In a resume I set in Roboto and downloaded from Google Docs, the "fi" in "first" is stored as one character, U+FB01 (fi). The PDF looks normal, but a search for the word "first" in its extracted text finds nothing. I wrote up why "first" can break in a Google Docs PDF. This post looks at the same files from Python: what pypdf and pdfminer.six return, and a short check you can run on an uploaded PDF.
The files are one sample resume set in 32 fonts, each imported into Google Docs and downloaded as a PDF (the same file File > Download > PDF gives you). I'd already looked at what an open-source resume parser read from the same files. For this test, I ran extract_text() with default settings in pypdf 7.1.1 and pdfminer.six 20260107 on all 32 PDFs.
Ligatures: 16 of 32 fonts
16 fonts stored at least one letter pair as a single character. Both libraries returned the same ligatures, with the same counts, in every file. 13 of the 16 PDFs contained six ligatures each. Source Sans 3 had just one, an "ff".
The words it hit were ordinary resume words: "first" (30 times across the files), "first-response", "Shelfie", "conflict-free" and "office". In the 15 files where "fi" became one character, a regex for \bfirst\b matched zero times. In the other 17 it matched three times.
Unicode NFKC normalization fixes this. It maps U+FB01 back to "f" and "i", and after it every one of the 32 files matched "first" three times.
import unicodedata unicodedata.normalize("NFKC", "\ufb01rst") # 'first' If you're choosing a font for your own resume, nine passed every check in the parser test: the PDF stayed on one page in the font you picked, the text came out clean, and the parser read all 46 fields correctly. They are Arial, Book Antiqua, Calibri, Cambria, EB Garamond, Garamond, Georgia, Times New Roman and Trebuchet MS. A one-page checklist for a first resume covers what to include and how to write the bullets.
Private-use characters: Inter
Inter had no ligatures, but four of its characters came out as code points from Unicode's private-use area, which have no agreed meaning. pdfminer returned the phone number like this:
\ue081312\ue082 555\ue0880143 The brackets and the hyphen in (312) 555-0143 are U+E081, U+E082 and U+E088. The colon after "GPA" is U+E092. NFKC leaves private-use characters alone. After normalization, \(\d{3}\) \d{3}-\d{4} found the phone number in pdfminer's output for 31 of the 32 files. Inter was the one it missed.
One word per line from pypdf
In 31 of the 32 files, most lines of pypdf's default output held a single word:
checks first pdfminer.six joined the same words into normal lines, and pypdf with extraction_mode="layout" kept them on one line too. So with pypdf's default settings, a job title or a date range can span several lines.
In all 32 PDFs, pdfminer.six and pypdf's layout mode kept a space between "checks" and the next word, whether it came out as "first" or "first". The run-together "checksfirst" in the parser test came from how that parser put the text back together.
A check for uploads
Pass the PDF's file path to this script. It counts Latin ligatures (U+FB00 to U+FB06) and private-use characters (U+E000 to U+F8FF):
import re import sys import unicodedata from pdfminer.high_level import extract_text LIGATURES = re.compile("[\ufb00-\ufb06]") PRIVATE_USE = re.compile("[\ue000-\uf8ff]") text = extract_text(sys.argv[1]) ligatures = LIGATURES.findall(text) private_use = PRIVATE_USE.findall(text) print(f"ligatures: {len(ligatures)} {sorted(set(ligatures))}") print(f"private-use characters: {len(private_use)}") # NFKC turns U+FB01 back into "f" + "i"; it leaves private-use characters alone clean = unicodedata.normalize("NFKC", text) print("'first' found after NFKC:", bool(re.search(r"\bfirst\b", clean))) On the Roboto file it reports 6 ligatures (fi, fl and ffi). On Inter it reports 4 private-use characters. On Arial it reports none of either. The last line looks for "first" only because the sample resume contains it; swap in a word from your own file. Normalize with NFKC before you search, and if a phone number comes back with private-use characters like Inter's, flag that field for review.
These results come from one sample resume, one export path and two libraries. A different document or a later Google Docs version can come out differently, so run the check on your own files.
To check a finished resume without writing code, upload it to ReadableResume's free checker and see what an ATS parser reads from your resume.
I work on ReadableResume, a free resume template site. The 32 PDFs come from a font test we ran on one sample resume in October.
I used AI to help draft this post.

