Get the tables out of a PDF.
Tables out of a report and into a spreadsheet, with every number checked against the totals printed next to it.
No account needed to start. You only pay when you like what you see.
- .pdf. PDF. Checked for a text layer first. A PDF written by software keeps its table structure; a scan is read as an image and told apart from the first kind.
- .docx. Word document. Tables are lifted out of the document; the prose around them is kept as context.
- .pptx. PowerPoint deck. Tables and the data behind native charts are read slide by slide.
- .rtf. Rich text. Tables extracted, formatting discarded.
- .png. PNG image. Screenshots of a dashboard or a table are read, and every figure is checked against the totals in the same image.
- .jpg. JPEG photo. A photograph of a printed report or a handwritten sheet is read into a table.
- .jpeg. JPEG photo. The same as .jpg.
- .heic. iPhone photo. Accepted as it comes off the phone. Safari usually hands it over as a JPEG.
- .webp. WebP image. Read like any other screenshot.
- .tiff. TIFF scan. Multi-page scans are read page by page.
- .txt. Plain text. Read as notes, or as data when the contents turn out to be a table.
- .md. Markdown. Markdown tables become tables; the rest becomes context for the analysis.
- .zip. Zip archive. Unpacked, and each file inside is treated as though you had uploaded it yourself.
What comes back
31 tables out. Nineteen of the twenty-two that could be checked add up, and the three that do not are named.
4,118 rows and 18,442 figures were recovered from 60 pages. 22 tables print a subtotal, which is the only check available without the original open beside you: 19 reconcile exactly, and the other three are reported with their cause — a footnote read as a data row, a document whose own total was rounded differently from its rows, and a column containing a figure that could not be read. 84 figures across both files are marked unreadable rather than guessed, and 78 of them come from the scanned statement, which carries roughly eighty times the error rate of the report.
What happened when each table was added up
Nine tables print no total, so nothing can be checked
- 01Three of the 22 checkable tables do not add up, and each has a different cause.
- 02The scan carries eighty times the uncertainty of the report.
- 0361 of the 84 unreadable figures are in one table.
01
It works out which kind of PDF it has
A PDF written by software carries a text layer and its tables come out almost exactly. A scan carries pixels and has to be read. These are different jobs with different error rates, so the file is classified first and the answer says which one you got.
02
It rebuilds the table, not the page
Columns separated by whitespace, rows split across a page break, a header repeated on every page, a footnote sitting inside the grid. What comes back is one table per table, not one block of text per page.
03
It checks the figures against the page
Where the document prints a subtotal, the extracted rows are added up and compared to it. A column that does not reconcile is flagged on the spot rather than shipped as a clean-looking spreadsheet.
It expects a mess.
Most files that arrive here have several tables on one sheet, a title block above the data, totals stranded in the middle of it, and dates in more than one format. That is the normal case, not the awkward one.
- Every table as a sheet in one .xlsx
- CSV per table
- Reconciliation against the totals printed in the document
- Anything unreadable, listed with its page
| Order | Date | Customer | Amt |
|---|---|---|---|
| 10241 | 07/02/26 | John Smith | $84.00 |
| 10242 | 2 July 2026 | SARAH LOPEZ | $41.00 |
| 10242 | 2 July 2026 | SARAH LOPEZ | $41.00 |
| 10243 | 45840 | s. lopez | 41 |
| Not supplied | Not supplied | SUBTOTAL | $166.00 |
| 10244 | 2026-07-03 | Amara Osei | $127.50 |
Questions it works through
- Which tables are in here, and where do they start and end?
- Do the extracted numbers add up to the totals on the page?
- Which figures could not be read confidently?
- Can this run every month on the same report?
Questions people ask
Will it work on a scan?
Yes, and less well, which it tells you. A PDF with a text layer extracts close to exactly; a photographed or scanned one is read visually and lands in the interpreted tier, where every figure is re-derived against the document's own totals before it is shown.
What about a table split across pages?
It is stitched back into one table, with the repeated header removed. A row broken across the page boundary is rejoined rather than left as two half rows.
Can it do a hundred of them?
Zip them and drop the zip. The rules worked out on the first are saved, so the same monthly report is minutes rather than a fresh start every time.
Related
Upload a PDF
No account needed to start. You only pay when you like what you see.