Choosing a Format

Prefer the original format.
If your content already exists in a specific format, keep it as-is.

Choosing a Format for New Content:

Tabular Data

1. Stand Alone Resource (CSV/Excel)
   • Tables are small to medium sized
   • The tabular data is the primary content
   • Users will ask questions about the data (not query it directly)
   
   

   Currently only tables are extracted...surrounding text is ignored

2. Tables Inside Documents (PDF/Word)
   • Tables are part of a larger narrative
   • Context matters (e.g., explanation + table together)
   • Tables are small to medium sized
3. Agents (CSV/Excel)
   • Data is large or frequently queried
   • Users need to:
     • Filter
     • Lookup values
     • Aggregate data
     
     Examples:
   • “Find the value in column C where column F = X”
   • “What’s the average pressure for pumps in region A?”
4. Other Data Sources
   If your data lives in:
   • Databases
   • APIs
   • Other structured systems

👉 Contact us to discuss integration options.

Decision Guide

Frequently Changing Content

Dynamic resources (HTML/Markdown)

Best for:

• Content that changes frequently
• Unstructured or miscellaneous content

Benefits:

• Renders cleanly in the browser
• Easy to edit and maintain

Everything Else

PDF, Word (.docx), Text (.txt), Markdown (.md), etc.

Note: PDF files open in the browser. Other file types download when opened.

Best Practices

Word

• Use predefined styles (Heading 1, Heading 2, etc.)

HTML

• Use semantic HTML
  • Headings (<h1>, <h2>, etc.)
  • Navigation elements (to exclude menus from parsing)
• Add title attributes to images and links
• Ensure tables include a header row
• Avoid using tables for layout
• Use inskill-skip (attribute or class) to exclude irrelevant content (e.g. cookies banners, shopping cart, language selectors, etc.)

PDF

• Prefer tables with visible borders/lines for better detection

Excel

• Named tables or filtered ranges improve table detection
• Top row should be the headers when possible
• Avoid vertical headers