Should AI alt text describe the picture or what clicking it does?
Imagine reaching a button with a screen reader and hearing ‘a black printer on a white background.’ You know what the icon looks like. You still do not know whether it prints this page, opens a printer shop or shows setup instructions. W3C’s existing image guidance uses ‘Print this page’ for a printer icon that activates printing. It also gives an empty alt attribute to a logo beside link text that already names the destination, so the reader does not hear the same information twice. These are established accessibility examples, not findings about a particular AI tool. For an AI alt-text helper, I would want the draft made with the link destination, button action and nearby text in view—not just the uploaded picture. That is a proposed requirement. A fluent description of the pixels can be the wrong thing to say here. Before accepting a batch of generated labels, try navigating one ordinary task by its announced links and buttons. Can you find where to go and predict what selecting it will do? Where should an AI helper ask for context instead of guessing a label?
Comments
Give the helper the same magnifying-glass icon in two places: one opens product search, the other enlarges a product photo. If both come back as ‘Search,’ the shop owner has a batch of labels to repair. I’d try that pair before handing over the whole image folder. A helper that only accepts pictures should ask what each button does; otherwise you still have to supply the missing context, just after it has written the wrong answer.