Product Image SEO for Ecommerce: What Moves Listings and What Doesn't

Ecommerce is the one place where image SEO has an unambiguous line to revenue. A shopper searching for a product in Google Images and landing on your page is a buying-intent visit that never appeared in a keyword tool, and it either happens or it does not depending on work that takes minutes per product.

It is also where the most equity gets destroyed by platform defaults. As of July 2026, the typical mid-sized store is losing image visibility to three things: filenames generated by the upload pipeline, variant images duplicated across dozens of URLs, and product schema with a missing or broken image value.

Start with what Google requires, not what it prefers

There are two separate systems with different consequences for getting it wrong, and confusing them wastes effort.

Product structured data governs rich-result eligibility in organic search. Google's Product structured data documentation lists image among the required properties. A missing, relative or non-crawlable URL there can make an item ineligible for the visual treatment its competitors get. Requirements worth honouring: absolute URLs, crawlable paths, and multiple images at different aspect ratios — commonly 1:1, 4:3 and 16:9 — so Google can pick what suits the surface.

Merchant Center feed rules govern Shopping. These are hard validation gates rather than ranking factors: minimum dimensions, no promotional text or watermark overlays, no borders, and the image must accurately show the product being sold. Break one and the item is disapproved, which is a binary loss of a paid or free listing rather than a soft ranking effect.

Fix these two before anything else on the page, because both have cliff-edge failure modes rather than gradual ones.

The variant-image decision

This is the most common structural mistake in ecommerce image SEO, and it is easy to get right once the rule is clear.

Different appearance means different image and different URL. A shirt in six colours is six images. Each can accumulate its own image-search visibility for its own colour query, and a shopper searching for the burgundy one should find the burgundy one.

Identical appearance means one image, referenced from many pages. Six sizes of the same shirt in the same colour do not need six copies of the same file. Duplicating it splits whatever equity it earns across near-identical URLs, multiplies crawl demand, and gives Google a deduplication problem it will resolve by picking one arbitrarily.

Platform defaults frequently do the wrong thing here, generating a distinct upload per variant regardless of whether the image differs. Auditing this is a query against your own product data, not an SEO tool.

What to shoot, and in what order

Volume is not the goal — coverage is. A useful priority order, given a limited budget per product:

Priority	Shot	Purpose	Generated imagery acceptable?
1	Primary on clean background	Feed compliance, listing thumbnail	No — must be the real product
2	Two to three alternate angles	Answers "what does the back look like"	No
3	Scale or in-hand shot	Prevents the biggest returns driver	No
4	Detail crop of the critical feature	Texture, closure, connector, finish	No
5	Lifestyle / in-context	Aspiration, category pages, ads	Yes, with care
6	Category and collection headers	Navigation, editorial	Yes
7	Blog and guide imagery	Content marketing	Yes

The line between rows 4 and 5 is the important one. Everything above it depicts the item a customer is buying and must be a photograph of that item. Everything below it depicts context, where the subject is a setting rather than a product, and where generated imagery is both legitimate and considerably cheaper than a location shoot.

Getting this backwards — generating product shots to save on photography — creates a misrepresentation problem, a returns problem and a trust problem well before it creates an SEO problem. Our post on commercial use and copyright covers where that line sits legally.

Filenames and alt text at ecommerce scale

Google Search Central's image guidance names descriptive filenames explicitly, and ecommerce is where the default behaviour is worst. Stores routinely ship 1a2f9c-3.jpg, IMG_0042.jpg, or a SKU hash — thousands of times.

The pattern worth enforcing is product name, variant, view:

oak-dining-chair-walnut-front.jpg
oak-dining-chair-walnut-detail-joint.jpg
oak-dining-chair-walnut-in-room.jpg

Hyphens rather than underscores, lower case, no stop words, no keyword stuffing. The full reasoning is in our filename guide.

Alt text follows the same discipline with an important difference: it should not simply repeat the product title, because you have four to eight images on the page and repeating the title across all of them tells a screen reader user nothing about which is which. Describe what this image shows — "walnut finish oak dining chair seen from the front", "close-up of the mortise joint at the seat rail". The alt-text guide covers the judgement calls.

This is exactly the work that does not survive contact with volume. A store adding 200 SKUs a quarter, each with six images, is 1,200 filename-and-alt-text decisions nobody has scheduled. Which is why it ends up automated or skipped, and skipped is the default.

Category and collection pages are the forgotten half

Product pages get the attention; category pages get the traffic for the broader, higher-volume queries. And their imagery is usually a grid of product thumbnails with whatever alt text the template generated, which is frequently the product title repeated verbatim.

Three things are worth fixing there.

Thumbnail dimensions. A grid of 40 products serving full-size images is the most common cause of a slow category page. The thumbnails should be served at thumbnail size through srcset, and everything below the first row should be lazy-loaded. The first row should not be — it usually contains the LCP element.

Alt text that distinguishes. In a grid, alt text repeating the product name for every tile is technically present and practically useless: a screen reader user hears forty product names with no indication of what they look like. If the template can only produce the title, that is better than nothing, but a colour or material qualifier costs one field and makes the grid navigable.

A genuine category image. Collection pages frequently have room for one editorial header image, and this is the clearest legitimate use for generated imagery in ecommerce — the subject is a mood or a setting rather than a specific purchasable item. It gives the page something to be about visually without misrepresenting anything.

Accessibility is a compliance question in retail

Worth stating separately because the risk profile differs from a content site. Retail sites in several jurisdictions have been the subject of accessibility complaints and litigation, and images with missing or meaningless alt text are among the most frequently cited defects because they are trivially detectable at scale by automated scanners.

The relevant standard is WCAG, whose non-text content requirement is met by providing a text alternative that serves an equivalent purpose. For product imagery that means describing what the image shows, using alt="" on genuinely decorative elements, and never leaving the attribute off entirely — an omitted attribute can cause a screen reader to announce the filename, which on an ecommerce site is usually a SKU hash.

The convenient part is that this is the same work image SEO asks for. There is no separate accessibility pass; a descriptive alt attribute satisfies both, which makes it one of the few tasks with two independent business cases behind it.

Discovery: the failure that looks like a ranking problem

Before concluding your images underperform, confirm Google can see them at all.

JavaScript injection. If images are loaded by a client-side gallery component and absent from the server HTML, Google may render the page and find them — or may not, at the scale and frequency you would like. Use URL Inspection and read the rendered HTML. Product images belong in the initial response.

Blocked paths. A Disallow on a CDN or asset directory silently removes every image behind it from image search. This is depressingly common on stores that added a CDN after launch.

Missing sitemap coverage. Image entries in your XML sitemap give Google a discovery path independent of page crawling, which matters most for large catalogues where crawl budget is a real constraint. Our image sitemaps guide covers the markup.

Lazy loading the wrong image. The primary product image is frequently the LCP element. Lazy-loading it delays render on the highest-value page type you have — see images and Core Web Vitals for why that one attribute matters so much.

The realistic ecommerce workflow

Three of the four buckets here are configured once. Format, dimensions, responsive variants and delivery live in your platform or CDN and then apply to every future upload — our format comparison covers the choices. Schema is a template edit. Feed compliance is a validation rule.

The fourth bucket — filename, alt text, embedded metadata — is per-file, per-product, forever, and no build step can invent it because it does not know your product names or which angle it is looking at.

For the lifestyle, category and editorial rows of that table, SEOpix generates the image with the descriptive filename, alt text and EXIF block already written, so context imagery arrives described instead of joining the backlog. For the product shots themselves, photograph the product — that is not a corner to cut.

Audit two things this week: whether your product schema image values are absolute and crawlable, and whether cosmetically identical variants are duplicating image URLs. Both are quick, and both are worth more than a general optimisation pass. If context imagery is the part slowing you down, ten images a month are free and the pricing tiers cover catalogue volume.