How to extract PDF tables without losing rows
Dynamite Docs, 2026-08-30
Why PDF rows disappear
A PDF stores text at page coordinates. It does not always store a real table with cells and rows. A visible row can be several separate text fragments, while a wrapped description can look like two records to an extractor.
Rows also disappear when lines sit close together, a page break cuts through a record, or the tool mistakes a subtotal for the end of the table.
Set a row identity rule
Choose the field or combination that marks a new record. In a bank statement, a date plus an amount may identify a row. In an invoice, an item code or quantity and line amount may do it. Text below that marker can remain part of the same record.
Do not depend on horizontal borders alone. Many PDFs draw faint rules as decorative graphics or omit them entirely.
Count source rows before cleaning
Record a raw count for each page, including legitimate blank-field rows. After removing repeated headers and subtotals, record the expected data-row count. Compare that number with the final table.
Counts will not prove that every value is correct, but they quickly expose missing and duplicated records.
Treat wrapped text as part of the row
When a description continues on the next visual line, join it only if the anchor columns on that line are empty and the vertical spacing matches a continuation. Preserve a space between fragments so words do not run together.
If a continuation lands at the top of the next page, use the previous page state before deciding whether it starts a new record.
Detect duplicates at page boundaries
Some reports repeat the final row of one page as the first row of the next. Compare stable fields such as date, reference, description, amount, and balance. Remove a boundary duplicate only when the complete row matches, not because two transactions share the same amount.
Use totals and sequence checks
Reconcile line totals, running balances, record numbers, or printed page subtotals. Sequence gaps often reveal a lost row. A duplicated row usually creates an equal and opposite reconciliation difference at the point where it entered the table.
Review row structure in Dynamite Docs
Dynamite Docs infers the table schema, keeps the PDF next to the extracted result, and lets you merge, split, or correct rows before export. Start with the pages that contain the longest descriptions and the busiest page break.
When the hard sample preserves its row count and reconciles, process the full file and export the reviewed table.
Related workflow: Accounting document automation and data extraction.
Try it yourself. Upload a PDF, scan, or image and let Dynamite Docs infer the schema.