Fonts that break resume parsing

Font advice for resumes is almost entirely about taste: serif or sans, 10pt or 11pt, avoid Comic Sans. There is a second, stranger category of font problem that has nothing to do with how the page looks and everything to do with what comes out of it. These are the ones worth knowing about, because you cannot catch them by proofreading.

1. A font that embeds correctly and still loses characters

We shipped a typeface called Carlito for a while — the metric-compatible clone of Calibri, properly licensed, correctly embedded. The purchased PDF was flawless. The in-app preview, which we build by rasterizing that same PDF, rendered the buyer’s name as Sen or Backend Eng neer: the subset dropped the letter i, the full stop, and the digits 2, 3, 4, 5, 7 and 9.

Everything that could have caught it passed. Text extraction passed. The scorer passed. The font licence check passed. The failure lived in one specific consumer of the file — and the lesson generalises well past our pipeline: a font can be valid, licensed, embedded, and still be handled differently by the next program that opens your PDF. We removed it and shipped Open Sans instead.

What this means for you is narrow but useful: prefer typefaces that have been through millions of documents — Arial, Helvetica, Calibri, Georgia, Times, Open Sans, Inter, Lato, Source Sans. Not because obscure fonts are unprofessional, but because obscurity is untested.

2. A typographic refinement that deletes every digit

font-variant-numeric: tabular-nums makes figures line up in columns. It is the right choice for a table of numbers, and under some variable fonts — Inter among them — a resume set with it renders every digit correctly and extracts with none of them. Dates, percentages, headcounts, revenue figures: absent from the text a parser receives, present on the page you proofread.

You cannot set this property yourself in most builders, but you are affected by it if your builder does. It is one of three CSS rules we forbid outright in our own templates, with a test that fails the build if one comes back.

3. Letter-spacing that splits words into letters

Tracked-out uppercase headings — E X P E R I E N C E — are a design cliché for a reason: they look considered. Past roughly 0.08em the extractor stops treating the glyphs as one word and emits them individually, which means the heading a parser is searching for does not exist in the text.

And 0.08em is a ceiling, not a guarantee. Bold uppercase Merriweather split into one run per character at 0.08em, so our editorial template sets 0.03em instead. The general rule: any new typeface has to be measured, not reasoned about from the number.

4. Ligatures, which nobody can fully avoid

Most good text fonts draw fi, fl and ffi as single glyphs. Some extractors reconstruct the letters from the font’s character map; others infer letters from glyph positions and produce Certi fi ed. You can see this in our own extraction dump on what an ATS reads from your PDF, which we publish rather than clean up.

This is not usually fatal — a recruiter searching certified is one of many ways you get found — but it is the clearest illustration of the general principle: the text in your PDF is reconstructed, not stored, and reconstruction has edge cases.

Practical rules

• Pick a common font. Boring is a feature here. If the builder offers a font you have never seen in a document, that is a reason to skip it, not a reason to try it.
• Do not use icon fonts for contact details. A phone glyph from an icon font extracts as a private-use character or as nothing, and if the number lives inside the same run it can go with it. Write +91 98765 43210, not a symbol.
• Avoid all-caps for anything that must be matched. Your own name, section headings and job titles are worth setting in normal case; the tracking that usually comes with small caps is what breaks them.
• Never accept a preview you cannot read. If your builder shows you a rendering, look at it properly. Ours exists because a PDF can be perfect while the picture of it is missing every i.
• Extract your finished file and read it. It is the only check that covers all four of these at once.