llm.txt Maintenance Guide

What is llm.txt?

llm.txt is a standardized file that helps AI models (like Claude, ChatGPT, Perplexity) understand your website’s content, structure, and purpose. It’s served at https://humaine.studio/llm.txt and acts like a comprehensive README specifically for LLMs.

When to Update llm.txt

Update the file whenever you make these changes:

✅ MUST Update

• Add/remove blog post → Update “Blog Posts” section with count and new entry
• Add/remove primary page → Update “Primary Pages” section
• Add/remove experiment → Update “Experiments” section
• Major tech stack change → Update “Technical Stack” section
• Change site purpose/positioning → Update “Site Overview” section

📝 SHOULD Update

• Update PostHog/analytics setup → Update “Analytics & Tracking”
• Add new CI/CD workflow → Update “CI/CD Pipeline”
• Change site philosophy/focus → Update “Core Philosophy” and “Key Messages”
• Update contact methods → Update “Contact & Collaboration”

⚪ Optional Update

• Minor CSS/styling changes → Usually not needed
• Typo fixes in posts → Not needed
• Image additions → Not needed unless it’s a new feature

Update Checklist

When updating llm.txt:

1. Update the content sections
   • Verify blog post count matches _posts/ directory
   • Add new posts with date and topics
   • Update primary pages if added/removed
   • Update experiments if added/removed
2. Update metadata
   • Change “Last Updated” date to current date
   • Review “Site Overview” for accuracy
   • Check “Author Information” is current
3. Update technical details
   • Verify tech stack is accurate
   • Update CI/CD pipeline if workflows changed
   • Check analytics configuration
4. Commit with clear message

   git add llm.txt
   git commit -m "Update: llm.txt with [what changed]"
   

Quick Update Commands

# Check blog post count (should match llm.txt)
ls -1 _posts/*.md | wc -l

# Check last modified date of llm.txt
grep "Last Updated" llm.txt

# List all root-level pages
ls -1 *.md *.html | grep -v README

# List experiments
ls -1 experiments/

Integration with CLAUDE.md

When updating CLAUDE.md, ask yourself:

• Does this change affect site structure, content, or purpose?
• If yes → update llm.txt as well
• Both files should tell consistent stories, but:
  • CLAUDE.md = Instructions for Claude Code (how to work on the project)
  • llm.txt = Information for all LLMs (what the site contains)

Automation Opportunities

Consider these future enhancements:

1. Script to auto-count content

   # scripts/check-llm-txt.sh
   POST_COUNT=$(ls -1 _posts/*.md | wc -l)
   LLMTXT_COUNT=$(grep -c "^[0-9]\." llm.txt)
   if [ $POST_COUNT -ne $LLMTXT_COUNT ]; then
     echo "⚠️  llm.txt may be out of sync"
   fi
   

2. Pre-commit hook to remind about llm.txt updates

3. GitHub Action to auto-update “Last Updated” date

Verifying LLM Discovery

How llm.txt Works

llm.txt is a convention, not a web standard. It’s inspired by robots.txt and humans.txt. LLMs that support it will:

1. Check for the file at https://yourdomain.com/llm.txt
2. Read and parse the content
3. Use it to better understand your site when answering questions

Important: Not all LLMs automatically check for llm.txt. It’s an emerging convention.

Testing File Accessibility

1. Manual URL Test (Most Important)

   # After deploying, verify the file is publicly accessible
   curl https://humaine.studio/llm.txt
   
   # Or just open in browser:
   # https://humaine.studio/llm.txt
   

   ✅ Expected result: Should see your llm.txt content as plain text ❌ If 404 error: File not being served by Jekyll/GitHub Pages

2. Check Jekyll Build

   # Build locally and check output
   bundle exec jekyll build
   ls -la _site/llm.txt
   
   # Preview locally
   bundle exec jekyll serve
   # Then visit: http://localhost:4000/llm.txt
   

3. Verify in GitHub Pages
   • Go to repository Settings → Pages
   • Confirm site is published
   • Click “Visit site” → manually append /llm.txt to URL

Testing with LLMs

Once your file is accessible, test if LLMs can find and use it:

Test with Claude (claude.ai or Claude Code)

Prompt: "Can you read the llm.txt file from humaine.studio and tell me what you learned about the site?"

Expected: Claude should fetch the URL and summarize the content.

Test with Perplexity

Prompt: "Look at humaine.studio's llm.txt file and describe what the site is about"

Perplexity has web access and should be able to fetch public URLs.

Test with ChatGPT (with browsing enabled)

Prompt: "Visit https://humaine.studio/llm.txt and summarize what this site offers"

Note: ChatGPT requires explicit URL mention to fetch content.

Common Issues & Fixes

Problem: llm.txt returns 404

• Cause: File in wrong location or Jekyll not copying it
• Fix: Ensure file is at repository root (not in subdirectory)
• Jekyll note: Files at root are automatically copied to _site/ unless excluded in _config.yml

Problem: llm.txt shows as HTML instead of plain text

• Cause: Jekyll is processing it as markdown/template
• Fix: Add to _config.yml under include: to copy as-is: ```yaml include:
  • llm.txt ```

Problem: LLM says “I can’t access external URLs”

• Cause: Some LLMs don’t have web browsing enabled
• Fix: Copy/paste the llm.txt content directly into the conversation

Problem: LLM doesn’t find useful info

• Cause: LLM may not automatically check for llm.txt
• Fix: Explicitly mention the file: “Check the llm.txt file at humaine.studio”

Monitoring Discovery

# Check if llm.txt is being accessed (if you have analytics)
# PostHog query: Filter events for URL path = "/llm.txt"

# Check GitHub Pages build logs
gh run list --workflow=jekyll.yml --limit 1
gh run view [run-id] --log

# Verify file in deployed site
wget https://humaine.studio/llm.txt -O -

Best Practices for Discoverability

1. Keep file at root - https://humaine.studio/llm.txt (not in subdirectory)
2. Use plain text markdown - No fancy formatting that breaks parsing
3. Update regularly - Stale info is worse than no info
4. Link to it - Mention in footer or about page for humans to discover too
5. Test after every deploy - Quick curl check to ensure it’s still accessible

Example: Adding a link in your footer

You could add this to _includes/footer.html:

<a href="/llm.txt" title="Machine-readable site information">llm.txt</a>

This makes it discoverable by humans too!

---

Last Updated: 2026-01-18