MoeLog LLMs.txt

A clean, efficient WordPress content index plugin built for AI language models.

This plugin follows the emerging llms.txt proposal for WordPress, providing AI tools with a structured site directory and noise-free Markdown versions of public content — without HTML clutter.

Features

• llms.txt-style index: Automatically generates /llms.txt, listing published, non-password-protected posts and pages.
• On-demand Markdown output: Append .md to any post or page URL (e.g. https://example.com/about.md) to retrieve a clean Markdown version of that content.
• HTML-to-Markdown cleanup:
  • Strips header, footer, nav, aside, script, style, and other non-content elements.
  • Converts the core content to standard Markdown using a DOMDocument-based recursive converter.
• Self-contained URLs:
  • Root-relative (/path) and protocol-relative (//host/path) URLs become absolute, so images and links still resolve once the Markdown is read outside the site's HTTP context. Root-relative paths resolve against the site origin, exactly as a browser would read them — a subdirectory install is never prepended.
  • Document-relative URLs (image.jpg, ../uploads/a.jpg) are left untouched, since resolving them needs the containing document's own address. WordPress editors do not normally emit them.
  • Links to your own published posts and pages are rewritten to their .md counterparts, keeping AI crawlers on the Markdown fast lane. Category, tag, attachment, query-string, external, and (on subdirectory installs) out-of-install URLs are left untouched, so no link is rewritten into a 404.
• Smart noise filtering:
  • Temporarily removes the_content filter callbacks from configured noise sources (YARPP, ad injection plugins, related post widgets, etc.) before rendering Markdown.
  • Only the article body reaches the AI — no wasted context window.
• Wiki redlink demotion:
  • When the Moelog Wiki Links plugin is active, [[term]] entries with no matching Wikipedia article are emitted as plain text rather than as links to a search-results or article-creation page. On a rendered page those are styled red, but Markdown keeps no styling — an LLM would just see an ordinary-looking link and could mistake a search page for a citable source.
  • Entries that do exist keep their real Wikipedia URLs, so no information is lost.
  • The filter is registered only while a .md response is being built and removed in a finally block, so visitor-facing pages and the database are untouched. It goes through the public mwl_link_html hook, so neither plugin depends on the other being installed.
• SEO-safe:
  • .md responses include X-Robots-Tag: noindex, preventing search engines from treating them as duplicate content.
  • No Disallow in robots.txt — AI crawlers like GPTBot and ClaudeBot can access .md files freely.
• Cache-friendly:
  • .md responses include Cache-Control: public, max-age=3600.
  • /llms.txt is cached for 30 minutes and automatically rebuilt at the origin after content, site identity, or permalink settings change.
• Protected-content safe:
  • Password-protected, private, and unpublished content is excluded from both /llms.txt and .md responses.

Requirements

• WordPress 6.0 or later.
• PHP 7.4 or later.
• Pretty permalinks are required for .md endpoints. With Plain Permalinks, /llms.txt falls back to the original HTML URLs.
• PHP DOM is recommended. If it is unavailable, Markdown output safely falls back to plain text.

Installation

1. Upload the moelog-llms-txt folder to /wp-content/plugins/, or install via the WordPress admin panel.
2. Activate MoeLog LLMs.txt from the Plugins screen.
3. Visit /llms.txt and one public .md URL to verify the endpoints.

If .md URLs return 404, go to Settings → Permalinks and click Save Changes once as a troubleshooting step. Activation normally flushes the rewrite rules automatically.

Verify it's working

• Visit https://yoursite.com/llms.txt to confirm the index is generated correctly.
• Pick any post URL and append .md (e.g. https://yoursite.com/hello-world.md) to see the clean Markdown output.

Design Philosophy

As AI agents and intelligent assistants become primary traffic sources, traditional HTML pages — bloated with DOM structure, CSS, ads, and tracking scripts — are increasingly hostile to machine consumption.

This plugin opens a dedicated AI fast lane for your WordPress site: no heavy llms-full.txt dumps, no reinventing the wheel. AI crawlers first get a lightweight /llms.txt overview, then pull only the specific /*.md content they need — precise, low-overhead, and context-window friendly.

Customization

If another plugin injects noise into post content, extend the callback-name patterns from your own small plugin or mu-plugin. Do not edit includes/class-llms-txt.php, because updates would overwrite that change:

add_filter( 'moelog_llms_noise_patterns', function ( $patterns ) {
    $patterns[] = 'My_Ads_Plugin::append_ad';
    return $patterns;
} );

Use the narrowest callable or class/method fragment possible. Broad patterns such as jetpack can remove useful gallery, shortcode, or formatting callbacks along with sharing widgets. After changing these patterns, compare the Markdown output of a real post before and after the change.

To keep wiki redlinks as links in the Markdown output instead of demoting them to plain text:

add_filter( 'moelog_llms_demote_wiki_redlinks', '__return_false' );

License

GPL-2.0+