How to write a Harvard-format resume that passes ATS in 2026

An Applicant Tracking System has three jobs. First, it parses your PDF — extracting text, identifying sections (Experience, Education, etc.), and structuring fields (dates, employer names, job titles, locations). Second, it matches the parsed data against the requirements of the job posting. Third, it ranks your application relative to others in the pool, surfacing the top N to recruiters.

The parsing step is where most resumes lose points. A 2024 survey by Jobscan tested 1,800 resumes uploaded by real candidates against Workday's parser and found that 60% had at least one field misparsed — most commonly job titles read as part of the previous employer's name, dates split across two roles, or entire sections (Skills, Education) skipped because of a layout choice the candidate considered cosmetic.

The candidate didn't get filtered. Their data got filtered. By the time a human sees the file, half the inputs the algorithm scored on never made it through the parser.

Once parsing succeeds, matching is mostly mechanical: the ATS tokenises the job description, builds a weighted vocabulary (required skills, preferred skills, must-have credentials), and computes a similarity score against your parsed resume. Modern systems use semantic embeddings, not just exact matches — «Python» in your resume matches «data engineering» in the JD because the embedding model knows they cluster. But that's gravy. The base game is still keyword recall, and recall depends on parsing.

Modern resume parsers were trained largely on US single-column chronological resumes — exactly the shape the Harvard format produces. A 2023 Lever engineering blog post (since taken offline but cached) explicitly mentioned that their parser model was fine-tuned on a corpus that included Harvard OCS sample resumes. This isn't conspiracy — it's just that the format is the de facto standard for the population of resumes the parsers were built to handle.

The properties that make the Harvard format ATS-friendly:

• Single column. Multi-column layouts confuse 90%+ of parsers — they read across, mixing fields from different sections.
• Section headers in plain text (Education, Experience, Skills) match the parser's vocabulary directly. Custom or stylised section names risk being missed.
• Reverse-chronological ordering matches the parser's expected sequence; out-of-order entries are sometimes misdated.
• Times-family serif body font. Most parsers run OCR as a fallback when text extraction fails; serif fonts have better OCR recall than thin sans-serifs.
• No embedded images, icons, or graphics. Anything inside a PDF image layer is invisible to a text parser. Skill bars, headshots, and icon-prefixed contact info are pure noise.
• Standard date formats (Sep 2024 — May 2026). The parser has regex patterns for the common formats; deviating breaks them.
• Consistent hierarchy: institution bold, role italic, dates right-aligned. The parser uses font weight and position as structural cues.

Other resume styles can also be ATS-friendly — but you have to choose every property deliberately. The Harvard format gives you all of them by default.

The single highest-leverage move you can make for ATS scoring is tailoring your resume's vocabulary to the specific job description. This is not keyword stuffing — it's aligning your language to the vocabulary the ATS will tokenise from the JD.

Treat the job description as a vocabulary list. Your resume needs to use the same words. Synonyms are a tax.

Doing this thoroughly for every application is what separates candidates with 5% callback rates from candidates with 25% callback rates. It takes 20 minutes per application and changes the trajectory of your search.

Harvard's X-Y-Z formula — accomplished X, as measured by Y, by doing Z — was designed for human readers, but it happens to be optimal for ATS parsing too. Parsers run named-entity recognition (NER) on bullets to extract actions, quantities, and tools. The X-Y-Z bullet packs all three in a deterministic order, which is exactly what an NER model wants.

The strong bullets aren't just better for humans — they pack five to seven extractable entities each (numeric, organisational, temporal, technical), versus zero in the weak versions. ATS scores rise mechanically.

• Headers and footers. Most ATS parsers ignore header/footer fields entirely. If your contact info is there, it's invisible.
• Text inside images or PDF graphics. Anything in a non-text layer is invisible to the parser.
• Multi-column layouts. The parser reads left-to-right across both columns, jumbling fields.
• Tables (even invisible ones used for layout). Parsers extract row by row, mangling the structure.
• Custom section headers ('My Journey' instead of 'Experience'). The parser's section-recognition vocabulary is small.
• Date ranges with em-dashes encoded as Unicode rather than ASCII (— vs -). Some parsers tokenise on ASCII hyphens only.
• Special characters in employer names ('Goldman Sachs & Co.' with a typographic ampersand). The parser may split on it.
• Resumes saved as 'Print as PDF' from Word with image-quality JPEGs of the text instead of selectable text. Always export as PDF/A or use a vector renderer.
• Embedded fonts the parser's environment doesn't recognise. Stick to Times, Garamond, Georgia, Cambria, Calibri, Arial, Helvetica.
• More than two pages. Many ATS truncate beyond page 2 — your second page never gets scored.

Before submitting to a real job, test your resume:

The Harvard-format resumes our editor produces have been tested against Workday, Greenhouse, Lever, iCIMS, Taleo, SmartRecruiters, Jobvite, Ashby, and Recruitee. The PDF passes parsing in every case without modification.