CVBranch research · Original data

What 2,156 IT resume bullets actually look like

Sample: 238 parsed IT resumes · 2,156 experience bullets · August 29, 2026

Generic resume advice says "use action verbs" and "add metrics." We wanted numbers. CVBranch builds a deterministic bullet scorer for software and IT CVs — action, impact, scale, leadership — so we ran it across a fixed research corpus: 238 parsed resumes (218 from an India Indeed NER dataset, 20 from US/mixed pipeline samples) yielding 2,156 experience bullets. This is not a hiring-outcome study and not representative of every industry. It is an honest baseline of how much proof IT resumes carry today — and why tailoring toward a job description should strengthen evidence, not invent it.

0%Median bullet quality score

28%Lines with a recognized action verb

0.1%Lines with measurable impact

96.7%Bullets scoring below 20%

Headline findings

Half of all bullets score zero on our rubric (median 0%, mean 4%). Most lines read as duties, fragments, or parse debris — not achievements.
Action verbs are common; outcomes are almost absent. 28% of bullets open with a past-tense verb such as "developed" or "implemented," but only 0.1% pair impact language with a defensible metric.
Skills often lack bullet proof. Across canonical skills detected on these CVs, 40% appear only in the skills section — not in experience bullets — while 48.4% have at least one bullet mention.
Parse shape matters. 54% of job entries had an empty bullets[] array with experience text living in a header block instead — so raw bullet counts under-state how much text exists and over-state how structured it is.

Quality score distribution

Each bullet receives a 0–100 score from CVBranch's scoreBulletQuality heuristic: action verb, measurable impact, leadership, technical depth, and scale signals. The histogram is harsh — by design. We prefer false negatives over cheering generic duty lines.

00-09%

1,480

10-19%

605

20-29%

30-39%

40-49%

50-59%

2,156 scored experience bullets · only 7 lines (0.3%) reached 40% or higher

Action without impact

The most frequent verbs are exactly what career guides recommend — yet the impact rate stays near zero:

developed — 157 bullets
implemented — 72 bullets
designed — 61 bullets
created — 41 bullets
deployed — 12 bullets

"Developed" and "implemented" describe activity. Recruiters and engineers interviewing you want scope and outcome: what shipped, for whom, at what scale, with what measurable effect. Our scorer treats a metric alone (e.g. a bare percentage) as weak unless impact language is present — only 0.5% of lines even contain a number, and almost none clear the full impact bar.

Weak vs stronger lines (anonymized)

Representative patterns from the corpus — employers and names removed:

Typical weak lines (score 0%)

"Application Development Associate" — job title pasted as a bullet
"Working on all major and minor enhancement requests as part of maintenance and support" — duty, no scope
"[Employer] — Bangalore, Karnataka —" — location fragment from parse
"Involved in analysis, design, development, integration and testing of application modules" — verb list, no outcome

Stronger lines from the same corpus (40%+)

"Saved 30% time and cost of testing by automation using Selenium WebDriver, Java, Cucumber" — metric + method
"Implemented master–slave architecture to improve Jenkins performance" — action + system + outcome direction
"Migrated millions of customers to Azure Cloud through FastTrack program" — scale + platform

The gap is not vocabulary — it is evidence density. Strong lines are rare in this sample (0.6% hit our "exceptional evidence" bar). That is an opportunity when you tailor your resume to a job description: pick the bullets that already contain facts and rewrite toward the posting, instead of adding skills you cannot discuss.

CV structure in the sample

2.7 jobs per CV on average (median 2)
9.4 bullets per CV in structured bullet arrays
7.5 bullets per job when the job entry includes a bullet list
7.7 canonical tech skills detected per CV (skills catalog matching)

Methodology

Corpus. India Indeed NER resume corpus (SRBHR); plus US / mixed pipeline parse samples. Parsed to CVBranch's CvMatchInput JSON; no live user uploads.

Bullet scoring. Deterministic scoreBulletQuality in @resume-ai/matching — same signals used in Composer and Matcher Lab. Noise filtered (very short lines, URL-only lines, skills-header debris). Scores are heuristic, not human-labeled at scale; a 40-bullet calibration set exists for regression tests only.

Skill evidence. Canonical skills from skills.canonical.json; each skill tagged as skills-section-only vs bullet/header proof via collectCvSkillEvidence.

Reproducibility. Regenerate stats with pnpm research:corpus in packages/matching (run-corpus-research.mts). Optional Ollama pass classifies sample lines for qualitative labels — not used in the headline percentages above.

Limitations. IT-biased; India-heavy in the larger set; parse quality varies; we do not measure interview rates, ATS pass-through, or recruiter preferences. Do not cite as "all resumes worldwide."

What to do with this

If you are applying to software and IT roles, assume your competition includes hundreds of duty-list bullets. Differentiation comes from verified outcomes on the lines you keep — and from matching those lines to each job description. CVBranch scores bullets while you tailor, surfaces skill gaps with evidence questions, and composes a one-page version for that application.

Related guides

See how your bullets score against a real JD

Tailor your resume with CVBranch

Free guest session · Built for software and IT CVs and job descriptions