Every resume you submit through a careers page is read twice: once by software, once — maybe — by a person. The software goes first, and its verdict decides whether the person ever sees your name. Yet almost nothing is published about how these parsers actually behave. So we tested them: 1,100 real resumes (donated, anonymized), four commercial parsers that sit behind a large share of Fortune 500 career sites, and a simple question — what survives?
What a parser sees
A parser does not see your layout. It extracts a text stream from the PDF, then runs segmentation models to label runs of that stream: this is a name, this is a company, these dates belong to that company. Everything you think of as design — columns, boxes, icons — exists only as its side effects on text order. If the extracted stream reads in the wrong order, the labels land on the wrong text, and your work history quietly reassembles itself into fiction.
The failures are invisible from your side. The PDF looks perfect. The confirmation email arrives. The parsed record — the only version of you the recruiter's software ranks — has your job titles attached to the wrong employers.
The failure modes, ranked
- Two-column layouts. 38% of them interleaved: parsers read straight across the page, stitching your sidebar skills into the middle of your job descriptions.
- Text boxes and shapes. Text inside drawing objects extracted out of order or, in one parser, not at all. Headers built as graphics simply vanished.
- Icon fonts. A phone glyph from an icon font extracts as a private-use codepoint — the parser sees garbage where your phone number's label should be, and two of the four then rejected the whole contact block.
- Tables for alignment. Cell-by-cell extraction turned "Company | City | Dates" rows into three disconnected fragments.
The PDF looks perfect. The parsed record — the only version the software ranks — is fiction.
What survives everything
Across all four parsers, one shape was effectively indestructible: a single column, real text, standard section headings, dates on the same line as the role. Zero structural misparses in our sample. This is not a coincidence — it's the layout the segmentation models were trained on, because it's the layout most resumes use. Boring is the moat.
The uncomfortable summary: the most important reader of your resume is a machine with narrow taste, and it is cheap to satisfy. Spend your creativity on the content. Let the layout be boring on purpose.