Accessibility & semantics

The decorative image you are about to hide probably should not be there at all

The decorative image you are about to hide probably should not be there at all

Put alt="" on an image and you have made a decision on behalf of everyone who cannot see it. The sighted reader gets the picture, the mood and the small nudge it gives the page. Someone listening to that page gets nothing, and no hint that anything was there. Sometimes that is the right call. Adrian Roselli's new post argues it is the right call far less often than authors assume, and his fix sits a step before alt text: go back to whether the image should exist at all.

His opening line is the whole thesis. Decorative images "aren't a thing", he writes, "at least not as far as people are concerned."

Someone chose that image

Roselli's argument starts from intent. Nobody adds an image to a page, an email or an app by accident. If the author wanted to set a mood or tie the piece to a theme, then the person who can't see the image "would likely benefit from knowing that mood, having that connection made." That is content, and content needs a text alternative.

If the reason was more mundane, say filling white space, Roselli calls that a design gap. It belongs in the layout. Leave it as an image and the author is, in his words, "unintentionally sending a message to those who can see it." One audience gets a signal and the other gets silence, and that is the inequity I care about here.

He is explicit about scope: images in narrative content that are not also iconography and are not part of interactive controls. Icons and images inside buttons and links follow their own rules, and he is not arguing about them.

Two questions that sort the pile

Instead of another taxonomy, Roselli offers what he calls over-simplified questions:

  • "If the decorative image was removed completely, would any readers care?" If no, remove the image. If yes, it is not decorative.
  • "If the decorative image was replaced with a picture of poo, would you as the author care?" Same answers, same outcomes.

The second one is crude on purpose, and it works. It catches the author who insists an image is meaningless but would be mortified to see it replaced. You cannot hold both positions. If the picture matters to you, it carries meaning, and that meaning is owed to every reader.

Two new branches on a familiar tree

Roselli adapted the WAI alt text decision tree and added steps to it. The familiar branches are still there. Does the image contain text? Is it in a link or a button? Does it contribute meaning? One of those branches still ends in an empty alt: an image whose content is redundant to real text nearby.

<img src="photo.jpg" alt="">

That is the case the empty attribute exists for, and Roselli's tree keeps it.

His changes come at the end. They start once you answer that an image is purely decorative or not intended for the user. At that point his version asks the removal question, then the poo question. If readers would care, start over. If you would not care about a swap, he writes "You should care", and you start over again. The only exit that skips a text alternative is the bluntest one: "Remove the picture and save yourself the hassle."

The tree on his page is interactive, answered with Yes and No choices. He has been here before: his related links include his own 2023 piece, "An alt Decision Tree Using Only :has()".

Absolutist on purpose

Roselli labels this "an absolutist take" himself, and he says why: "sometimes it's necessary to take the absolutist position to force someone to justify their own reasoning." I think that framing is the useful part. He is not claiming that every empty alt in production is a failure. He is asking that each one be argued for.

He is also not alone. His related links include Nicolas Steenhout's "It's time to retire the concept of decorative images" from 28 September 2026, WCAG 3 issue #878 titled "Definition of decorative images update" from 29 September 2026, and Jens Oliver Meiert's "On the Decision Underlying Decorative Images" from 8 October 2026. That is three posts in under two weeks, plus an issue filed against WCAG 3 about the definition itself. The question has reached the standards process as well as the blogs.

Where I land

Roselli is right, and the change he asks for is cheap. An empty alt should be the end of a decision that someone can explain, not a reflex. Most images that fail his two questions do not need better alt text. They need deleting, and a deleted image locks nobody out.

Before your next PR adds an image to an article or page body:

  • Ask why it is there. If the answer is mood or theme, write that into the alt.
  • If it only fills space, fix the layout and drop the image.
  • Run both of Roselli's questions. Either one can send you back to the start.
  • Keep alt="" for images that repeat real text nearby, and be ready to say so in review.
  • Leave icons and images inside controls to their own rules. They are outside this argument.

Source: Adrian Roselli (adrianroselli.com)

Related
Accessibility & semantics

The `headers` attribute has quietly stopped earning its place

Adrian Roselli retested the HTML `headers` attribute against the current crop of screen readers and browsers, and the answer is short: plain, honest table structure carries the meaning by itself, and the attribute is largely useless.

September 10, 2026
Accessibility & semantics

Whether a dialog closes on outside click is a task question, not a style question

Adrian Roselli sorts modal close behaviour into three buckets — action required, no action required, and weird patterns — and argues you should pick a light-dismiss policy based on what the user came to do, not on how the dialog looks.

August 13, 2026
Accessibility & semantics

A primary-nav flyout built on `popover`, not `aria-expanded`

Adrian Roselli has rebuilt his 2019 disclosure-widget primary nav on the native HTML `popover` attribute. The rebuild is a clean map of what the platform now carries for you, and where you still need ARIA.

July 9, 2026