AI Summary: Alt text is an accessibility requirement and durable context. Use context-specific alternatives, empty alt for decoration, action text for functional images, and full equivalents for charts; alt text does not control a vision model or guarantee image retrieval, ranking, or AI citation.
Start with purpose and context
W3C WAI says images need text alternatives for their information or function. Its decision tree starts with purpose and context, not a fixed formula. [1] [2]
Write the shortest alternative that preserves the image's job. A photo, button icon, chart, and decorative flourish need different treatment. Do not copy filenames, stuff keywords, or describe irrelevant detail.
Choose the right alternative
For informative images, describe essential information. For linked or functional images, describe the destination or action. For decoration or redundant imagery, use alt="". [1] [2]
For text in an image, repeat the words as real text where possible. For graphs, provide a text equivalent. [1]
| Purpose | Implementation | Avoid |
|---|---|---|
| Informative | Essential content in alt | Keyword stuffing |
| Functional | Action or destination | Appearance-only labels |
| Decorative | alt="" | Repeating nearby text |
| Complex data | Full equivalent plus context | Image as the only record |
Keep multimodal pages interpretable
HTML alt is a contextual textual equivalent or replacement. WHATWG documents srcset, sizes, and picture for resource selection; that does not replace an alternative. [3]
Google says standard <img> elements help crawlers process images, CSS images are not indexed, and responsive images should retain a fallback src. It uses alt text with computer vision and page contents to understand subject matter, and recommends contextual wording without keyword stuffing. [4]
This supports a text channel, not a claim about model weighting. Keep important facts in page text.
Validate the rendered contract
Inspect the rendered DOM. Confirm alt survives the CMS, decorative assets are empty, functional images announce actions, and charts have text equivalents. Test broken or disabled images, screen readers, keyboard focus, narrow layouts, duplicates, and text-only extraction.
A multimodal-ready page remains understandable without the image. That is accessibility and content quality, not a model promise.
References
- W3C WAI: Images Tutorial
- W3C WAI: An alt Decision Tree
- WHATWG: Images in HTML
- Google Search Central: Image SEO best practices