About PDF to Excel

Add up to 20 PDFs — each up to 50 MB and 300 pages — and each one comes back as a spreadsheet. By default every PDF page becomes its own worksheet, named Page 1, Page 2 and so on; one checkbox merges them into a single sheet instead, which is what you want for a statement or ledger that runs across many pages.

The .xlsx output is a real Excel 2007+ workbook: numeric cells, percentage formatting, column widths sized to the content, and a frozen bold header row when the first row looks like one. The .csv output is flat UTF-8 with a byte-order mark, which is what stops Excel from mangling accented characters when it opens the file.

A PDF stores no table structure at all — just glyph runs, each with a transformation matrix that says where on the page it sits. Rebuilding a grid from that takes three passes, all of them running on pdf.js, Mozilla's PDF engine:

1. Rows. Text runs are sorted top to bottom and grouped by baseline, with a tolerance scaled to the page's own typical glyph height — so a 6 pt footnote table and an 18 pt heading are both split into the right lines.
2. Columns. Every run's horizontal extent is projected onto a single axis. Wherever a corridor of whitespace survives that projection across the whole page, there is a column boundary. This is why prose collapses to one column — a gap between two words in one line is filled by text in the next — while a table keeps its structure. Lines holding a single run, such as titles and page numbers, are excluded from the projection so one full-width caption cannot erase every boundary on the page.
3. Cells. Each run is dropped into the column it starts in, runs within a cell are re-joined with the spaces the PDF omitted, and empty rows and columns are trimmed away.

Numbers are then recognised conservatively, because a wrong guess corrupts data silently. $1,234.50, 12 345,67, 20% and the accounting negative (45.00) all become numeric cells. Anything ambiguous stays text: an identifier like 007 keeps its leading zero, and a digit group that is not a clean set of thousands — 12 34 — is left alone.

The workbook itself is written here rather than by a spreadsheet library: an .xlsx is a ZIP of XML parts, and generating those parts directly keeps the code that downloads to you small.

Scanned PDFs. A scan is an image of a page with no text layer, and there is no text-recognition step here. Such a file is reported as a scan rather than silently converted to an empty sheet.

Tables with no gaps. Column detection relies on whitespace. A table whose columns are separated only by ruled lines, with text running edge to edge, will come out merged. Preview each file before downloading — it takes a second and shows exactly what the download will contain.

Cells that wrap. A cell whose text wraps onto a second line is read as two rows, because on the page that is what it is. Rotated and sideways text is skipped, since it cannot be placed on a rectangular grid.

Nothing leaves the page. The three passes described above, the number recognition and the ZIP that becomes your .xlsx all execute in your browser tab; no byte of the PDF, and no filename, is sent to a server, logged or stored by PDF to Excel. Passwords typed into the unlock box go to the parser in the same tab and nowhere else.

Because there is no round trip, the tool keeps working with the network disconnected once the page has loaded, and a 50 MB PDF starts converting immediately instead of spending a minute uploading. Removing a file from the list, or closing the tab, releases the in-memory copies — nothing is written to storage on your device, so nothing persists between visits.

This describes what PDF to Excel does with your files. It says nothing about what a browser extension, a network proxy or the device itself might see — only that this tool does nothing with your PDF beyond converting it for you to download.