Alt text for e-commerce product images

E-commerce is where alt text is simultaneously most valuable and most neglected. Valuable, because product images are the product page — a shopper who cannot see them cannot buy. Neglected, because a catalogue has forty thousand images and nobody has ever been given four weeks to describe them.

The result is the familiar compromise: alt="product-image-1", forty thousand times.

What alt text does on a product page

Three jobs, and they are not equally important.

It lets a blind shopper buy the thing. This is not an abstraction: a screen reader user landing on a product page with six unlabelled images has no idea whether they are looking at the front, the back, the fabric detail or a lifestyle shot. Alt text is the difference between a purchase and a bounce.

It is a legal exposure. Accessibility lawsuits against retailers are routine, and image alt text is among the first things audited. That is not a reason to write good alt text, but it is often the reason a budget finally gets approved.

It helps image search. Product images rank in Google Images, and image search converts — someone searching for a picture of a thing is frequently trying to buy the thing. Alt text is one of the strongest signals about what an image depicts.

A formula that works

For product photos, freeform advice (“describe the image”) is less useful than a pattern:

[Product name] — [what this specific shot shows]

So, for the third image in a listing:

• ❌ alt="shoe"
• ❌ alt="running shoes buy online cheap trainers sports footwear"
• ✅ alt="Aurora running shoe in cobalt blue — side view showing mesh upper"

The product name anchors it. The second half is what makes each image in the gallery different — and it is the half everyone skips, which is why product galleries so often ship six images with identical alt text.

For variant-heavy catalogues, include the attribute that distinguishes the variant: the colourway, the size, the material. That is exactly the information a shopper cannot get any other way.

The mistakes, ranked by how common they are

Identical alt text on every image in the gallery. The whole point of a gallery is that the images differ. If the alt text does not differ, five of the six images are noise.

Keyword stuffing. It has not worked for SEO in over a decade, and it turns a screen reader into a spam machine. If your alt text reads like a search query, rewrite it.

Describing the model instead of the product. “Smiling woman in a park” tells the shopper nothing about the jacket they are trying to buy. The product is the subject; the model is context.

Alt text on decorative images. Category banners, spacer graphics, the little icons next to your delivery promise — most of these should have an empty alt="" so screen readers skip them. See how to write good alt text for where that line sits.

Ignoring text baked into the image. If your promo image says “20% off this week”, that text is the content — and a shopper using a screen reader currently does not know there is a sale.

Forty thousand SKUs and no time

Everything above is easy to agree with and impossible to action by hand at catalogue scale. So the real question is not “what is perfect alt text” but “what do we do about the forty thousand”.

The answer is to stop treating them equally.

Tier 1 — write by hand. Best sellers, hero images, the top of every category. A few hundred images. This is where revenue and legal risk concentrate, and human writing pays for itself.

Tier 2 — generate a draft, then edit. The long tail. An AI pass gives every image a factual description in minutes, and a human skims for nonsense. Not perfect; vastly better than what you have now.

Tier 3 — templated. Where you have structured product data — and in e-commerce you always do — you can generate serviceable alt text mechanically: {product_name} — {colour}, {shot_type}. Inelegant, and better than product-image-1 on every axis that matters.

Most catalogues need all three. The failure mode is attempting tier 1 quality across all forty thousand, achieving nothing, and shipping with none.

Where AI actually helps

An image model is good at the literal layer: it will tell you the shot shows the side of a blue shoe with a mesh upper. That is genuinely the sentence you were never going to write forty thousand times.

What it does not know is your product name, your colourway, your SKU. It knows it is a shoe; it does not know it is the Aurora in cobalt. Which points at the workflow that actually works:

1. Run an AI pass to get the descriptive half of every image.
2. Prefix the product name and variant from your product database — you already have it.
3. Spot-check, fix the nonsense, ship.

That is a couple of days for a catalogue that has been shipping empty alt attributes for five years.

PicsTag does step 1 in the browser: drop in the images, or feed a CSV of your image URLs with the product ID in the asset_id column, and export a file that joins straight back onto your product data on that ID. Nothing is uploaded — which, for unreleased product photography, is usually the whole ballgame.

The one metric to watch

If you want to know whether any of this worked, do not measure alt text coverage. Measure image search impressions in Search Console before and after, and run a screen reader down one of your own product pages.

The first number tells you whether Google understood your images. The second tells you whether a customer could have bought from you. Coverage tells you neither.