Resources
Why a PDF table becomes the wrong Excel grid
A PDF table is usually drawn text and rules, not a spreadsheet. The converter infers rows from Y coordinates and columns from X gaps, so wrapped cells, blank planned values, and spanning headers become merged rows, dropped columns, or split numbers.
Key takeaways
- PDF stores glyphs and lines. Excel stores a grid. Those are different objects.
- Row shift happens when several printed lines share a Y band, so they land in one cell with embedded newlines.
- A column disappears when the header is only a spanning label or the cell has no glyph to attach to.
- On Papermeld's published operations table, 8 line items plus TOTAL survived, including numeric -6237.25. That is one tidy printed table, not an audit statement.
What a PDF table actually is
A typical report PDF does not contain a table object. It contains characters placed at coordinates, and sometimes horizontal or vertical rules. The word "table" is a human reading of that drawing. Excel needs rows, columns, and cell types. Someone, or some program, has to invent that grid.
Viewer copy-paste does not do that work. It copies the appearance: an image, or a stream of line breaks in the order the glyphs were drawn, which is often not left-to-right then top-to-bottom.
How extraction infers a grid
Text-layer extraction clusters glyphs by Y to guess a row, then by X to guess a column. A wrapped description in the item column sits in two Y bands. If the next line item starts in the same band, the extractor joins them. In Excel that looks like two printed rows jammed into one cell, with a newline in the middle.
Blank cells are the opposite problem. A planned-value cell that was left empty on purpose has no glyph. The extractor may skip that column entirely, then shift later numbers left. Period headers that are only a drawn rule, not text, fail the same way.
Numbers that look like decoration
Thousands separators, trailing cents, leading dollar signs, em-dash blanks, and numbers in parentheses for negatives are common on printed statements. They are not Excel types. A converter that treats each fragment as a separate glyph run can split 39678 into 3 and 9,678, or keep a parenthesized loss as text. On the publishedPapermeld operations workbook, E2 is numeric 39678 and G6 is numeric -6237.25. That is what that file did. It does not mean every statement will land the same way.
Scanned tables are a different job
An image-only PDF has no text layer, so there are no glyphs to cluster. Optical recognition has to invent characters first. Codes and quantities are easy to scramble: a header that reads 42 QTY can come back as 420TY. Papermeld treats that as a searchable-text job. Word or Excel is a second request after you have read the codes. The published receiving-sheet case kept RX-17, LM-04, PK-91, and TOTAL 152 in a searchable PDF. It is not a workbook.
How to check a conversion
Count the printed data rows and the Excel data rows. They should match, plus a TOTAL row if the source has one. Spot-check a negative, a decimal, and a blank. If a number is stored as text, or two line items share a cell, do not total the sheet until you fix those rows against the page.
FAQ
Common table-extraction failures
Why does a copied PDF table paste as a picture?
Most PDF viewers copy the appearance of the table, not the cell grid. You get an image or a jumble of line breaks. A converter has to rebuild rows and columns from glyph positions before Excel can sort them.
When will rows merge into one cell?
When several printed lines share a Y band, or a wrapped description sits in the same vertical slice as the next item. The extractor treats that band as one row. Newlines then appear inside a single Excel cell.
When will a column disappear?
When a header is a spanning label, a period column is only a drawn rule, or a blank planned cell has no glyph. There is nothing to attach a column to, so the grid comes back short.
What did Papermeld keep on the published operations table?
On the 3792-byte synthetic operations PDF, the workbook kept 8 line items plus TOTAL. Variance -6237.25 and total actual 195132.25 are numeric cells. The blank planned cell on Safety audit stayed blank. That is one tidy printed table, not a multi-page audit statement.
What about scanned tables?
A scan has no text layer, so there are no glyphs to cluster. Add a searchable text layer first and read the codes. Then ask for Excel if you still need a grid. One-step scan-to-Excel is not a promised path.
Start in the workspace
Convert one table, then check the grid.
Sign in or create an account. Sign in first, try conversions free, and pay when you hit the usage limit. Ask for Excel from a text-layer PDF, and compare the cells with the printed rows before you sort or total.