Using AI to describe thousands of images, across 120+ brands

A global drinks company with well over a hundred brands had an accessibility problem hiding in plain sight. Every product shot, campaign image and banner on its websites needed alt text. That is the short written description a screen reader reads aloud to a blind visitor. For a few dozen images, it is an afternoon's work. Across 120+ brands, 180 markets and several languages, doing it by hand had become impossible.

The brief

Find a simple, repeatable way to write accurate alt text for thousands of images, in every language the business publishes in. Without hiring an army of copywriters. And without churning out bland descriptions that fill the box but help nobody.

The approach

• Tried the technology first, then rebuilt the process around it. Bolting AI onto the old way of working rarely holds up.
• Piloted it on one brand, Guinness. Small enough to learn from properly, big enough to be a real test.
• Spot-checked the results during the pilot, then fixed what those checks exposed.
• Rebuilt the quality check around what actually mattered. Not whether the box had been filled in, but whether each description was accurate and good enough to publish.
• Rolled it out across the rest of the business once the process and the checks were holding up.

Why it matters for a smaller business

Very few small businesses have thousands of images to describe. But most have a version of the same problem. A job that genuinely matters, that nobody has time to do properly, and that quietly gets skipped. The lesson transfers directly, because the tool was only half of it. What made it work was redesigning the process around the tool, proving it on one small pilot, and checking whether the results were actually any good. Same order of operations, whatever the size of the business.