Image Alt Text at Scale: Writing Descriptions for 5,000 Product Images Without Going Insane

Run a crawl on almost any store with more than a few hundred products and the same line shows up in the report: thousands of images with empty alt attributes. It's the most ignored issue in e-commerce SEO, and it gets ignored for an understandable reason — nobody wants to hand-write five thousand image descriptions, and nobody believes it moves the needle enough to justify the pain. Both halves of that are wrong. The needle moves, and the pain is a process problem, not a volume problem.

Alt text does three jobs at once. It's the text screen readers speak aloud to visually impaired shoppers, which makes it an accessibility requirement before it's an SEO tactic. It's what Google Image Search uses to understand what a photo shows, which is a real traffic channel for product-heavy niches — furniture, apparel, parts, anything people shop with their eyes. And it's what renders when an image fails to load, which on flaky mobile connections is more often than you'd like. A store where most product images carry no alt attribute is invisible in image search and hostile to a slice of its customers, simultaneously.

Filename and alt attribute are two different signals

The first confusion to clear up: the image filename and the alt attribute are separate signals, and most stores botch both. The filename is what Google sees in the URL — IMG_4823.jpg tells it nothing, while walnut-standing-desk-60-inch.jpg tells it plenty. The alt attribute is the description in the HTML. They should agree with each other and with the page they sit on, because agreement is what makes the signal believable.

Filenames are the harder half to fix retroactively, because renaming an image changes its URL — every rename needs the old URL redirected or the image loses whatever image-search equity it had. That's why the practical order of operations on an existing catalog is: fix the alt attributes everywhere, and fix filenames only going forward, as new images are uploaded. Retroactive mass renames create more breakage than benefit on most stores.

What good alt text actually looks like

The rules are short, and every one of them exists because stores break it at scale:

Describe what the image shows, specifically. "Walnut standing desk with dual-motor frame, 60-inch top" beats "desk" and destroys "buy standing desk best price standing desk sale."
Front-load the distinguishing detail. In a gallery of eight photos of the same desk, the alt text should say what's different about each shot — the cable tray close-up, the height-range side view — not repeat the product name eight times.
Keep it under about 125 characters. Screen readers get clipped or tedious past that, and Google doesn't reward essays.
Skip "image of" and "photo of." The tag already says it's an image.
Leave decorative images empty — deliberately. A divider graphic or background texture should carry alt="", an explicit empty attribute, so screen readers skip it. Empty-on-purpose is correct; missing-by-neglect is not.

The failure mode at scale is templated garbage: a plugin that stamps every image with the product title, so all eight gallery images say the same thing and every variant swatch says "T-Shirt." That's technically "alt text coverage" and practically worthless — it's the image equivalent of duplicate meta titles.

Alt text, WebP, and your schema

Two adjacent technical details ride along with any serious alt text project. First, format: if you're touching every image anyway, it's the natural moment to convert to WebP, which cuts file size 30–40% against JPEG and directly improves Largest Contentful Paint. Alt attributes carry over untouched — the alt lives in the HTML, not the file.

Second, structured data: product schema has its own image requirements, and Google is picky about them. The image property should reference high-resolution originals (Google asks for images that are at least 50,000 pixels when multiplied width by height, in multiple aspect ratios ideally), and those URLs must be crawlable — not blocked by robots.txt, not lazy-loaded in a way that hides them from the parser. We regularly find stores whose visible images are fine but whose schema points at thumbnail-sized crops, quietly disqualifying them from image-rich results. Alt text, schema image references, and the rendered gallery should all describe the same product the same way.

How to do 5,000 without going insane

The reason this job feels impossible is that people picture it as 5,000 independent writing tasks. It isn't. A product catalog has structure, and the structure does most of the work:

Inventory first. A full crawl produces the actual list — which images are missing alt entirely, which have templated duplicates, which are decorative and should be empty. The real number of images needing hand-written descriptions is always smaller than the raw count.
Work by template, not by image. Category thumbnails, primary product shots, gallery angles, variant swatches, blog images — each group has a natural description pattern built from data the catalog already holds: product name, key attribute, shot type. The pattern gives you a correct, specific baseline; human judgment then sharpens the images that matter most.
Prioritize by traffic. Best-sellers and top category pages get the most careful treatment. The long tail gets accurate, pattern-based descriptions. Perfect uniformly is the enemy of done correctly.
Verify on the rendered page. What the media library says and what the theme ships are not always the same thing — some themes drop alt attributes on gallery images no matter what's in the database. The crawl after the fix is the only proof that counts.

This is exactly the kind of job that rewards a systematic pass over a heroic one — the same logic as any bulk product SEO work: structure the work by pattern, execute in one sweep, verify with a crawl. Our team runs full media library cleanups this way — alt attributes, filenames on new uploads, decorative images zeroed out, schema image references reconciled — across WordPress and BigCommerce catalogs.

What to expect afterward: image search impressions climb over the following weeks as Google recrawls the catalog, accessibility audits stop flagging the store, and — the part nobody predicts — product pages themselves get a modest lift, because the alt attributes add exactly the kind of specific, relevant on-page text a thin product page was missing. Five thousand images is not five thousand tasks. It's one inventory, a handful of patterns, and a verification crawl.