Alt text for AI-generated images: do you say it’s AI, and what do you describe?

Write alt text for an AI-generated image the way you would for any other: say what it shows and why it is on the page. Mention that it is AI-generated only when that fact is part of what the reader needs to know.

That follows WebAIM’s advice that the type of image, such as photo, illustration, or painting, belongs in alt text only when it is important to the content. “Generated by AI” is a kind of image type, so the same test applies. The rest of this post covers when it passes and what to write when there was never a real scene to photograph.

When to say it’s AI

Ask whether the origin changes how someone understands the post. A launch announcement for an image generator, a case study on AI imagery, or a comparison of generated and photographed versions all qualify. A generated texture behind a quote card, or a concept illustration for a newsletter about budgeting, does not. The origin changes nothing about what the image communicates, so leave it out.

Disclosure is a separate question. Whether you must label AI imagery depends on each platform’s current rules, your client contracts, and your own editorial policy, so check those directly. Alt text is a poor place to meet any of them, because most readers never see it. Put a disclosure where everyone can read it: the caption, the post text, or the platform’s own label if it has one.

What to describe when nothing is real

A generated image has no original scene, so there is nothing to be faithful to except the image itself. Describe what is visible and what it is doing on the page. The W3C’s tutorial on informative images says the text alternative should convey the meaning the image carries, which is usually not a literal inventory. Its example is a stock photo of a happy family used to suggest a family-friendly company. The alternative is “We’re family-friendly,” and the tutorial says the individuals should not be identified.

That rules out the prompt. A prompt records what you asked for, not what came back. Put the most important element first, as the W3C’s image tips suggest, and keep it to a phrase or a sentence. A hero image of glowing neural pathways for a post on AI search is fine as “Abstract illustration of glowing blue lines branching like a network.”

If the image contains garbled lettering or an obvious glitch, fix it or regenerate it before publishing. Do not transcribe nonsense words into the description. The same habit applies to alt text for stock photos: describe the picture on the page, not the story behind it.

Realistic people deserve extra care

A photorealistic generated face can pass for a customer or an employee. Do not give it a name, a job title, or a quote in the alt text, and do not describe it in a way that implies a real person. If readers could reasonably take the portrait for a real customer, that is a case where saying “AI-generated” matters. Our notes on images you didn’t take cover the broader habit of describing what is there without claiming more.

A three-question check

Before you publish, ask what the image is for, whether its origin changes the meaning, and whether nearby text already says it. If the image only adds atmosphere, it may be decorative rather than informative, and blank alt text is the right call.

If you do mention the origin, use it as a modifier, as in the third example above. Starting with “Generated image of” falls into the “image of” habit screen readers make unnecessary. The foundation for all of this is in why alt text matters.

SnipAlt can draft a description of any generated image on your screen: snip it with Ctrl+Shift+A (⌘⇧A on a Mac), then edit the draft to match what the post needs, including whether the origin belongs in it.