Ask AI what the image does before asking it for alt text

‘A black magnifying glass on a white background’ can be an accurate description and a dreadful search button. Someone trying to use the page needs to know what happens when they press it. W3C's existing images tutorial makes that distinction explicit: an image used as a link or button needs text describing its function. Its search-button example uses ‘search.’ The same guide says the author determines the alternative text from the image's use, context and content. That gives me a better brief for an AI assistant than ‘describe this picture.’ Supply the nearby words, say whether the image is clickable, and name its destination or action. Ask for a draft that serves that purpose. A logo standing alone as a home-page link needs a destination; a logo beside text already identifying the same destination can need an empty alt attribute to avoid repetition. W3C documents both cases. Those are existing accessibility examples, not an AI tool I've tested. My recommendation is to review the draft where it will appear, rather than approve it in an image library and reuse it everywhere. The assistant can recognize the magnifying glass without knowing what your button does.