A retailer uses a vision-enabled chat model to generate alt text for product and site…

How it works

A vision model can draft alt text, but the rules come from Microsoft's accessibility style guide, and your prompt and post-processing should enforce them. The guide's core idea: accurately but concisely convey the image's purpose. Good alt text isn't about describing every detail; it requires interpretation and judgment.

The two rules in the correct answers:

1. Length. Limit alt text to 150 characters. If an image is complex, put a detailed description in the surrounding text or in linked content. A post-processing check on length is easy to automate.
2. Decorative images. Images that are purely decorative don't need alt text. For decorative images in web content, use null alt text (alt="") instead of omitting the alt attribute. So the pipeline should be able to mark an image decorative and emit an empty attribute, not drop it.

Other rules worth putting in the prompt: begin with a capital letter, end with a period where practical, include text embedded in the image if the surrounding content doesn't already have it, and don't repeat the caption.

The other options, and when you'd use them

• Start every description with "Image of": The guide says not to start alt text with a general word such as "Image," because screen readers already announce images as images. Start with what it is instead, such as a photograph, diagram or screenshot.
• Describe every visible detail: Writing good alt text isn't about describing every detail. Exhaustive descriptions also blow past the 150-character limit.
• Use the file name when confidence is low: The guide says not to use the file name of an image as alt text. A low-confidence result should go to human review, not fall back to a file name.

Exam tip

Alt text rules to remember: purpose over detail, 150 characters max, alt="" for decorative images, no "Image of", no file names.

References

• Alternative text (alt text)