We Counted the Verbs on 800 Resumes. One Word Is on 45% of Them.

The word is **"Developed."** It shows up on 45.4% of the 800 real resumes we analyzed. That's 999 times in all, across the 14,669 lines those resumes break into. The rest of the top of the list is just as familiar: "Worked" (43%), "Created" (38%), "Performed" and "Provided" (both 37.1%), and "Managed" (34.9%). The verbs people reach for to describe their work turn out to be the same handful almost everyone else reached for.

These aren't bad words. The problem is that the ten most common verbs all describe tasks, what you did, not outcomes. And they almost never carry proof. Across all 14,669 lines, counting the section headers and skills lists that have nothing to quantify, just 2.3% contain a concrete number: a percentage, a dollar amount, or a multiplier. A verb this common does nothing to set you apart. Trading it for a stronger one only helps if the achievement underneath is real. In the next few sections we rank the full list, show which verbs come with numbers and which don't, and make the case that a precise, true rewrite beats a flashier verb every time.

Key points

• "Developed" appears on 45.4% of resumes. That's 999 occurrences across 800 documents. "Worked" (43%), "Created" (38%), "Performed" and "Provided" (37.1%) follow.
• Only 2.3% of the 14,669 resume lines we analyzed contain a real number behind the claim. A percentage, a dollar amount, or a multiplier.
• Verbs that name a result come with proof about 4x as often as verbs that name activity. 62.5% of "reduced" lines include a hard figure, versus 14.2% for "worked" and 7.1% for "demonstrated."
• The average score from a production ATS was 52.2 out of 100, and none of the 300 resumes we scored reached 80. ATS means applicant tracking system, the software that screens resumes. The weakest area by far: proving achievements with numbers (32.3/100). The strongest: keywords (70.7).
• Resumes with at least one real figure scored 4 points higher on average than resumes with none. The gap holds up statistically (details in the Methodology).

The 15 most overused resume verbs

Across 800 real resumes, the same short list of verbs opens bullet after bullet. These are the 15 that appear on the most resumes, ranked by the share of documents that use each at least once.

Almost every verb at the top of this list describes an activity, not a result. Developed, worked, created, performed, maintained, and managed all answer "what did you do?" They don't answer "what changed because you did it?" They're task verbs. They restate the duty from the job description and stop there. That's exactly why they cluster so tightly. Developed alone shows up on 45.4% of resumes and 999 times overall, and that's because it fits almost any role and commits to nothing specific. Swap one of these for another and most bullets read the same.

The verbs that actually name a result sit far lower on the list. "Increased" appears on just 17.1% of resumes, "improved" on 12.2%, and "reduced" on only 7.4%. Here's the trade-off: writers who reach for those verbs tend to back them up. 62.5% of "reduced" sentences include a hard figure, versus 18.3% for "developed." Across all 14,669 resume lines we measured, only 2.3% contain a number at all. The fix isn't to hunt for a flashier verb. It's to attach a real outcome to the one you already have.

The proof gap: claims without numbers

Across 800 real resumes and the 14,669 lines they break into, only 2.3% contain a concrete impact figure: a percentage, a dollar amount, a headcount, or a multiplier. Most of those lines are section headers and skills lists, which never carry a number by design. But even inside the achievement claims, where a number should live, hard evidence is the exception. This is the proof gap. Resumes are full of confident statements about what someone did, and nearly empty of proof for how much it mattered.

The verb a candidate reaches for predicts whether a number follows it. Outcome verbs sit next to figures far more often than effort verbs do. When we sort the lines by their leading verb, the spread is stark. The top of the table carries proof several times as often as the bottom.

The honest read is causal in one direction only. Verbs like reduced and improved name a change, and a change invites its own measurement. "Reduced" almost demands the reader ask by how much?, so writers tend to answer. Verbs like worked and handled describe activity with no built-in scale, so no number gets pulled along. That's correlation, not proof of causation. Swapping handled for reduced does not conjure a metric you never had. The lesson runs the other way: start from the result you can measure, then pick the verb that carries it.

What a production ATS engine sees

Frequency counts show what resumes say. To see how they hold up under evaluation, we ran a 300-resume sample through the production scoring engine behind Resumello's resume evaluation, an 8-dimension assessment of structure and content. The sample was drawn evenly from short, medium, and long resumes.

The results were not kind. The mean overall score was 52 out of 100 (median 56). Not one of the 300 resumes scored 80 or above. 31 scored below 40. And 227, three-quarters of the sample, landed in the 40–59 band: readable, parseable, and unremarkable.

The dimension detail tells the sharper story. The strongest content dimension across the sample was keyword richness, at 70.7/100. A decade of "optimize your keywords" advice has plainly landed. The weakest content dimension was quantified achievements, at 32/100. Job seekers have learned to name the right skills. They have not learned to prove their impact. That's the same proof gap the verb analysis found, measured a second, independent way.

The gap is worth points. Resumes containing at least one concrete impact figure (a percentage, dollar amount, or multiplier anywhere in the text) averaged 54.9 against 50.9 for those with none. That's about four points on a 100-point scale from the presence of a single number. The gap is small but statistically solid (the exact figures are in the Methodology), with the usual caveat: this is a pattern we observed, not a controlled experiment. Two dimensions (contact information and formatting) scored artificially low, and that's because the public datasets strip personal details and flatten layout. We exclude both from every claim above (see Methodology).

The verbs almost nobody uses

Some of the strongest verbs in the language are also the rarest. Across the 800 resumes in this study, spearheaded shows up on just 1.8% of them. Launched, drove, and expanded each land at 1.6%. Grew and cultivated sit at 0.8%. Pioneered and influenced trail at 0.5% apiece. These verbs describe starting something, owning it, and moving a number. And almost no one is using them.

That rarity is the opportunity. A recruiter has read "managed" and "developed" a hundred times in a morning. A resume that says it spearheaded a program or grew a segment reads differently. The word carries weight precisely because it's uncommon, and because it claims authorship rather than participation. Against a baseline where only 2.3% of all 14,669 lines in this study carry a concrete impact figure at all, a rare verb paired with a real number stands out twice over.

But the weight only holds if the claim is true. "Spearheaded" on a project you contributed to reads as inflation the moment anyone asks a follow-up question. A strong verb attached to nothing measurable is just a louder version of the same empty line. A verb upgrade is a differentiator, not a costume. That's why Resumello rewrites the wording only as far as what actually happened supports it: the right verb for the role you really played, backed by the figure you can actually defend in the room. Rare and true beats rare and borrowed every time.

Methodology

Data sources. Two public datasets hosted on Hugging Face, both accessed 2026-07-28: a categorized resume corpus spanning 24 job categories [1], and a resume-to-job-description matching corpus [2]. Neither dataset lists license metadata on its page. Both are widely used in public NLP work, and we use them here for aggregate statistical analysis only. No resume text is reproduced in this study. No Resumello user data was used. The product does not retain free-tool uploads, so there is nothing from our own users to draw on here, and we would not have used it if there were.

Preprocessing. Both corpora are heavily duplicated. Deduplication by exact text match reduced corpus [2] from 6,241 rows to 642 unique resume texts of at least 400 characters, and corpus [1] from 962 rows to 166 unique texts, of which 158 met the length floor and did not overlap the first corpus, for a combined N of 800. Text was segmented on line breaks, bullet markers, and sentence boundaries. Segments under 15 characters were discarded, yielding 14,669 segments.

Measures. For every action verb we report two numbers. doc_pct is the share of the 800 resumes containing the verb at least once (case-insensitive, whole-word match of the past-tense form, so "led" matches "Led" but not "misled"). impact_pct is the share of that verb's segments that also contain a concrete impact figure, defined exactly as: a percentage sign, a dollar sign, a quantity suffixed k/m/million/billion, or an Nx multiplier. Across all segments, 2.3% contain such a figure. The overall rate runs low because most segments are headers, skills lists, and contact lines that were never meant to carry one, and verb-bearing achievement claims quantify at the higher per-verb rates reported above.

ATS scoring. A stratified 300-resume subsample (100 drawn from each corpus length tercile, fixed random seed) was evaluated by the production scoring engine behind Resumello's resume evaluation: eight dimensions scored 0–100, plus an overall score. All 300 evaluations completed without error. The public corpora strip contact details and flatten formatting, so the contact-information and formatting dimensions score artificially low (means 10.8 and 35.2). We disclose them for completeness but base no claims on either.

Statistical notes. Overall scores: mean 52.2, median 56, SD 8.9. The comparison of resumes with at least one impact figure (n=97, mean 54.9, SD 6.5) against those with none (n=203, mean 50.9, SD 9.6) gives a difference of 4.0 points, 95% CI [2.1, 5.8] by Welch's method with a normal approximation. This is an observational association in a single corpus, not a causal estimate. Per-verb impact rates are reported only for verbs with at least 30 segments. No significance testing was applied to per-verb contrasts.

Limitations. The corpus skews technical, so verbs common in engineering and data roles are likely overrepresented against, say, nursing or sales. Segment splitting is heuristic. A run-on bullet can land two thoughts in one segment or split one across two. "Contains an impact figure" is a proxy for quantification, not a judge of it: a figure can be present but vague, and a strong result can be stated in words we do not count. Scoring reflects a single evaluation engine, and other ATS platforms weigh dimensions differently. We measured what we could count consistently across 800 resumes, and we're telling you exactly how.

Data availability. Both source datasets are publicly available at the references below. Aggregated statistics behind every figure in this study are available on request.

What to actually do with this

Three moves follow directly from the data:

1. Trade one task verb for one outcome verb, where it's true. "Reduced," "improved," and "initiated" carried a concrete figure three to four times as often as "worked" or "handled." The verb doesn't create the achievement, but the right verb pulls the number out of you.
2. Put one real figure in your top three bullets. A percentage, a dollar amount, a headcount, a time saved. The STAR method is the fastest structure for surfacing that figure from work you've already done. In our scoring sample that single habit separated the above-average resumes from the pack.
3. Steal from the empty column. The verbs almost nobody uses, spearheaded, launched, grew, are free differentiation for the people who can claim them honestly.

Or check where you stand in thirty seconds. The free ATS checker scores your resume on the same 8 dimensions used in this study, including the quantified-achievements dimension most resumes fail, with no signup. And if a verb in the overused list above is doing heavy lifting on your resume, its synonyms page shows exactly when each stronger alternative earns its place. For the structural side of getting past the filter, start with how to pass the ATS and the five mistakes that get resumes auto-rejected.

References

1. Resume Dataset (mirror of the widely circulated Kaggle "Updated Resume Dataset"). Hugging Face: `Sachinkelenjaguri/Resumedataset`. https://huggingface.co/datasets/Sachinkelenjaguri/Resumedataset. Accessed 2026-07-28.
2. Namuangtoun, C. Resume–Job Description Fit. Hugging Face: cnamuangtoun/resume-job-description-fit. https://huggingface.co/datasets/cnamuangtoun/resume-job-description-fit. Accessed 2026-07-28.

Suggested citation: Resumello Research (2026). We Counted the Verbs on 800 Resumes. One Word Is on 45% of Them. Published July 28, 2026. https://resumello.com/content/resume-verbs-study-2026