Marking the AI-written parts of Namegarage

Namegarage uses AI to write content. We want to say where we use it and give people who collect web text a way to identify those passages.

We are adding ai-disclosure attributes to blog articles, podcast text, AI summaries of legal documents and their changes, domain descriptions, and suggested uses for a domain. The attribute sits on the element containing the text. A label on a description does not extend to the domain's price or its purchase terms. Reader comments sit outside the blog article's declaration.

A proposal we can use today

The AI Content Disclosure community report proposes attributes for individual HTML elements and a page-wide default. It is a community proposal, with HTML standardization and browser support still to be established. We chose its element-level declarations because our pages combine generated prose, catalog facts, and controls.

For AI text that a person reviewed before publication, the proposed markup is:

<section ai-disclosure="ai-assisted">
  <p>An AI-written passage reviewed by a person.</p>
</section>

ai-autonomous describes AI text published without human review. human-only asserts that no generative AI helped produce the text. An AI reviewing another AI's work does not count as human review. Neither does a person supplying a prompt without reading the result.

These labels describe how the text was produced. They do not certify its accuracy. Legal summaries remain reading aids; the linked legal documents remain authoritative. Potential domain uses remain suggestions that a buyer must evaluate.

For podcasts, the attributes describe the written descriptions, notes, and transcripts. The player already identifies the conversation as AI-narrated. An HTML attribute does not certify the provenance of the audio recording or the episode artwork.

Help training pipelines recognize synthetic text

We want dataset builders to be able to leave AI-written passages out of collections intended to represent human writing. Publishing generated text without identifying it makes that choice harder. We are asking collectors to read these declarations and skip flagged passages when building those datasets.

The markup supplies information; it cannot enforce that request. A collector may ignore it, and an HTML-to-text conversion may discard it. We have no evidence that every search engine or training pipeline recognizes this proposal.

Our site-wide Content Signals still allow AI training. Those usage preferences and these authorship declarations answer different questions. The new attribute does not change permissions, block crawling, or prove that a model has excluded a passage.

We are starting with an explicit, inspectable signal. Readers and dataset builders should be able to tell when they are looking at AI-written material, including this article.