Blog ›

The best PDF to Excel converters compared: Excel's own import, Acrobat, Tabula, online converters and Python

30 September 2026 · 8 min read

"PDF to Excel" is one of the most searched document tasks and one of the most disappointing: the converter produces a spreadsheet, and only later do you notice the wrapped descriptions that became extra rows and the German amounts that Excel read as dates. We compared the main options on the things that actually decide whether the numbers are right, and we're honest about which of them keeps your file on your computer.

What to judge a converter on

• Real tables, not demo tables. Merged header cells, wrapped multi-line cells, tables that continue across pages, columns aligned by whitespace only. Every converter handles the clean demo; few handle a real supplier statement.
• Numbers as numbers. "1.250,00" must become 1250 in a German locale and stay a number, not text and not a date.
• Privacy. Bank statements and invoices are about as sensitive as documents get. Uploading them to a converter site is a decision, not a default.
• Scans. A photographed table needs OCR first; most free tools silently produce nothing.
• Repeatability. If the same PDF arrives monthly, the tool should let you do it again without redoing the clean-up.

The comparison

Tool by tool

Excel's built-in PDF import: the one most people don't know about

In Microsoft 365 on Windows (and recent Excel for Mac), Data → Get Data → From File → From PDF opens a navigator listing every table Excel detected, per page, with a preview. Because it goes through Power Query, the split, trim and type-change steps you apply are saved, so next month's statement is a file swap and a Refresh. For ruled tables in text PDFs it is the best free option and it never leaves your machine. It cannot read scans and it struggles with whitespace-aligned columns.

Acrobat Pro: the safe paid choice, with OCR

Acrobat's Export PDF → Spreadsheet is the most polished single button and runs OCR on scanned pages as part of the export. If you already pay for Acrobat, use it. If you don't, the subscription is hard to justify for occasional tables.

Tabula: draw a box around the table

Tabula runs locally in your browser (it needs Java). You draw a rectangle around exactly the table you want, choose Lattice for ruled tables or Stream for whitespace ones, and export CSV. The rectangle is why it beats automatic tools on messy pages: footnotes and headers never get in. One table at a time, text PDFs only.

Online converters: fast, but you're uploading your bank statement

The convert-in-browser sites are quick and require nothing installed. Accuracy is fine on simple tables and poor on merged cells and multi-page tables, with little control when it goes wrong. The bigger point is privacy: the file is transmitted and processed on their servers, usually deleted after some hours according to their policies. For a public price list, fine. For a bank statement, an invoice with client names, or anything under an NDA, prefer one of the local options.

Python: for volume, with a warning

If you have fifty PDFs or the same one every week, a ten-line pdfplumber script (see our extraction guide) beats everything above. Camelot adds a per-table accuracy score so you can flag pages for review. The warning: scripts are only as good as the checks you add, so always recompute a printed total.

AI chat: good at reading, dangerous with numbers

Uploading a table screenshot to ChatGPT or Claude and asking for CSV works far better than it should, including on scans. The failure mode is subtle: models occasionally "fix" a number that looks like a typo, drop a row that seems duplicated, or silently round. Use it for the layout and then verify every column total against the PDF.

Where pdfagents.ai fits

Our own tools are in this comparison because they exist now. Drop a PDF on pdfagents.ai, say “put the tables into Excel” (or CSV), and you get a workbook with one sheet per table. Three things it does that the quick converters usually leave to you: lines that wrapped inside a cell are merged back into their row, repeated cells are placed by column position rather than by counting spaces, and amounts are written as real numbers whether the document uses 1.250,00 or 1,250.00, so SUM works straight away. It runs in your browser, so a bank statement never leaves your computer.

Where it is behind: it finds tables by alignment, so a table with ruling lines but ragged text can confuse it; it does not yet recompute the printed totals for you; and it handles one file at a time, with no saved query to refresh next month, which is where Excel's own import still wins. For an invoice where you want the number, date and total rather than the line items, the key-fields tool is the better fit. For bank statements, which a general table tool handles only roughly (wrapped descriptions, separate debit and credit columns, summary boxes), see bank statement converter: it compares the tools built for statements, which also check the totals.