How to convert PDF to CSV (free, and programmatic with Python and Node.js)

Converting a PDF to CSV means pulling the rows and columns of a table out of a PDF and writing them as comma-separated text, one line per row. The catch: a PDF does not store a table as a table. It stores glyphs at fixed x and y positions on a page, so every converter has to guess where the cells are. This guide covers the free browser method, three Python libraries, a Node.js approach, and what changes when the PDF is scanned.

If you only need a quick one-off and the PDF has a clean table, the fastest path is the free PDF to text tool: it runs in your browser, extracts the text layer, and you paste the result into a spreadsheet. For accuracy and automation, use Python.

Which PDF to CSV method should you use?

Pick your method based on the PDF type and whether you need to automate. Digital PDFs with ruled tables are the easy case and any tool handles them. Scanned PDFs need OCR first. Automation and large volumes favor Python.

Accuracy varies with table quality. Ruled tables with one value per cell extract cleanly; merged cells, multi-line cells, and multi-column layouts reduce accuracy for every tool.

How to convert a PDF to CSV online for free

Use a browser-based extractor when you have a single digital PDF and do not want to install anything. The pdf4.dev PDF to text tool reads the text layer with pdfjs-dist, entirely on your device, so the file never uploads to a server. That privacy property matters for invoices, payslips, and bank statements.

The trade-off: a text extractor returns the text in reading order, not a structured grid. For a simple table where each row sits on its own line, you paste the output into a spreadsheet and use "Text to columns" to split on the delimiter. For dense or multi-column tables, the column boundaries do not survive, and you should switch to Python.

If your goal is a real spreadsheet with data types rather than plain CSV, read how to convert PDF to Excel alongside this guide, since the extraction step is identical and only the write step changes.

How to convert a PDF to CSV with Python

Python has the strongest open-source table extractors. Camelot targets tables with ruling lines, pdfplumber gives you cell-level control, and Tabula wraps a Java engine with a one-line helper. All three write CSV without a paid API.

Camelot exposes read_pdf and a to_csv method per table, and its flavor argument picks the detection strategy. Use lattice for tables with visible lines and stream for whitespace-aligned columns.

import camelot
 
# lattice = tables with ruling lines, stream = whitespace-aligned
tables = camelot.read_pdf("report.pdf", pages="all", flavor="lattice")
 
for i, table in enumerate(tables):
    table.to_csv(f"table_{i}.csv")
 
print(f"Extracted {tables.n} tables")

Camelot and pdfplumber are both actively maintained and installable from PyPI. Camelot needs Ghostscript for its lattice mode; pdfplumber has no external dependency. For messy input, run both and keep the cleaner result.

How to convert a PDF to CSV in Node.js

Node.js has no table extractor as strong as Camelot, so you reconstruct rows yourself from positioned text. Read the text items with pdfjs-dist, group them by their y-coordinate into rows, sort each row by x, then join with commas. This is the same engine that powers the browser tool.

import * as pdfjsLib from "pdfjs-dist/legacy/build/pdf.mjs";
import { writeFileSync } from "node:fs";
 
const doc = await pdfjsLib.getDocument("report.pdf").promise;
const rows = [];
 
for (let p = 1; p <= doc.numPages; p++) {
  const page = await doc.getPage(p);
  const content = await page.getTextContent();
 
  // Group text items into rows by rounded y position
  const lines = new Map();
  for (const item of content.items) {
    const y = Math.round(item.transform[5]);
    if (!lines.has(y)) lines.set(y, []);
    lines.get(y).push({ x: item.transform[4], text: item.str });
  }
 
  // Sort rows top-to-bottom, cells left-to-right
  for (const y of [...lines.keys()].sort((a, b) => b - a)) {
    const cells = lines.get(y).sort((a, b) => a.x - b.x).map((c) => c.text);
    rows.push(cells);
  }
}
 
const csv = rows
  .map((r) => r.map((c) => `"${c.replace(/"/g, '""')}"`).join(","))
  .join("\n");
writeFileSync("output.csv", csv);

This handles the common case of one value per visual line. It does not infer merged cells or column spans, so wide or irregular tables still need manual tuning. The replace(/"/g, '""') step follows RFC 4180 quoting so values containing commas or quotes stay intact.

Why do PDF to CSV conversions break, and how do you fix them?

Conversions break because the extractor guesses cell boundaries and the guess fails on non-grid layouts. The four most common failures and their fixes:

Merged cells collapse into one column. Fix: switch from stream to lattice mode so ruling lines define the exact cell edges. Multi-line cells split into extra rows. Fix: post-process by merging rows that share a key column, or set a higher row tolerance in pdfplumber. Numbers import as text (currency, thousands separators, parenthesized negatives). Fix: clean the strings before writing, for example turning (1,200) into -1200. Wrong delimiter for your locale. Fix: set the delimiter explicitly, since RFC 4180 uses a comma but many European tools expect a semicolon.

For a deeper look at the extraction step shared with spreadsheet output, see how to extract text from a PDF and the complete guide to PDF conversion.

How do you convert a scanned PDF to CSV?

A scanned PDF is a picture of a page with no text layer, so extractors return empty rows. Confirm by trying to select text: if nothing highlights, it is scanned. You must run OCR to add a text layer before any table extraction works.

Tesseract is the free, open-source OCR engine and covers many languages. Its table structure detection is basic, so for dense scanned tables a paid service like Amazon Textract or Microsoft Document Intelligence returns cleaner cells, typically billed per page. After OCR produces a searchable PDF, feed it to Camelot or pdfplumber exactly as you would a digital PDF.

The best PDF to CSV conversion is the one you never run

Every method above is a reconstruction: the data started as structured rows, got flattened into positioned glyphs in a PDF, and now you spend effort guessing the grid back. If you own the document pipeline, keep the structured data as the source and generate the PDF from it, rather than parsing PDFs back into data later.

That is what PDF4.dev does. You send JSON to the render API, a Handlebars template lays it out, and you get a PDF. The CSV or database row that produced the document stays the canonical copy, so there is no round-trip to reverse.

curl -X POST https://pdf4.dev/api/v1/render \
  -H "Authorization: Bearer p4_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{"template_id": "invoice", "data": {"total": "1200.00"}}'

This does not replace PDF to CSV when you receive PDFs from third parties, and bank statements or supplier invoices still need extraction. But for documents you produce yourself, generating from data beats parsing it back every time.

Summary

For a digital PDF and a one-off, use the free PDF to text tool and split into columns. For accuracy and automation, use Camelot in lattice mode, or pdfplumber for irregular tables. In Node.js, reconstruct rows from pdfjs-dist text positions. For scanned PDFs, run OCR first. And when you control the source data, generate the PDF from it so you never have to convert back.