What problem are you trying to solve?

The current HTML specification provides the alt attribute on images to supply alternative text for users who cannot see visual content. However, there is no way to indicate whether the alt text was authored by a human or generated automatically by a machine (e.g., via AI or computer vision).

As machine-generated alternative text becomes more common across platforms (e.g., social media, CMS tools, accessibility overlays), users, particularly those who rely on screen readers, have no visibility into the origin or likely quality of the description. This creates issues such as:

• Trust and reliability: Machine-generated descriptions may be inaccurate, misleading, or contextually wrong. Users may want to treat them differently or verify them.

• Accessibility auditing and compliance: Tools and organizations performing accessibility checks need a way to differentiate between human-verified content and auto-generated fallbacks, especially when evaluating WCAG conformance.

• User experience: Screen reader users may prefer to skip, treat differently, or receive additional cues for machine-generated content.

• Moderation and transparency: Platforms and regulators may want to monitor when and how machine-generated content is presented as accessible.

By providing a standardized way to declare the provenance of alt text (e.g., human-authored vs. machine-generated), we can improve transparency, trust, and tooling across the web without breaking backward compatibility.

What solutions exist today?

Current solutions rely on users manually indicating in the alt text itself that it was AI-generated (e.g., "Image description generated by AI"). This approach is entirely discretionary, lacks standardization, and is not machine-readable or enforceable through any formal mechanism.

• Non-standardized: There is no consistent format or convention for indicating origin. Phrases like “AI-generated” or “autogenerated” vary in wording, placement, and language.
• User-dependent: Adoption relies entirely on the discretion of individual authors or platforms, this means it's often ommitted,
• Not machine-readable: Accessibility tools, screen readers, and auditing software cannot reliably parse or act on this information as it currently stands,
• No accessibility semantics: Embedding origin metadata inside alt text may confuse users and does not provide meaningful cues to assistive technologies.
• Difficult to enforce or verify: Without a formal attribute, it's difficult to track or validate the provenance of alt text across large-scale systems.

How would you solve it?

I propose introducing a new attribute alt-origin to the <img> (and potentially <area> and other media-related) elements in HTM, which currently use an alt identifier.

Proposed Values/Attributes

Values (enumerated or tokenized string):

• "machine" – The alt text was generated automatically, e.g., by AI, computer vision, LLM or template logic.
• "human" – The alt text was written or reviewed by a human.
• "unspecified" or omitted – Origin is unknown or not declared (backward-compatible default).

Anything else?

No response