PDF tables to Markdown without losing the meaning
A reliable table conversion preserves which value belongs to which header. Borders that look right are not enough. This guide shows when GFM is suitable, when to keep HTML or JSON, and what to verify against the PDF.
Short answer: Use a GFM pipe table only when every row has one stable value per column. If the PDF contains merged cells, multi-row headers, footnotes inside cells, or a table that continues across pages, keep a structured companion export and review the source before publishing.
What Markdown tables can actually represent
The GitHub Flavored Markdown specification defines a table as one header row, one delimiter row, and zero or more body rows separated by pipes. It does not define row spans or column spans. That makes simple grids portable, but it also means a visually faithful conversion can still attach a number to the wrong header.
| PDF structure | Safe Markdown choice | Review required |
|---|---|---|
| Single header, rectangular rows | GFM pipe table | Check row and column count |
| Merged cells or multi-row headers | HTML or structured JSON companion | Rebuild header relationships |
| Table continues on the next page | One logical table plus page notes | Remove repeated headers carefully |
| Cell footnotes or formulas | Markdown table plus notes below | Verify every marker and symbol |
A five-minute verification pass
- Count the columns in the original header and in the Markdown delimiter row.
- Check the first row, one middle row, and the last row on every source page.
- Search for merged labels, indentation, ditto marks, and blank cells that carry meaning.
- Join cross-page tables only after confirming that units and column order stay unchanged.
- Keep the original page number beside any value that will be quoted, calculated, or fed into another system.
| Metric | 2025 | 2026 |
| --- | ---: | ---: |
| Revenue | 120 | 136 |How pdfmd handles the workflow
pdfmd first analyzes the PDF in the browser, then sends a text PDF to Cloudflare conversion or renders a fully scanned PDF for Moondream OCR. The result can be exported as clean Markdown or with source metadata. A text-PDF page costs 1 point; a scanned page costs 4 points; failed conversions cost 0 points.
Conversion is a first draft, not a table auditor. Source-map JSON preserves only metadata the provider actually returned; it does not prove that a merged header was reconstructed correctly. Review the financial-statement sample or your own PDF before using extracted numbers.
Official sources and evidence boundary
These links support format and platform behavior. pdfmd processing and point costs are governed by this site's pricing and privacy pages.
Keep the review checklist
Download the Markdown checklist and reuse it next time. Export-format changes are recorded in the public changelog.
Frequently asked questions
Can Markdown preserve merged PDF cells?
No. GFM tables have no rowspan or colspan. Use HTML or a structured companion export and explain the relationship in text.
Should repeated headers on a second PDF page be deleted?
Only after confirming that the second page continues the same table with the same units and column order.
Does source-map JSON prove that a table is correct?
No. It records available page, block, method, offset, and confidence evidence; it does not validate header-to-value relationships.
How much does PDF table conversion cost?
A completed text-PDF page uses 1 point. A completed scanned page routed through OCR uses 4 points. Failed conversions use 0 points.
Convert, then audit the risky rows
Read the checklist first. When you are ready to process a real file, Google sign-in and spendable page points unlock cloud PDF conversion.