Alt text

An image that conveys information needs alt text describing what it conveys, in context.

An image that is purely decorative — a divider, a texture, an icon beside a label that already says the same thing — takes `alt=""`, which tells assistive technology to skip it entirely.

Omitting the attribute is different from setting it empty. With no attribute, a screen reader has nothing to work with and commonly reads the filename, which is worse than silence.

An image inside a link with no other text must describe the link's destination, because it is the only thing the user has to act on.

What the image communicates here, not an inventory of what is in it. The same photograph needs different alt text on a news article and on a product page.

As short as the meaning allows. A sentence is usually enough; a paragraph belongs in the page, not in an attribute.

No "image of" or "picture of". The technology already announces that it is an image, so the prefix is repeated on every one.

Text that appears in the image must be in the alt text, since it is otherwise unavailable to anyone who cannot see it.

For a chart or a diagram, the alt text gives the conclusion and the page gives the data. A description of the axes helps nobody.

A caption, in a `figcaption`, is shown to everyone and adds context the image does not carry on its own.

Alt text replaces the image for those who cannot see it and is shown to nobody else.

When a caption fully describes the image, the alt text can be empty rather than repeating it — otherwise a screen reader user hears the same sentence twice.

The `title` attribute is neither: it produces a tooltip on hover, is unavailable on touch devices, and is announced inconsistently. It is not a substitute for `alt`.

Missing attributes are mechanically detectable: an `img` element with no `alt` is a finding without exception.

The filename-as-alt-text pattern — `alt="IMG_4832.jpg"` — is equally detectable and is always wrong.

Very long alt text, beyond a couple of hundred characters, usually means content that belongs on the page.

Whether the text is accurate requires looking at the image. This is the part no automated check does, and it is the part that decides whether the attribute is useful.