We Analyzed 2,483 Real Resumes. Most Don’t Contain a Single Number.

I read resumes for a living. After a few thousand, you stop reading them like a hiring manager and start reading them like a pattern-matcher — because that is what a recruiter's first scan actually is. Six seconds, looking for two or three signals. Most resumes don't carry those signals, and they get passed over for reasons that have nothing to do with the person's actual ability.

I wanted to know whether that hunch held up at scale, so we ran the numbers. We took 2,483 real resumes spanning 24 different industries — from sales and finance to nursing, teaching, and construction — and measured each one with the exact detectors that power Resume Annex's scoring engine. No surveys, no opinions, no AI guessing. Just deterministic text analysis you can rerun yourself.

The results were worse than I expected.

54.8% of resumes don't contain a single number that matters

This is the headline, and it is the one worth sitting with. Across 2,483 resumes, more than half — 54.8% — did not contain one quantified result. Not one percentage. Not one dollar figure. Not one multiplier. The average resume carried just 1.85 quantified figures across its entire work history.

Here is why that matters. When a recruiter scans your bullet, "Responsible for managing the regional sales territory" and "Grew regional sales 34% to $4.1M in 18 months" describe the same job. One is a job description. The other is evidence. The first tells me what you were assigned; the second tells me what happened because you were there. In a six-second scan, evidence is the only thing that survives — and most resumes don't have any.

"Responsible for" is on 41% of resumes

If quantification is the thing resumes are missing, weak filler language is the thing they have too much of. 40.9% of resumes used the phrase "responsible for" — the single most common way to turn an accomplishment into a chore. And it is not alone: 78.3% of resumes used at least one weak opener from a list that includes "responsible for," "helped," "assisted," "worked on," "duties included," and "participated in."

These phrases all do the same damage: they put a passive frame in front of work that was probably active. "Helped reduce churn" hides whether you reduced it. "Assisted with the migration" hides whether you ran it. Recruiters have read these phrases ten thousand times, and they read them as a tell — the candidate either didn't own the outcome or can't articulate that they did.

Only 11.6% of statements open with a strong verb

We broke the corpus into 113,571 individual statements and checked the first word of each. Only 11.6% opened with a strong action verb (led, built, drove, delivered, reduced, increased, launched, and so on). A further 3.5% opened with weak filler. The remaining ~85% opened with something else entirely — a noun, a job title, a tool name, a date.

That last number is the quiet killer. A resume where most lines start with "Microsoft Excel," "Team of five," or "January 2021" reads as a list of facts about a job. A resume where lines start with verbs reads as a record of things a person did. The second one gets read to the bottom of the page. The first one gets skimmed and set down.

Which fields quantify — and which don't

The quantification gap isn't evenly spread. Fields that live and die by numbers — business development, banking, finance — quantify the most. Fields where the value feels harder to count — design, teaching, the arts — quantify the least. But notice that even the best category still has more than 1 in 4 resumes with zero numbers.

We ran the same analysis against a completely separate dataset of 940 resumes (a more IT-heavy corpus), and the pattern not only held — it got more extreme: 94.7% of those resumes had zero quantified results. Two different datasets, same conclusion. The quantification gap is not a quirk of one sample. It is how most people write resumes.

Why this happens

None of this is because job seekers are lazy or bad at their jobs. It's because nobody ever told them the rules, and the default way humans describe their own work is to describe their responsibilities — the things they were supposed to do — rather than their results. You list your duties because that's how you'd answer "what's your job?" at a dinner party. But a resume isn't answering that question. It's answering "what should we expect if we hire you?" — and only results answer that.

There's also a quiet myth doing damage here: the belief that an applicant tracking system auto-rejects you on keywords, so the game is keyword-stuffing. Recent studies show ATS rarely auto-reject on content. The real filter is the human doing the six-second scan. Optimizing for a robot that mostly isn't filtering you, while ignoring the recruiter who is, is exactly backwards.

How to fix it on your own resume tonight

You don't need a rewrite. You need a pass with three rules:

• Put a number on your top 5 bullets. Percentage, dollar, count, time saved, anything measurable. If you genuinely can't measure the outcome, measure the scope ("across 3 sites," "for a 40-person team"). Half of all resumes have zero numbers; adding five puts you ahead of the field instantly.
• Delete "responsible for," "helped," "assisted," and "worked on." Start the bullet with what you actually did: led, built, cut, grew, shipped, negotiated. The active verb usually makes the sentence shorter and stronger.
• Make every bullet open with a verb. If a line starts with a tool, a title, or a date, rewrite it so the first word is something you did. This single habit fixes the 85% problem.

That's the entire difference between a resume that gets skimmed and one that gets read. Not design, not length, not keywords — evidence and verbs.

Full methodology & limitations

I'd rather you trust this than take my word for it, so here's everything:

• Data: The primary sample is the public "livecareer" resume corpus (2,483 resumes, 24 categories) hosted on Hugging Face. The cross-check sample is a separate 962-resume dataset, of which 940 met our minimum-length filter. Both are public, anonymized resume collections — no Resume Annex user data was used.
• Detectors: "Quantified result" = a percentage, dollar amount, or multiplier (the regular expression and word lists are taken directly from our scoring engine). "Weak opener" and "strong verb" use the exact phrase lists in the product. We report results at the whole-resume level wherever possible, so the findings don't depend on how we split bullets.
• Limitations: Our quantification detector intentionally ignores scale numbers like team sizes, so the true "has a number somewhere" rate is a bit higher than 45% — we measured achievement metrics specifically. Public resume datasets skew slightly toward certain industries and may not perfectly represent every job market. And a resume that quantifies well can still fail for other reasons. The point of the study isn't that numbers are everything; it's that their near-total absence is the most common, most fixable problem in the pile.

If you want to see your own resume measured against these exact signals — quantification, verb strength, filler language — that's the free check below. It takes about ten seconds and it's the same lens I'm describing here.