How to Optimize Images For eCommerce: From Alt Text to AI Shopping Visibility at Any Catalog Size

Most teams treat alt text as a box to tick. Write something, move on, and if the CMS filled the field with the file name, that counts. At ten products the approach survives. At four thousand SKUs with variants, feeds and a migration in the history, it produces a catalog where the images are technically present and functionally invisible. 

What it takes to optimize images for eCommerce is not a longer checklist of tips, it is a system: whether bots can fetch the image at all, whether anything describes it, whether the page survives the weight of it and whether the product data around it is structured enough for a shopping engine to use. 

This guide covers that system and how to audit it across a whole catalog rather than one product page at a time.

Why You Need to Optimize Images for E-commerce SEO

Image SEO is usually framed as alt text plus compression, which describes two tasks inside a system that has five parts: whether a crawler can reach the file, whether anything describes it, whether the weight lets the page render, whether the structured data around it is machine readable and whether a screen reader user can shop at all. Each part has its own audience, and only one of them is Google organic search.

Traffic and Conversion From Image and Shopping Search

Visual search is a real entry point rather than a rounding error. Google said in October 2024 that people use Lens for nearly 20 billion visual searches a month and that a fifth of those are shopping-related. It is Google’s most recent published number, so treat it as a 2024 baseline. The shopping surface behind it is larger than most catalogs assume: the Shopping Graph holds more than 50 billion product listings, with over two billion refreshed every hour. What decides whether your photo joins them is the data attached to it.

Accessibility and Legal Exposure

Missing alt text is the second most common accessibility failure on the web. In the WebAIM Million 2026 analysis of a million home pages, 53.1% had images with missing alternative text, and across the 66.6 million images sampled, 16.2% carried no alt attribute at all.

eCommerce is measurably worse than average. WebAIM’s Shopping category covered just over 95,000 home pages and averaged 71 detected errors per page, which is 26.6% above the overall sample. By platform the spread is wide: Shopify sites averaged 75.1 errors, Magento 75.8 and PrestaShop 143.2.

The legal exposure is not theoretical either. UsableNet counted more than 5,000 digital accessibility lawsuits filed in the US during 2025, and eCommerce accounted for close to 70% of all ADA web lawsuits. Among the top 500 eCommerce companies, 35.8% received at least one.

In the EU the European Accessibility Act has applied to services placed on the market since June 28, 2025, and eCommerce is explicitly in scope. Its exemption covers businesses under ten employees and two million euro turnover, which excludes essentially every catalog this article is written for.

Visibility in AI Shopping and Generative Engines

Here is where the common assumption breaks. Generative shopping engines do not look at your product photograph and work out what it is. They read structured product data and render the image from a URL that data supplies.

OpenAI’s product feed specification makes image_url a required field and sets no requirements at all for dimensions, file size, resolution or aspect ratio. Its help documentation on how products get surfaced describes structured metadata such as price and product description, and mentions no image analysis. Perplexity’s shopping experience runs on a platform integration with Shopify plus a merchant program for product specifications.

Computer vision does exist in this picture, just at the other end of the transaction. Google Lens and Perplexity’s Snap to Shop are what a shopper points at an object. Neither is how your catalog gets read.

That leaves a specific conclusion for anyone optimizing a catalog. Structured data and feed quality decide whether a product is eligible to be surfaced. Image quality decides whether it is convincing once it appears, and whether it renders at all. Adobe measured this gap directly in April 2026: across US retail sites, product pages scored 66 out of 100 on machine readability, the lowest of any page type, against 75 for home pages and 74 for category pages. Over the same quarter, AI-sourced traffic to retail grew 393% year over year and converted 42% better than other channels.

If you want to see which AI crawlers are actually reaching your product pages rather than assuming, that is a log-level question. Our guide to tracking AI crawler activity covers the setup.

Alt Text and File Naming: Getting It Right at eCommerce Scale

Two signals do most of the descriptive work on an image, and both break in the same place: at the point where images stop being uploaded by a person and start arriving through a pipeline.

File naming has one rule worth stating and then leaving alone: name the file descriptively before upload, with hyphens, as Google’s image best practices put it, my-new-black-kitten.jpg rather than IMG00023.JPG. Once a file sits in the CDN with a generated name, renaming it across a catalog costs more than it returns, so this belongs to the upload process rather than to a remediation project. Alt text is the one worth spending time on.

Writing Alt Text That Is Both Accessible and Crawlable

Good alt text describes what a shopper would need to know if the image never loaded: material, colour, cut, finish and the angle when a gallery has several. Google asks for “useful, information-rich content that uses keywords appropriately” without stuffing, and the practical test is whether the string reads sensibly out loud. “Women’s waxed cotton field jacket in olive, front view” works for a screen reader and for Google Images. “Jacket product image” fails both.

Two catalog-specific rules matter more than the general advice.

Variants need differentiated alt text. If a jacket comes in four colours and every variant page carries the same alt string, three of those pages describe an image that is not on them.

Gallery images need distinct alt text from the main image. A six-image gallery with six identical alt strings gives a screen reader user six copies of the same sentence and gives Google Images nothing to distinguish them.

Why Bulk Feeds and Templated Uploads Skip Alt Text

The alt attribute is the field most likely to be dropped in transit, and the data shows exactly what it looks like when that happens. In the HTTP Archive Web Almanac 2025, about 13 to 14% of images have no alt attribute and 30% use an empty one. Of the images that do have text, roughly 20% carry ten characters or fewer.

The giveaway statistic is in the same chapter: around 8.5% of alt texts end with a file extension such as .jpg or .png. Nobody types that. It is what a CMS writes when it is told to populate alt automatically and has nothing to work with.

Four pipelines account for most of the loss: PIM exports with no alt column, marketplace feed imports that carry the image URL but not the description, migrations where the old alt data sat in a table the script never touched, and bulk uploaders that build the img tag from the file path.

The fix is upstream. Alt text belongs in the PIM as a product field, generated from attributes you already hold and carried through every export, because anything written at the storefront layer gets overwritten at the next sync.

If you need to see what is currently in those fields before deciding, you can bulk export all image alt attributes from a crawl and read the whole set at once.

Finding Every Missing Alt Tag Across the Catalog

Spot-checking works up to a few hundred SKUs. Past that, the problem is not that nobody looked, it is that nobody could look at all of it.

A crawl solves this if image processing is switched on before the crawl runs. In JetOctopus that is the “Process images data” checkbox in Advanced settings when you create a new crawl, and every image report depends on it. From there the Images data table carries the columns that answer the question: Page URL, Full src, Image Alt Text, Is link and Is lazy loading, with a filter on Image Alt Text set to Is Empty to isolate the gaps.

The “Is link” column matters more than it looks. When an image is wrapped in a link, its alt text is doing anchor text duty, so an empty alt there costs you twice. Our walkthrough on how to find and export all images without alt attributes covers the filter setup.

Export every missing alt with its page URL

Try Free Trial

Technical Image SEO: Formats, Compression and Speed

This section is hygiene rather than strategy. It is also where most catalogs are still losing, which is why it cannot be skipped.

Choosing the Right Format

WebP and AVIF are the default now, and the gap between that statement and reality is large. Among Largest Contentful Paint images measured in the Web Almanac 2025, JPG still accounts for 57% and PNG for 26%, while WebP sits at 11% and AVIF at 0.7%. On the single most performance-critical image on the page, more than eight in ten sites are still serving a legacy format.

Browser support is no longer the reason. AVIF is supported by 95.36% of global browser usage, and WebP has been universally supported for years.

The savings are documented. Google’s own studies put WebP lossless images 26% smaller than PNG and WebP lossy images 25 to 34% smaller than comparable JPEG at equivalent quality. AVIF goes further, though the most quoted figure, roughly 60% smaller than JPEG, comes from an imgix benchmark published on web.dev in 2021 rather than from Google research, so quote it with that attribution.

A JPEG fallback still earns its place when your CDN cannot generate modern formats on the fly, and when images are consumed outside the browser, in a Merchant Center feed for instance, where the safest common denominator beats the last 30% of file size.

Compression and Core Web Vitals

The reason format choice matters is that images usually are the LCP. In the 2025 data, an image is the LCP element on 85.3% of desktop pages and 76% of mobile pages.

On eCommerce specifically, desktop Core Web Vitals pass rates ranged from 33% on WooCommerce to 76% on Shopify, and LCP on these sites typically maps to hero imagery and product grids.

Two rules cover most of the damage. Serve thumbnails at thumbnail size rather than shipping a 2000 pixel product photo scaled down in CSS, which is the single most common waste on a category grid. And set fetchpriority=”high” on the image you believe is the LCP element, which Google recommends explicitly in its LCP optimization guidance.

Our article on Core Web Vitals optimization goes deeper on the measurement side, with input from sixteen practitioners.

Responsive Images and Lazy Loading for Product Pages

A product page is not one image. It is a gallery, a set of variant swatches, a recommendation carousel and a grid of related products, which means every decision here multiplies.

Responsive Images and srcset for Product Galleries

On a gallery the temptation is to declare every size for every image, which inflates the HTML and makes the browser parse variants it will never use. Three practices hold up at scale: a small number of meaningful breakpoints rather than one per device, an honest sizes attribute, since a wrong value makes the browser pick a larger file than it needs, and <picture> when you are switching formats rather than resolutions, so a browser without AVIF support falls back cleanly.

Lazy Loading Without Hiding Images From Crawlers

Lazy loading is now standard, with the Web Almanac reporting that 26.5% of images use loading=”lazy” while 68% use no loading attribute at all. The problem is where it gets applied.

Between 16% and 17% of pages lazy-load their own LCP image. Google’s position on that is blunt: “Never lazy-load your LCP image, as that will always lead to unnecessary resource load delay.”

The crawlability risk is separate and more damaging. Google’s lazy loading documentation, updated in December 2025, states that Google Search does not interact with your page. Anything that needs a scroll, a click, a hover or a tap to load will not be seen. Native loading=”lazy” and IntersectionObserver are both fine. A gallery where thumbnails only load after the user clicks a “view more images” control is not.

The way to verify is to check whether src attributes appear in the rendered HTML, either through Search Console’s URL Inspection tool on a sample or through a crawl with JavaScript rendering on. A crawl also lets you check the whole estate at once: the Images data table has a dedicated Lazy load images subset and an “Is lazy loading” column, which is how you find lazy load images and analyze them rather than sampling by hand.

Making Images Crawlable and Indexable at Scale

None of the above matters if bots cannot reach the file. On a large catalog, unreachable images accumulate quietly, because a broken product photo is visible to a shopper and invisible to a report.

Image Sitemaps: Still Worth It?

Yes, with a caveat that changes what they are for. Google still supports and recommends image sitemaps as a way of telling Google about images it might not otherwise find, and treats a separate image sitemap and image tags inside an existing sitemap as equally fine.

What changed is the payload. Only two tags survive, image:image and image:loc, with a limit of a thousand images per URL. The caption, title, geo_location and license tags were deprecated in 2022. An image sitemap in 2026 is therefore a discovery mechanism and nothing else, a list of image URLs, with every piece of descriptive metadata now travelling through structured data, IPTC metadata or the page itself.

They earn the effort in two situations: when product images sit on a separate CDN domain, and when images load through JavaScript in a way ordinary discovery may miss. Where images sit in the HTML on crawlable product pages, standard discovery already covers it. On the page side, the same crawl will tell you which indexable pages are missing from your sitemap.

Auditing Broken, Slow or HTTP Images Across Thousands of URLs

Three problems pile up on large catalogs and none of them surfaces in a normal report.

Mixed content, meaning images still referenced over http:// on an HTTPS site, the leftover from a migration declared finished too early. External images from domains you no longer control, which accumulate through supplier feeds and old campaigns and fail silently when the other side reorganizes. And dimension mismatches, where declared width and height do not match what is served, which is a layout shift in waiting.

A crawl with image processing enabled covers all three. The Images table separates internal from external images, the Content report shows average images per page, and the content problems list includes “HTTP Images on page (mixed content)” as a named issue. To isolate insecure images directly, filter Full src to contain http://, which is the method in our guide to identifying and fixing HTTP images. Filters on height and width surface the dimension mismatches.

Audit every product image on your site in one crawl

Audit Your Image

Structured Data for Product Images

Structured data is where the image stops being a picture and becomes a field that a shopping engine can use. It is also where the requirements differ by surface, which catches teams out.

Image Fields in Product Schema

In Google’s merchant listing documentation, image is a required property. Google recommends multiple high-resolution images with a minimum of 50,000 pixels when multiplying width by height, in 16×9, 4×3 and 1×1 aspect ratios, and requires that image URLs be crawlable and indexable.

That 50,000 pixel floor works out at roughly 224 by 224 for a square image, which is very permissive. The binding constraint sits elsewhere, in the feed.

The Merchant Center change is the one with a date attached. Feeds that currently pass with a 250 by 250 apparel image will start producing warnings and then disapprovals, and Merchant Center also rejects images with watermarks, promotional text or borders, or where the product occupies less than 75% of the frame.

Three mistakes make the field unreliable at scale: image URLs that need a session or a referrer and so cannot be fetched, a single image in the array where the page shows a gallery, and markup referencing an image the page does not display, which Google’s guidelines rule out directly.

How Structured Images Feed Google Shopping and AI Answer Engines

This closes the loop on the AI section above. When a generative engine shows your product, the image comes from a URL in a feed or in structured data, and the decision to show that product came from the text around it.

The work splits cleanly. Eligibility comes from product data that is complete, current and machine readable, which is the same job as feed quality. Conversion comes from an image that is fetchable without authentication, sized for the strictest surface you sell through and honest about the product. Our guide to optimizing eCommerce product pages for SEO covers the wider picture this sits inside.

Product Variants and Duplicate Images: The eCommerce-Specific Trap

A catalog with colour and size variants produces the same image at many URLs by design. Four sizes of one jacket share a photograph. The question is whether that matters, and the documented answer runs against the instinct.

Google’s image best practices, updated in March 2026, ask you to “consistently reference the image with the same URL, so that Google can cache and reuse the image”. Serving the same photograph from one stable image URL across variant pages is the recommended behaviour, not a risk. Generating a unique image URL per variant page for the same file is the thing to avoid.

There is no Google documentation on image canonicalization or on any penalty for reusing a photograph across variant URLs, so treat claims in either direction with suspicion.

What is documented sits one level up, in the structured data. In Google’s product variant guidance, image is a property of the individual variant rather than of the parent ProductGroup, and Google’s own examples give each variant its own file: coat_small_green.jpg, coat_small_lightblue.jpg. The implication is clear enough for a decision.

The practical split follows from that. Where the variant changes what the product looks like, colour above all, unique photography earns its place and the markup expects it. Where it changes something invisible, size or capacity, reuse the same image URL and spend the photography budget elsewhere. In both cases the alt text describes the variant on that page, which is the one place where sharing a string is always wrong.

Final Thoughts

Alt text stopped being an accessibility checkbox some time ago. It is one field in a chain that decides whether a machine can describe your product to someone who never sees the page, and the same holds for the file name, the format, the feed and the schema.

The uncomfortable part is that none of this fails loudly. A catalog of thirty thousand images with no alt text looks perfect to everyone who can see. A product page that lazy-loads its own hero image scores fine in every report that does not measure it. The failures are quiet and cumulative, which is why the answer to how to optimize images for eCommerce at catalog scale is to audit the whole set on a schedule rather than to write better alt text on whichever pages somebody happened to open.

FAQ

Does image SEO affect Shopping and Merchant Center feed approval, not just organic rankings?

Yes, and Merchant Center is stricter than Search. From January 31, 2027 the minimum rises to 500 by 500 pixels across all categories, with warnings since April 2026. Watermarks, promotional text, borders and images where the product fills less than 75% of the frame cause disapprovals regardless of size.

Do decorative images like icons and banners need alt text at all?

They need an alt attribute, and it should be empty. An empty alt=”” tells a screen reader to skip the image, which is the right outcome for decoration. Omitting the attribute is what makes some screen readers announce the file name instead. Keep descriptive alt text for images that carry information a shopper needs.

Is reusing the same product photo across marketplaces an SEO risk?

Not for your own rankings. Marketplaces host their own copy on their own domain, so it is a separate URL under their control rather than a duplicate on yours. The risk is commercial: identical photography makes your page look interchangeable with the marketplace listing that often outranks it. Distinct lifestyle or detail shots are worth more here than any indexation concern.

Do AI-generated product images need different alt text treatment than photographs?

The alt text follows the same rules, since it describes what is shown. The caution is accuracy: an image showing a colour, texture or configuration the real product does not have creates a mismatch between page and item, and Merchant Center prohibits images that misrepresent colour, pattern or material. Disclosure rules vary by market and are worth checking separately.

How often should the image catalog be re-audited for missing alt text?

Tie it to the pipeline rather than the calendar. Run a full image crawl after every PIM change, feed integration, template release and platform migration, since those are the four events that strip the attribute. Between them, a monthly crawl catches drift from ordinary product uploads. Treat it like any other recurring technical check, because the work to optimize images for eCommerce only holds if it survives the next sync.