PDFGym
EN

Turn a PDF table into CSV

or drop one here
Choose a PDF first.

A PDF stores words at page coordinates, not cells in a grid. This page finds the tables hiding in that coordinate soup, lets you tick the ones you need, and downloads each as a CSV that is parsed back and matched cell by cell first. Scanned pages are refused outright instead of yielding empty files.

The supplier price list trapped in PDF

When the numbers you need sit in a document instead of a spreadsheet, retyping 400 rows is the failure mode. Extract the grid and check it.

An analyst receives supplier-pricelist-q3.pdf: 9 pages, three product tables, no Excel in sight. She opens it here, and the detector lists three tables with their pages and shapes. She ticks table 2 on page 4, leaves Spreadsheet safe switched on, and presses Convert table to CSV.

The download arrives as supplier-pricelist-q3-table-2.csv, every cell parsed back and matched before release. Ticking all three instead would have packed a ZIP with one CSV per table, each named the same way.

Four passes from PDF to verified CSV

  1. Open the PDF

    Press Choose a PDF, or drop the file on the page. The pages need selectable text; a bare scan has nothing to align.

  2. Tick the tables

    The Choose detected tables panel lists every grid found, with Select all and Deselect all for quick work. One tick downloads a single CSV, several ticks pack a ZIP with one CSV per table.

  3. Pick a cell safety mode

    Spreadsheet safe prefixes an apostrophe to cells starting with =, +, -, or @ so Excel and Sheets cannot run them as formulas. Exact text keeps every character untouched for raw data work.

  4. Verify and download

    Each CSV is parsed again and matched cell by cell against the detected grid, and ZIP packs are reopened too. Download CSV or Download ZIP appears only after everything matches, with a proof line counting rows, columns, and cells.

How tables get found and named

The detector hunts for repeated horizontal and vertical alignment in the native text layer: columns of words that line up, rows that repeat. Each find becomes a numbered table in document order, and the result pairs the source page render with the extracted grid so you can judge the inference yourself. That preview is the real check on detection quality; use it every time.

Names follow the numbers. Table 2 of the pricelist above downloads as supplier-pricelist-q3-table-2.csv, while several ticks share supplier-pricelist-q3-tables.zip with one CSV inside per table. The proof line then counts rows, columns, and cells read back exactly, plus how many cells earned formula protection. What the check proves is the round trip from grid to CSV and back, not that a merged heading belonged to one column rather than three.

Tables this detector misreads

Scans come first because they fail loudest: with no text layer the detector reports that no table of aligned selectable text was found, and suggests a native-text file instead of printing an empty CSV. Run scanned statistics through an OCR step elsewhere first; this tool starts where selectable words exist.

Inside native files, four layouts mislead. Merged headings get pinned to a single column. Rotated tables read sideways. Form fields scatter. Multi-column article prose can pose as a grid. There is no in-tool correction for any of these, so fix the stragglers in the spreadsheet after import. And treat Exact text mode with respect: it keeps leading formula characters untouched, so open those files only when you trust the source.

When CSV is the wrong container

When the destination is Excel rather than a pipeline, PDF to Excel runs the same detector and writes one table per worksheet. When the surrounding paragraphs matter as much as the grid, PDF to Markdown carries the whole document across with its structure. When you need to see the page rather than compute on it, PDF to images renders pictures instead.

Why money tables stay on your device

Price lists, payroll exports, and client ledgers are exactly the files that should never detour through a stranger server. Detection, extraction, and the parse-back check all happen where the PDF already sits, with no account and no retention window to trust.

The safety mode extends that caution into the spreadsheet. A price cell reading -5% or a code starting with @ could wake up as a formula in Excel; the apostrophe keeps it asleep as plain text. Between local processing and guarded cells, the numbers arrive without surprises at either end.

Short answers about PDF tables to CSV

Why did my scanned price list find nothing?
Scans carry pixels, not words, so the aligner has nothing to group. The tool says so plainly instead of shipping an empty CSV. Run the scan through OCR elsewhere, then bring the text version here.
A heading spans three columns but landed in one. Can I fix it here?
No. Detection infers boundaries once, and there is no correction pass in the tool. Check the two-view preview, download, and nudge the stragglers in your spreadsheet.
Which cells earn the apostrophe in safe mode?
Cells opening with =, +, -, or @, the characters spreadsheets can wake as formulas. The result note counts how many were guarded; other text keeps its characters with spacing normalized.
Which table is table-2 in the file name?
The second grid the detector found, counting through the document in order. The result note also names its page, so table-2 on page 4 is unambiguous.
Can I merge three tables into one CSV?
No. Each selected table downloads as its own CSV by design, so unrelated grids never fuse. Stack them in the spreadsheet after import.

More tools that free trapped data

The tools below run the same local way: detection on your device, outputs rechecked before download, and no account anywhere.