PDF to CSV

Misaligned columns come from the detection step rather than from the CSV writing, and the fastest fix is usually to narrow what you export. PDFChamp finds columns by looking for horizontal gaps wide enough to be gutters rather than word spacing, and by preferring the grid implied by rows that look tabular on their own. A full-width title or a paragraph of prose sitting inside the table block can still stretch one column across the whole page. Set the Pages chip to Range and export the single page you care about, so another page's layout cannot influence the grid, and use First only if a header block is being detected as its own table. If it still refuses to come out, PDF to text gives you the raw content to reshape by hand.

An empty detection on a PDF that visibly contains tables has two usual causes. The first is a scan: if the pages are images there are no text runs to cluster, so no table can be found — run OCR PDF over it to add a text layer, then come back. The second is a table drawn without real gutters, where cells are packed tightly or the visual structure comes entirely from ruling lines while the text runs nearly touch. Detection looks at text positions, not at the lines drawn on the page, so a heavily ruled but cramped table can read as ordinary prose. A block also has to reach at least two columns and two rows to count as a table at all, which filters out stray headings but discards very small tables too.

Merged cells do not survive as merges, because CSV has no concept of one. A cell spanning three columns lands in the first column it overlaps and the rest are left empty, so that row looks shorter than its neighbours until every row is padded to the same width on export. A row whose text wraps onto a second visual line is treated as two rows, since rows are grouped by vertical position and a wrapped line sits at a different height. Nested or stacked headers usually produce two header rows rather than one combined header. None of that is corrupted data, it is structure you may need to tidy in the spreadsheet — and the ten-row preview shows you exactly what you will get.

Every selected table is written into the same file one after another, separated by a single blank row, and the file is named after your source PDF. That keeps the export to one download, but a spreadsheet treats the whole thing as one sheet, so a file holding four tables of different widths will have ragged columns below the first blank row. There are two ways out: set the Pages chip to Range and export each page separately for one clean file per table, or use PDF to Excel instead, which puts each page — or each individual table, if you pick that layout — on its own worksheet. For a single table the concatenation never comes into play at all.

Cells are written exactly as they appear on the page, as text. Currency symbols, thousands separators, percent signs, footnote markers and trailing spaces all come through verbatim, so a spreadsheet may import a column of amounts as text rather than as numbers depending on its locale settings. That is deliberate — reinterpreting values would risk changing them. Use your spreadsheet's text-to-columns or value conversion once the data has landed. The file is UTF-8 and standard CSV quoting is applied, so cells containing commas, quotes or line breaks are escaped correctly and cannot shift your columns. If a European spreadsheet splits nothing at all, tell it on import that the delimiter is a comma.

The document never leaves your browser, and there is no row, table or page limit in the tool. Detection and CSV generation both run in the tab: pdf.js reads the text runs, PDFChamp's clustering builds the grid, and the CSV is assembled locally before the download is triggered. Nothing is posted anywhere, which is exactly why people use it for invoices, payroll exports and anything else they would rather not hand to a conversion service. It is free, needs no account, and the CSV carries no added rows, headers or branding of any kind. The realistic ceiling is memory: a very long document holds every detected table in the page at once while you review the preview.