PDF Automation: pypdf Read & Extract Text

By the end of this lesson you can:

• Open a PDF and read its page count and metadata with pypdf.
• Extract text from one page or a whole document.
• Search extracted text and pull out fields with regex.
• Recognise the limits of text extraction (scanned PDFs need OCR).

A PDF isn't a document of text — it's a set of drawing instructions: "put this glyph at x=72, y=640." There's often no concept of "paragraph" or even reliable reading order. Two PDFs that look identical can extract very differently.

pip install pypdf      # modern, pure-Python PDF library

Opening and inspecting

from pypdf import PdfReader

reader = PdfReader("invoice.pdf")
print(len(reader.pages))            # number of pages
print(reader.metadata.title)        # document title (may be None)
print(reader.metadata.author)
print(reader.metadata.creation_date)

reader.pages is a list-like of page objects; reader.metadata holds title/author/dates (any of which can be None — always guard).

Extracting text

# one page (0-based)
first_page = reader.pages[0]
print(first_page.extract_text())

# the whole document
full_text = "\n".join(page.extract_text() or "" for page in reader.pages)
print(len(full_text), "characters extracted")

Note the or "": extract_text() can return None for image-only pages, which would crash a join. Guard it. Pages are 0-based here (it's a Python list), unlike openpyxl's 1-based cells.

Encrypted PDFs

reader = PdfReader("protected.pdf")
if reader.is_encrypted:
    reader.decrypt("the-password")   # supply the password you were given
text = reader.pages[0].extract_text()

Many statements are password-protected. decrypt unlocks a PDF you have the legitimate password for — for reading your own documents, not bypassing protection you aren't entitled to.

Pulling out fields with regex

import re

text = reader.pages[0].extract_text() or ""

# find an invoice number like "INV-2026-0042"
m = re.search(r"INV-\d{4}-\d{4}", text)
invoice_no = m.group() if m else None

# find a total like "Total: $1,234.56"
m = re.search(r"Total:\s*\$([\d,]+\.\d{2})", text)
total = m.group(1).replace(",", "") if m else None

print(invoice_no, total)

Once you have the text, it's a string — your regex and parsing skills do the rest. This is how you read 500 invoices and pull every total into a spreadsheet automatically.

Goal: scan a folder of invoice PDFs, pull the invoice number, date, and total from each, write a CSV summary, and log any that don't parse — a genuine accounts-payable automation.

import re, csv, logging
from pathlib import Path
from pypdf import PdfReader

logging.basicConfig(level=logging.INFO, format="%(levelname)s %(message)s")
log = logging.getLogger("invoices")

PATTERNS = {
    "invoice_no": re.compile(r"Invoice\s*#?\s*([A-Z0-9-]+)"),
    "date":       re.compile(r"Date:\s*(\d{4}-\d{2}-\d{2})"),
    "total":      re.compile(r"Total:\s*\$?([\d,]+\.\d{2})"),
}

def extract(pdf: Path) -> dict | None:
    reader = PdfReader(str(pdf))
    text = "\n".join(p.extract_text() or "" for p in reader.pages)
    if not text.strip():
        log.warning("%s: no text (scanned image?) — skipping", pdf.name)
        return None

    record = {"file": pdf.name}
    for field, pattern in PATTERNS.items():
        m = pattern.search(text)
        record[field] = m.group(1).replace(",", "") if m else ""
    if not record["total"]:
        log.warning("%s: no total found", pdf.name)
    return record

def run(folder: str, out: str) -> None:
    records = []
    for pdf in sorted(Path(folder).glob("*.pdf")):
        rec = extract(pdf)
        if rec:
            records.append(rec)
            log.info("%s → %s", pdf.name, rec["total"] or "?")

    with open(out, "w", newline="", encoding="utf-8") as f:
        w = csv.DictWriter(f, fieldnames=["file", "invoice_no", "date", "total"])
        w.writeheader(); w.writerows(records)
    log.info("wrote %d invoices → %s", len(records), out)

run("invoices", "invoice_summary.csv")

INFO inv-001.pdf → 1200.00
INFO inv-002.pdf → 950.50
WARNING inv-003.pdf: no text (scanned image?) — skipping
INFO wrote 2 invoices → invoice_summary.csv

Read the code

Compiled regex patterns in a dict keep the extraction rules readable and easy to extend. The two guard clauses are the real lesson: an empty extraction means a scanned image (logged and skipped), and a missing total is flagged but doesn't crash the batch. The output is a CSV — straight into Lesson 15's tools or Lesson 19's Excel report. This turns hours of manual data entry into seconds, while honestly surfacing the documents a human still needs to look at.

pypdf's PdfReader opens a PDF: len(reader.pages) for the count, reader.metadata for title/author/dates (guard for None), and page.extract_text() for the text (also guard for None on image pages). Once you have text it's just a string — use regex to pull fields like invoice numbers and totals, always handling the "not found" case. Extraction is messy and order isn't guaranteed, so test against real samples; and remember scanned PDFs are images needing OCR, not text extraction. Pipe the results to CSV/Excel and you've automated document data entry.