How it works
- 01Open the PDF.
- 02The panel lists the pages that contain a ruled grid.
- 03Download — one CSV, or a zip if there are several tables.
- 04Check the result against the document before you rely on it.
It reads the lines, not a picture of them
A PDF's ruling lines are stored as exact vector coordinates, so where the columns are is already written down in the file. This reads those coordinates directly. The usual approach — rendering the page and hunting for lines in the pixels — is approximate, needs megabytes of computer vision, and is strictly worse for a document that already knows the answer.
Ruled tables only, and that is a deliberate limit
A table with visible lines has told you where its columns are. A table laid out with spacing has not, and recovering one is a research problem whose failure mode is plausible-looking wrong data. A spreadsheet that opens cleanly and contains the wrong numbers is worse than no spreadsheet, because nobody re-checks it. So this reads ruled tables and says so plainly when it finds none.
A grid is not always a table
A chart's axes, a decorative frame and a form's boxes all make a perfectly good lattice with no data in it. Those are found, recognised as empty, and dropped — rather than exported as a CSV full of blank rows that looks like the tool half-worked.
The CSV is properly quoted
A field containing a comma is quoted, and a quote inside a field is doubled, as the CSV standard requires. Skipping this is how a table containing "1,234" arrives as two columns in every spreadsheet that opens it — and it is skipped surprisingly often.
What it will get wrong
Merged cells are not understood: a cell spanning two columns lands in one of them. A table continuing across a page break comes out as two separate tables. Both are stated in the panel rather than left for you to discover in the data, and both are why the advice is to check the result.
It does not work on scans
A scanned table has no ruling lines in the file — only a photograph of some. The panel detects this and says so, instead of returning an empty result that reads as "there are no tables here".
Questions
- Can it read a table with no lines?
- No, deliberately. Guessing at columns from spacing alone produces data that looks right and is not, which for numbers going into a spreadsheet is the worst possible outcome.
- What if my document has many tables?
- You get a zip with one CSV per table, named by the page it came from and numbered so they sort in document order.
- Can I get one file with everything in it?
- Not yet. Turning on the page-number column makes the separate files easy to concatenate yourself, which is the nearest thing today.
- Is my document uploaded to convert it?
- No. The table is found and written by your own browser, and you can disconnect from the network first to confirm it.