Alt text is the photo, in words

Every product photo on a listing is there to answer a question — what colour it really is, what it is made of, how big it is, how it looks in use. A shopper who can see the photo gets the answer without reading a word. A shopper who cannot gets whatever was typed into one small field attached to that image, and nothing else.

That field is the alt text. A screen reader reads it aloud, a browser shows it in place of an image that failed to load, and it is the only part of a photograph that exists as words. Writing it well costs a few minutes per product. It is also easy to skip, because the person writing it can always see the picture.

The product name is not a description

The easy fill is the product's name: "Aurora pendant". It is not wrong, exactly — it is redundant. The title said it a few centimetres higher up. A shopper listening to the page hears the name, then hears it again for every image in the gallery, and learns nothing the photos were added to teach.

Write instead what a sighted shopper learns by looking. For the first image, that is usually the whole object, its colour and its material: "Aurora pendant in brushed gold, a teardrop of green glass on a fine chain." It is the sentence a friend would say if you asked them what the photo shows. That is the whole test.

Each photo says what the others do not

A good gallery is a sequence, not a pile: each image earns its place by showing something the one before it could not. The alt text should follow the same rule. If the second photo is a close-up of the clasp, its alt text is about the clasp — how it fastens, what it is made of — not a second description of the whole pendant.

Written that way, a set of alt texts reads like a short tour of the product. Written as six copies of the name, it reads like a stutter. And if you cannot say in words what one image adds, that is worth knowing too: it may be the image to cut.

Name the colour the way the listing does

Colour is the easiest place for alt text to drift from the rest of the page. The variant is called "sage", the alt text says "green", and a shopper choosing between colourways by ear cannot tell whether those are the same thing. Use the variant's own name, and add a plain gloss when the name is poetic: "sage, a soft grey-green".

If each colourway has its own photos, each set names its own colour. That sounds obvious, and it is the first thing lost when one block of text is copied across every variant.

Put scale into words — real measurements only

A photograph does not carry scale on its own, and a listener gets even less of it. If an image shows the product against something of known size — in a hand, on a wrist, beside a mug — say so, because that comparison is the reason the image exists. If the listing states a measurement, the alt text can repeat it.

What it must not do is guess. "About the size of a coin" helps only if it is true, and nobody re-reads alt text once it is written. Take numbers from the product's own specification, or leave them out.

Say what is not in the box

Lifestyle photos are staged. The pendant is shown on a model in a linen shirt, beside a ceramic dish, on a table that is not for sale. A sighted shopper reads those as props without thinking about it. A listener hears a list of objects and has to guess which of them they are buying.

So describe the product first and the setting as setting: "Aurora pendant worn on a fine chain, styled with a linen shirt." If a prop could plausibly be part of the order — a stand, a pouch, a gift box — and it is not, say that as well. A photo is a promise about what arrives, and its alt text is part of the promise.

Generated photos need it just as much

This matters more, not less, for images nobody on your team shot. A generated marketing set can put your product in a scene nobody arranged, with props nobody chose. Nitaris hands over the photos and the film; it does not write their alt text, and it should not guess at what your buyer needs to hear. That part is yours, and it takes about a minute per image.

A film needs words too

A video has no alt attribute of its own, so its words have to live beside it. A Nitaris film carries no dialogue unless you add a voiceover, and if you do, that line belongs in the description too, because most people meet the film with the sound off. Everything else it communicates is visual, and all of it is lost to a shopper who cannot see it. One sentence in the description does the job: what the film shows, in the order it shows it — "The pendant turns slowly under a single light, catching green through the glass, then settles to show the clasp."

What to leave out

• "Image of" and "photo of". Every word before the product is a word the listener waits through.
• Keywords. Alt text written for a search engine reads like a list of search terms, because it is one, and the shopper it was meant for has to sit through it.
• Marketing copy. "Stunning" and "perfect for any occasion" describe your hopes, not the photo.
• File names and SKU codes. If your store fills an empty field with something like the file name, then leaving it empty is not the neutral choice it looks like. Check what yours does.
• Decoration dressed up as information. A background texture or divider that shows nothing about the product should be marked as decorative, if your store allows it, so it is skipped rather than described.

Check it by listening

The test needs no special knowledge. Turn on a screen reader, open one of your own product pages, and listen to the gallery with your eyes closed. If you can picture the product — its colour, what it is made of, how big it is, what comes with it — the alt text is working. If you heard the product's name six times, you know what to fix first.

The short version

Alt text is the photo, in words, for a shopper who cannot see it. For each image, write what that image shows that the others do not: the colour by its variant name, the material, the scale when there is a real reference, and the setting as setting rather than as part of the order. Give the film one sentence beside it. Then listen to the page once, with your eyes closed.