E-commerce Ad Format Benchmarks: How to Compare Image, Carousel, Video, and Catalog Ads for Your Store

This page explains how to think about ad format benchmarks for an online store: single image, carousel, video, and catalog-driven ads. It gives you a framework for reading other people's benchmark tables, the format-by-format trade-offs that tend to hold across categories, and a method for building a benchmark from your own account, which is the only benchmark that predicts your own results well.

It does not give you a table of verified click-through rates, cost-per-purchase figures, or "winning format" percentages. Infinall AI has no live data source behind this page, and numbers published without one tend to be either stale, drawn from a narrow set of advertisers, or averaged across categories in ways that make them useless for any single store. Where this page mentions a magnitude, it is marked as illustrative, meaning a typical order-of-magnitude intuition and not a measured statistic. If you need hard industry figures, pull them from the platforms' own published reports or a data provider you trust, and check the date and the category definition on whatever you find.

Infinall AI's role here is on the production side. It turns a product URL into competitor research, strategy, ad copy, and creative (image ads, carousels, and video drafts) that lands in an Approval Queue for your review before anything is exported. It does not buy media, does not manage live ad spend, and does not tell you which format will win in your account. You run the ads in Meta Ads Manager, TikTok Ads Manager, Google Ads, or wherever you advertise, and read the results there.

A benchmark is an average, and averages hide the variables that matter most to a store. Before you compare your numbers to anyone else's, check these five things about the source.

First, category mix. A table that blends fashion, supplements, home goods, and consumer electronics is averaging products with completely different price points, purchase cycles, and visual demonstrability. A format that looks weak on average may be excellent for the one category you sell in.

Second, objective and funnel position. Click-through rate on a prospecting ad aimed at cold audiences and on a retargeting ad shown to cart abandoners are different animals. If a benchmark does not say which it measured, the number is nearly uninterpretable.

Third, attribution settings. Platforms report conversions under particular attribution windows, and those settings have changed over the years. A cost-per-purchase from one window is not comparable to another.

Fourth, survivorship. Public benchmark reports tend to come from advertisers using a given tool or agency, who may skew toward larger or more sophisticated accounts than yours.

Fifth, format confounds. Video ads often run on different placements, with different bidding, and different creative effort than static ads. A "video wins" finding may really be "the brand that invested in video also invested in better offers."

None of this means benchmarks are useless. It means they are good for a sanity check on whether a metric is wildly out of range, and poor for deciding which format to produce next. For that decision, your own test data is stronger evidence, as the last sections explain.

A single image ad is the cheapest format to produce and the easiest to test in volume, which is why it is the baseline. For a store with a product catalog, one image per product and angle can be made quickly, and small changes (background, crop, headline, offer badge) can be tested without reshooting anything.

What image ads do well: they communicate a single idea fast, they work across nearly every placement, and they let you isolate variables. Because there is one visual and one message, if one image outperforms another, you usually know why: the offer, the framing, or the product angle. They are also the natural vehicle for a clear price, discount, or social-proof line.

Where they struggle: demonstrating how a product works or moves, showing scale, or holding attention in feeds where motion dominates. A product whose value is "you have to see it in use," such as a kitchen gadget or a garment's drape, can underperform as a static image regardless of how well it is designed.

Practical benchmark guidance: when you evaluate image ads, compare like with like. Compare image ads against other image ads for the same product, objective, and audience, and note that a typical feed click-through rate for static image ads is often described as sitting in the low single digits of a percent or below. That is an illustrative range, not a verified statistic, and it varies widely by platform and category. Your conversion rate after the click depends more on your product page than on the ad format, so check that before blaming the creative.

Infinall AI's Full Campaign generates image ads as part of the pipeline: on the Free plan in Lite mode (up to 3 image ads and 1 carousel per campaign), and on Pro at "$30/month" in Standard mode (up to 6 image ads and 3 carousels). Everything lands in the Approval Queue for you to edit or reject before export.

Carousels and catalog ads solve a different problem from single images: showing more than one thing. They are worth separating, because the two behave differently.

A manually built carousel lets you control the story: card one hooks, cards two to five show product benefits, use cases, or a before-and-after, and the last card carries the offer. This suits products with several selling points, bundles, or a sequence worth explaining. The risk is that most viewers do not swipe through every card, so the first card has to work as a standalone ad. A common failure is putting the strongest benefit on card three.

A catalog or dynamic product ad is assembled from your product feed. Instead of designing each ad, you set a template and the platform fills it with products, often prioritising items a person viewed or similar ones. These tend to be most valuable for retargeting and for stores with many SKUs, because relevance comes from the data rather than from hand-built creative. Their quality depends heavily on the feed: clean titles, accurate prices and availability, and good primary images. A weak feed produces weak ads no matter how clever the template.

How to benchmark them: do not compare a catalog retargeting ad's results to a prospecting carousel's. Compare catalog ads against your own previous catalog setup (new template, new feed images), and carousels against other carousels aimed at the same audience. The audience and funnel stage are the dominant variables.

What Infinall AI does and does not do here: it generates carousel copy and creative from your product URL and puts it in the Approval Queue. It does not build or manage your product feed, does not connect to your catalog to run dynamic ads, and does not place media. Feed setup stays in your platform and your commerce system.

Video earns its place when motion is part of the argument: showing a product in use, demonstrating a transformation, conveying size, texture, or fit, or putting a human face on a claim. In feeds that are dominated by short video, a static ad can simply be skipped faster, which is the strongest practical case for video on platforms like TikTok and Reels placements.

The real cost of video is not the file; it is the volume of distinct ideas required. A store can test twenty image variations from one photo shoot, but twenty genuinely different videos need twenty concepts. This is why many stores that "tried video" ran two clips, saw an average result, and concluded video does not work, when the test was too small to say anything. Hooks, the first second or two, usually matter more than production polish, so plan to vary the opening rather than the whole piece.

Benchmark cautions specific to video: view-based metrics (three-second views, hold rate, average watch time) are diagnostic, not outcomes. A video with a strong hold rate and no purchases has a message or offer problem. Equally, a video that sells well with a modest hold rate is still working. Judge video on the same business outcome you use for other formats, and use the view metrics only to explain why.

Illustrative framing, not a verified figure: video cost-per-click is sometimes lower and cost-per-purchase sometimes higher than static in prospecting, but that depends so heavily on category and offer that you should treat any universal claim skeptically.

Infinall AI produces video ad drafts, and on Pro, video generation up to 4K. Drafts are drafts: they land in the Approval Queue, and you review for accuracy, brand fit, and platform requirements. It does not guarantee a draft will perform, and it does not replace live-action footage where authenticity is the point, such as real customer or founder footage.

The most reliable format benchmark is one built from your own account, because it controls for your category, price, audience, and offer. Here is a method that is deliberately simple.

Pick one product and one objective. Fix the audience and the offer. Produce the same core message in each format you want to compare: one image, one carousel, one short video, and a catalog variant if you have a clean feed. Keep the headline, offer, and landing page identical so that format is the main variable.

Decide in advance what you will measure and what counts as a decision. Primary metric: the business outcome you actually care about, usually purchases or cost per purchase. Secondary diagnostics: click-through rate and, for video, hold rate. Write down the minimum spend or number of conversions you need before you will call a winner, and honor it. Stopping a test at the first lead is the most common way to build a false benchmark.

Run each format for the same period, ideally concurrently, so that sales events and weekday effects hit all of them. Record results in a simple table: format, spend, impressions, clicks, purchases, cost per purchase. Repeat the test on a second product before generalising; a format that wins on one SKU may not win across your catalog.

Then use your table as your benchmark. Next quarter, compare new creative to these baselines instead of to an industry average. Over time you will learn things no public report can tell you, such as that your carousels carry your bundles and your short videos carry your newest launches.

Infinall AI can generate the creative variants for this kind of test and keep them in an Approval Queue where you check each one before export. It does not run the test, allocate budget between formats, or interpret your results. That analysis is yours, or your analyst's.