Everyone tells you to "use keywords from the job description." Almost nobody can tell you which words actually recur across real listings — because almost nobody has a job-listing corpus to measure. We do. ResumeGeni's live corpus holds roughly a million active job listings; on July 28, 2026 we measured a random 8,000-listing sample to see what employers actually write, and what that means for your resume.
1. Compound terms are what listings actually ask for
ATS keyword matching is literal: "management" on your resume does not credit you with "project management" when the listing asks for the compound. And listings ask for compounds constantly. Here is the measured share of listings containing each exact phrase:
| Compound phrase | % of listings | The stem alone appears in |
|---|---|---|
| communication skills | 34.9% | 53.9% ("communication") |
| customer service | 29.7% | 53.5% ("service") |
| cross-functional | 13.7% | 18.7% ("functional") |
| problem solving | 5.1% | 17.6% ("solving") |
| project management | 4.6% | 52.1% ("management") |
| supply chain | 4.3% | 4.9% ("chain") |
| process improvement | 3.9% | 14.3% ("improvement") |
| data analysis | 2.6% | 15.2% ("analysis") |
| machine learning | 2.4% | 16.8% ("learning") |
| quality assurance | 2.1% | 4.0% ("assurance") |
Read the "project management" row carefully. Over half of all listings say "management" somewhere — but only 4.6% mean project management specifically. If the listing writes the compound and your resume only has the stem, a literal matcher scores that as a miss. The reverse inflation — assuming the stem covers you — is exactly the false confidence a resume checker exists to remove.
2. The same role, different vocabularies
Employers describing the same job split into distinct dialects, and the split differs by field. Measured in job titles within the same sample:
- Nursing — a near-even split. "RN" appears in 84 titles; "registered nurse" spelled out appears in 74. The same license, two dialects, each covering roughly half the market. A resume that only ever says one of them literal-matches only half of the listings it could.
- Software — one dialect dominates. "Software engineer" appears in 261 titles versus 12 for "software developer" and 4 for "programmer." Body text shows the same skew: 5.3% of all listings mention "software engineer" against 0.3% for "software developer" and 0.2% for "programmer." Here, mirroring the dominant term matters more than covering every synonym.
- Trucking patterns the same way ("truck driver" vs "CDL driver" vs "Class A driver"), but our sample held too few trucking titles (53) to publish a reliable split — an honest limit of this measurement.
The practical rule falls out directly: check which dialect your target listings speak before assuming your field behaves like someone else's.
3. Certifications appear in a minority of listings — but cluster hard by field
Across the full sample, 26.8% of listings mention a certification or licensure signal (certification/licensure language, CDL, PMP, CPA, CISSP, CompTIA, Six Sigma, CPR/BLS/ACLS, AWS Certified, and similar). The field split is where it gets useful:
| Field (by title) | Listings in sample | With a cert/license mention |
|---|---|---|
| Healthcare/nursing | 426 | 65.3% |
| Trucking/CDL | 53 | 39.6% |
| Software/tech | 1,262 | 25.6% |
| Everything else | 6,259 | 24.3% |
In licensed fields, credentials aren't a bonus section — they're the gating vocabulary of the listing itself. In software, a certification appears in only one listing in four; skills language carries far more weight.
What this means for your resume
- Mirror compound phrases exactly as the listing writes them — once, in context, not stuffed. "Management experience" is not "project management experience."
- Check the listing's dialect before tailoring. In nursing, cover both "RN" and "registered nurse." In software, say "software engineer" unless the listing says otherwise.
- Weight certifications by field. In healthcare and trucking they gate the match; in software they're supporting cast.
- Measure instead of guessing. The free checker scores format, content, and keyword categories 0–100 against this same research corpus (methodology) — and our companion study, Inside a Resume Checker, shows keyword gaps are 44% of everything it flags.
Methodology
Random sample of 8,000 active English-language job listings drawn July 28, 2026 from ResumeGeni's live corpus (~1 million active listings of 4.6 million collected; Postgres TABLESAMPLE, active status, non-empty descriptions). Phrase shares are document frequencies: the percentage of listings containing the exact phrase at least once, case-insensitive. Title-dialect counts are raw title matches within the sample. Corpus collection and normalization are described on the research data dashboard. Limits: the live corpus skews toward large-volume employers (national retail chains contribute heavy boilerplate, which is why we report curated phrase measurements rather than raw bigram rankings); counts describe listing language, not interview outcomes; the trucking sample was too small for a dialect split, as disclosed above. See also our keyword benchmarks and the practical guide to finding and using ATS keywords.
