Proposal: explore section-level sitemap output

[KimHyeongRae0]

Proposal

Explore optional section-level sitemap output for larger sites.

Why

The current sitemap generator emits a single URL-level sitemap. That is a good default, but larger documentation or content-heavy sites may benefit from an additional section-level view that groups related pages by directory or content area. This can make generated discovery files easier for crawlers and downstream tools to reason about.

Possible Output

Examples:

• sitemap.xml: existing URL-level sitemap, unchanged.
• section-sitemap.json: section summaries derived from URL paths and page metadata.
• Optional references from ai-index.json or report output to the section groups.

Example shape:

{
  "sections": [
    {
      "path": "/docs",
      "title": "Docs",
      "pages": ["/docs/getting-started", "/docs/configuration"]
    }
  ]
}

Questions

• Should this be enabled by default or behind a new generator flag?
• Should sections be inferred from URL paths, configured manually, or both?
• Should the output be XML, JSON, or only added to existing AI-facing JSON assets?

Non-Goals

• No changes to the existing sitemap.xml default behavior.
• No search engine ranking claims.
• No required page metadata changes.

[rubenmarcus]

Thanks for the proposal. Before scoping this, I want to make sure we're solving a real problem rather than adding another generated artifact. A few clarifying questions:

1. Who consumes section-sitemap.json? The existing sitemap.xml is the spec'd standard; docs.json (our manifest) already groups pages and is consumed by tooling. Do you have a specific crawler or downstream tool that would actually use a new section format?

2. Overlap with docs.json. Our manifest output already exposes structure. Would it be more valuable to extend docs.json with explicit section metadata (titles, descriptions, page lists) than to add a separate file?

3. Trigger. Is the value primarily for AI crawlers (which is our focus) or for traditional SEO (which is not)? AI assistants typically traverse llms.txt / llms-full.txt rather than sitemap-style files.

If you can point to a concrete consumer that benefits, happy to scope this. Otherwise I'd lean toward not adding another generated artifact yet.