AI tools can now look at a picture and write a description of it in seconds, which makes them attractive for the millions of web images that have no alternative text. But the best known guidance on alt text says that “algorithms alone cannot understand what an image means within the context of the overall page.” Good alt text depends on why the image is on the page, and an automatic description of what is visible often misses that. Used as a first draft that a person reviews, AI can save time; used as a replacement for judgment, it can add noise for the people who rely on screen readers.

What alt text is for

WebAIM defines alternative text as “a textual substitute for non-text content in web pages.” Screen readers read it aloud, braille displays show it, and browsers show it when an image fails to load. Search engines also use it. The guidance is that every image should have an alt attribute, even if its value is empty (alt=""), because an empty value tells assistive technology to skip a decorative image, while a missing attribute may lead a screen reader to read the file name.

WebAIM also stresses that alt text is about both content and function. A photo of an astronaut inside a link needs alt text that conveys where the link goes, while the same photo next to a caption naming her may need none at all.

Where AI helps

AI image description is useful in several real situations:

  • Personal assistance. Apps such as Be My Eyes, with its Be My AI feature, and Microsoft’s Seeing AI describe photos for blind and low-vision users on demand. Here the user decides what to ask and can follow up. We cover these in our overview of AI and accessibility.
  • Drafting for publishers. A model can propose a description of a product photo or a screenshot that an editor then corrects.
  • Triage. Automated tools can flag images with no alt attribute so people can fix the most important first.
A man wearing headphones working with a braille book and a computer in a library
Illustration: assistive technology in use; screen readers and braille displays rely on alt text. Photo: Mikhail Nilov / Pexels

Where AI misses

Context is the hard part. The WebAIM article walks through one photograph and shows that the right text changes with use. As a plain illustration in an article, the person’s identity may be in the caption, so empty alt text can be enough. As a link, the alt must describe the destination. An automated captioner looking only at the pixels will usually produce the same description in every case.

Other common problems follow from how these systems work:

  • Describing the obvious. A generic “two people shaking hands” tells a screen reader user nothing the surrounding text does not already say. WebAIM’s handshake example concludes that empty alt text is best when the image conveys no important content.
  • Redundancy. If the caption or nearby paragraph already states the information, repeating it makes reading slower.
  • Charts and diagrams. A model may summarize the look of a chart but miss the values. WebAIM advises that complex images such as charts and maps need an equivalent elsewhere, such as an adjacent data table, with only a succinct alt on the image.
  • Text in images. Rasterized text should be real text where possible. If not, the alt needs to reproduce it accurately, and recognition errors matter.
  • Unverified details. Models can state a detail that is not in the image. Because the output reads smoothly, errors are easy to miss. Always check names, numbers and claims.
  • Identity and sensitive traits. Guessing a person’s identity, age or other personal traits from appearance is risky. Use the identity the publisher knows from the story instead.

A quick process for reviewing AI drafts

  1. Ask what the image is doing. Is it informing, linking, decorating, or showing data?
  2. Decide if alt is needed. If the image is purely decorative or its content is already in the text, use alt="". WebAIM’s test: if you could not use the image, what would you put in its place? If the answer is “nothing,” empty alt is probably right.
  3. Generate a draft if it helps, and then delete everything the surrounding text already covers.
  4. Keep it short. A couple of sentences is the practical ceiling; longer explanations belong in the body text or a linked page.
  5. Describe function for linked images. The text should say where the link goes or what a button does.
  6. Avoid “image of” and “link to.” Screen readers already announce the element type, and WebAIM advises against such phrases.
  7. Do not put captions in alt. WebAIM notes that an image inside a <figure> still needs its own alt text and that the information should not be duplicated through the caption.
  8. Test with a screen reader or ask a user who relies on one when the page is important.

Writing better alt text yourself

Start with the purpose. For news photos, name the subject and what they are doing: “Passengers queue at an airport check-in desk” is more useful than “people”. For product shots, give the features a buyer would see. For charts, state the finding in the alt and put the data in a table. For logos, use the organization’s name, since WebAIM says words like “logo” usually add nothing. Our own site follows the same rules: image descriptions say what is visible, not the headline.

What to watch next

  • Accuracy of AI image descriptions on charts, screenshots and text-heavy images.
  • Publishing tools that suggest alt text but require human approval before saving.
  • Regulation and accessibility standards that treat missing alt text as a compliance issue.
  • Feedback from screen reader users about which automatic descriptions help and which clutter.

Key takeaways

  • Alt text must reflect the image’s purpose on the page, not just what it shows.
  • Every image needs an alt attribute; decorative images use an empty one.
  • AI is useful for drafts and triage but needs human review.
  • Charts and complex images need a text equivalent beyond a short alt.
  • Never repeat the caption or add phrases like “image of.”

Frequently asked questions

Can AI write alt text for me?

It can draft it, but you should edit the result for context, accuracy and brevity before publishing.

When should alt text be empty?

When an image is purely decorative or its content is already fully presented in the surrounding text. An empty alt attribute tells screen readers to skip it.

How long should alt text be?

Short: usually a phrase or a couple of sentences. Longer descriptions of complex images should live in the page text or on a linked page.