shot types would close the gaps. It is a best-effort read over the images it managed to analyse, and it can be skipped entirely if that pass fails. What it does not do is check your finished file against a marketplace's rules. That step stays yours, whoever made the image.