# The Experience Bullet Standard, Graded Line by Line Before It Ships

*Grade every experience bullet pass or fail on seven fixed tests and every role block on lead and count, with verdicts a second reviewer will reproduce.*

- Canonical URL: https://www.refolk.ai/candidates/guides/experience-bullet-standard-line-grading
- Pillar: Positioning and materials
- Format: Standard
- Published: 2026-10-22
- Last reviewed: 2026-10-22
- Reading time: 22 min
- Keywords: how to tell if resume bullets are good, resume bullet points accomplishments vs duties, resume bullets without metrics, how many bullet points per job on resume, resume bullets sound generic AI written, resume bullet checklist

## Key takeaways

- An experience bullet passes only if it clears all seven binary tests: title-inferable, owned action, result or substitute evidence, baseline, specific noun, length, and sourced number.
- First-scan studies put the initial resume look at 6 to 11.2 seconds, mostly on titles, companies, and dates, so bullets get graded in the second pass, where InterviewPal measured a 1:34 median total review.
- Laszlo Bock's 2014 XYZ post explicitly asks you to provide a baseline for comparison, and Jobscan now lists missing context behind impressive numbers as a tell of AI-written resumes.
- Hiring managers punish generic writing more than AI use: 19.6% would reject a resume they believed was fully AI-generated, but 62% would reject AI resumes that lack personalization.
- In Refolk's index, Talent Acquisition Partner is the top recruiter title in 14 of 25 UK profiles versus 2 of 25 US profiles, so run the title-inferable test using your target market's wording.
- Count bands of 4-6 bullets for the most recent role, 3-4 for the previous one, and 1-3 for older roles only hold if you count bullets that pass all seven tests, never padding duties.

This is the pass or fail standard for the experience bullets on a resume: seven tests applied to every line, then two rules applied to every role block. It is for job seekers rewriting their own resume and for anyone reviewing one for them, and it is written so two graders working separately reach the same verdict on the same line. It does not cover parsing, layout, or interview prep. It decides whether each line on the page earns its space.

## What does a passing experience bullet have to prove?

A passing bullet proves that this specific person did a specific thing with a checkable effect, in words a stranger could not have produced from the job title alone. Seven binary tests check that, and a bullet that fails any one of them gets rewritten or cut.

The seven line tests:

1. **Title-inferable.** A stranger given only your job title could not have written this bullet.
2. **Owned action.** The bullet opens with a verb you personally performed.
3. **Result or substitute evidence.** The bullet names a metric or a specific substitute evidence type.
4. **Baseline.** Every number has a "from" value, a denominator, or a period.
5. **Specific noun.** The bullet names a system, artifact, audience, or count that AI could not have invented.
6. **Length.** One line, two at most.
7. **Sourced number.** You can name the artifact each number came from and explain how you reached it.

Then two role-block rules:

- **Lead.** The first bullet under each role is the strongest relevant passing bullet.
- **Count.** The number of passing bullets falls inside the band for that role's age.

### Why binary tests instead of a graded scale

Career-center rubrics grade on scales, and their own wording builds in disagreement. Two graders can read the same bullet against these columns and land a full level apart.

| Rubric | Column wording | Where graders split |
|---|---|---|
| ULM | Outstanding: "Quantifies and/or qualifies results, when applicable" | "When applicable" is undefined; "qualifies" lets almost anything pass |
| ULM | Average: flags "Duties include" or "Responsible for" | Clear, but only catches the most obvious duty phrasing |
| WPI | Outstanding: "Highlights relevant achievements & provides concrete results" | "Concrete" is never defined |

A scale is useful for teaching. As a gate, it doesn't work. Each test below is phrased as a question with a yes or no answer and a stated way to check it, so the discretion moves out of the grader's head and onto the page.

> **Rule:** One failure, one verdict
>
> A bullet that fails any of the seven tests does not ship as written. There is no partial credit: rewrite it until it passes all seven, or cut it.

## When do recruiters actually read your bullets?

Not in the first scan. The first-scan studies find reviewers spending a handful of seconds on titles, companies, and dates. Bullets get read in a second pass, and that pass checks them more than it skims them.

| Study | Sample | Method | Initial scan |
|---|---|---|---|
| Ladders, 2012 | about 30 recruiters (secondary report) | eye-tracking | 6s |
| Ladders, 2018 | not disclosed | two-stage, eye-tracking | 7.4s |
| ResumeGo, 2024 | 418 hiring professionals | self-report survey | 47% say 30-60s |
| InterviewPal, 2025 | 4,289 reviews, 312 reviewers | cursor tracking | 11.2s |
| Wonsulting, 2025 | 2 recruiters | hidden eye-tracker | under 10s |

What these studies say about bullets specifically:

- **The first scan barely touches them.** In the 2012 Ladders study, 80% of the six seconds went to a few data points: name, current title, previous titles, employment dates, and education. The 2018 gaze order ran from current title and company, to the previous one, to the dates, checking for steady progression.
- **The second pass is where bullets get graded.** InterviewPal logged a 1:34 median total review, with 38% of attention on the Experience section and a reported 27% increase in reading time for metric-based achievements. That study measured cursor movement in an AI-assisted interface, not eyes, so treat it as directional.
- **Self-reports run longer than measurements.** In ResumeGo's survey, only 1% of respondents said they spend less than 10 seconds. People describe their careful reads, while the instruments catch the first pass.

Some things are not established. No public study measures how long reviewers spend on the first bullet versus later ones, and none measures read rates by bullet length. The lead and length rules in this standard rest on convention and on the logic of verification, not on measured attention. I'd rather say that than dress up a convention as a finding.

#### Where a bullet gets judged

1. **First scan** - Titles, companies, and dates; bullets mostly unread
2. **Second pass** - Bullets read; numbers weighed against their context
3. **Shortlist** - Lead bullets decide whether the role reads as relevant
4. **Interview** - Each number and claim gets asked about directly

*Bullets are skipped in the first scan, checked in the second pass, and defended in the interview, so every line should be written to survive the check.*

The practical consequence is that a bullet doesn't need to grab attention. It needs to hold up when someone reads it closely and then asks about it.

## The seven line tests, one at a time

Each test is a yes or no question with a defined way to check it. Run them in order. Record every failure, not just the first, so the rewrite fixes everything in one pass.

### 1. Title-inferable: could a stranger write this from the job title alone?

Cover everything on the page except the job title and read the bullet. If someone who has never met you could have written it from the title alone, it fails. This is the test that separates accomplishments from duties.

- **What it proves:** the line describes you, not the role.
- **What it looks like when it lies:** rare wording wrapped around an ordinary duty, such as "Orchestrated stakeholder alignment." Rewrite it in plain verbs. If the plain version is a duty, the bullet fails.

### 2. Owned action: did you personally do the verb?

The bullet opens with a verb that you, not your team, performed. Automatic fails: "Responsible for", "Duties include", "Helped", "Worked on."

- **What it proves:** the credit belongs to you.
- **What it looks like when it lies:** a strong verb covering team credit, such as "Launched." Ask "what did you personally do?" If the honest answer is "attended" or "reviewed," it fails.

### 3. Result or substitute evidence: what changed, and how do we know?

The bullet names a metric, or one named substitute evidence type (see the next section). Write the evidence type next to the bullet on the grading sheet. If you can't name one, it fails.

- **What it proves:** something happened because of the action.
- **What it looks like when it lies:** scope with no change attached, such as "Managed 30 accounts." That is context, not a result.

### 4. Baseline: compared to what?

Every number on the line carries a "from" value, a denominator, or a time period. Laszlo Bock's 2014 LinkedIn post introducing his XYZ formula tells readers to "provide a baseline for comparison." Most retellings of the formula drop that instruction. A line with no numbers passes this test by default, but it still has to pass test 3.

- **What it proves:** the size of the change means something.
- **What it looks like when it lies:** "Increased conversion 40%" from a base of 5 users.

### 5. Specific noun: what did you do it to?

The bullet names a system, artifact, audience, or count that a language model could not have invented without your history. "Processes," "stakeholders," and "initiatives" don't count.

- **What it proves:** the line came from a real job, not a template.
- **What it looks like when it lies:** a tool name dropped in to look specific, such as "using Excel." Ask whether the noun changes what the reader believes about the result. If it doesn't, it fails.

### 6. Length: does it fit on one line?

Indiana State's career guide says to "use a 1-sentence phrase starting with an action verb." A Harvard-style guide (not Harvard's own) says "one line where possible, two lines at the absolute maximum." Ladders' 2018 study found that resumes with long sentences did badly. Three lines fails.

- **What it proves:** the claim is clear enough to state in one go.
- **What it looks like when it lies:** a one-liner that only fits because the baseline or the specific noun was cut. Fix that by trimming adjectives, not evidence.

### 7. Sourced number: where did this number come from?

For each number on the line, name the artifact it came from: a dashboard, a report, a ticket export, a review, an email. The defensibility test from practitioner guidance applies: "If you could not walk an interviewer through how you got the number, do not use it."

- **What it proves:** the claim will hold up in the interview.
- **What it looks like when it lies:** an estimate presented as a measured fact.

#### Anatomy of a passing bullet

1. **Owned verb** - What you personally did
2. **Specific noun** - The system, artifact, audience, or count you did it to
3. **Evidence** - What changed, as a metric or a named substitute
4. **Baseline** - The from value, denominator, or period
5. **Source** - The artifact the number came from, kept off the page

*The verb, noun, evidence, and baseline sit on the page; the source sits behind it, ready for the interview question.*

## Resume bullets without metrics: what counts as substitute evidence?

A bullet with no metric can still pass test 3 if it names a substitute evidence type: scope paired with a change, frequency, audience, before-and-after, adoption, speed, or quality. Scope on its own is context, not a result, and it never passes test 3 alone.

Career-office guidance mostly stops at prompts. JHU's Resume Power Bullets suggests asking whether you saved money, reduced costs, increased sales, or saved time. Most of the method for jobs without clean metrics comes from practitioners and vendors, not career offices, so treat these as accepted practice rather than research findings.

Substitute evidence types, and what each one proves:

- **Scope with a change.** Team size, budget size, number of accounts, transaction volume, or ticket volume, plus what changed within it. Proves scale. With no change attached, it's only context.
- **Frequency reconstruction.** A count rebuilt from how often something happened over a known period. Practitioner guidance calls this "a legitimate reconstruction, not an invented figure, as long as the underlying frequency is accurate." Proves volume.
- **Before-and-after.** The state before and after, stated without a percentage. Proves a change happened when you can't size it precisely.
- **Audience.** Who used, received, or approved the work. Proves reach and trust.
- **Adoption.** Whether others took the work up after you shipped it. Proves the work outlasted you.
- **Speed.** How much faster something happened, with a baseline. Proves efficiency.
- **Quality.** Fewer errors, fewer escalations, fewer rework cycles, with a baseline. Proves reliability.

How much of a resume should be quantified isn't settled. One vendor recommends that at least 50-60% of bullets be quantified, another puts the number on the first two bullets under each role, and no career office gives an accomplishment-to-context share at all. This standard sets no percentage. Every passing bullet needs evidence, and metrics are one kind of evidence among several.

The hard part of tests 3 and 7 is usually finding the evidence, not writing it down: the old report, the ticket counts, the email where someone said the new process cut their week in half. [Refolk](/candidates) writes the resume from your own history, so each figure starts out attached to the record it came from. That's the record test 7 asks you to name.

> **Tip:** Mine artifacts before you draft
>
> Sources disagree on drafting order: JHU builds up from duties, Bock leads with the outcome. Collect the numbers from artifacts first, because test 7 will send you back for them anyway.

## How many bullet points per job, and which one leads?

Lead each role with its strongest relevant passing bullet, and carry 4-6 passing bullets on the most recent role, 3-4 on the previous one, and 1-3 on older roles. Only bullets that pass all seven line tests count toward the band.

| Source | Guidance |
|---|---|
| Indiana State career center | 2-4 bullets per job or company |
| UC Davis (secondhand report) | 2 to 5 accomplishment statements per experience |
| Ranking-page convergence | 4-6 most recent, 3-4 previous, 1-3 older |
| AAAS S-STEM rubric | "Bullets are listed in order of importance" |

These bands are convention, not measurement. Nobody has published read rates by bullet position, so the bands encode a shared expectation among reviewers, not proof that bullet five goes unread. I use the ranking-page bands because they scale with recency, and recent roles are where the 2018 gaze data shows reviewers looking first.

### The role-block rules

- **Lead rule.** The first bullet is the passing bullet most relevant to the job you're applying for, not the most impressive one in general. It passes if a second grader, given the posting, would pick the same lead.
- **Count rule.** Count the bullets that passed all seven tests. The role passes if that count falls inside its band. Bullets that failed don't count, even if they're still on the page.
- **Context line.** A scope-only line ("Managed 30 accounts") may stay in a block as context. My policy is at most one per role, never in the lead position, and never counted toward the band.

#### Line triage before the role block

Horizontal axis runs from A stranger could write it to Only you could write it. Vertical axis runs from No evidence to Sourced result.

| Quadrant | What it means |
| --- | --- |
| Generic win | Add the specific noun, then re-grade |
| Ships | Count it and compete for the lead slot |
| Duty | Cut it, or rebuild it from an artifact |
| Context | Keep at most one per role as scope, never first |

*Only bullets that are both specific to you and backed by evidence count toward a role's band; everything else gets rebuilt, demoted, or cut.*

> **Watch out:** Padding to hit the band
>
> Reaching six bullets on your current role by adding duties fails the count rule. Three passing bullets beat six where three are filler, because the filler is what a close reader remembers.

## Why do resume bullets sound generic or AI-written?

They sound generic because they leave out what only you know: the named system, the real audience, the number with its baseline. Hiring managers object to that generic quality far more than to AI use itself, and tests 1 and 5 are built to catch it.

What hiring managers report noticing, in Resume Genius data:

- 51%: unnatural phrasing or tone
- 44%: repetitive or overly generic language
- 41%: vague or inflated descriptions
- 41%: buzzword-heavy writing

In the same data, 80% of hiring managers say they can often tell when a resume is AI-written. An Insight Global figure reported by Jobscan puts those who can tell immediately at 53%. Jobscan's own list of tells adds identical sentence structure across all bullets and "missing context behind impressive numbers." That second tell is test 4.

The rejection numbers point at specificity, not authorship. In TopResume's survey of 600, 19.6% would reject a resume they believed was fully AI-generated. In Resume Now's survey of 925, 62% would reject AI resumes that lack personalization, and 78% value personalized detail. In a separate survey of 929, 61% said AI resumes make candidates look more qualified than they are, and 49% said candidates frequently can't back up those claims. Reported rejection rates for AI-written resumes run from about 20% to 49%, mostly from secondary or vendor sources, so don't build policy on any one of them. One thing isn't established: no survey measures copying the posting's language word for word as a tell.

> A bullet a stranger could write from your job title alone is a bullet about the job, not about you.

### The title test depends on your target market

Test 1 asks what a stranger would infer from a title, and the same function goes by different titles in different markets. Refolk's index of professional profiles shows this clearly for recruiting itself, which is also the job of the people grading your resume.

| Measure | US | UK |
|---|---|---|
| Recruiter-titled profiles | 122,927 | 9,391 |
| Talent Acquisition Partner, top title (n=25) | 8% | 56% |
| Technical Recruiter share (n=25) | 32% | 12% |
| Resume or career coach profiles | 12,623 | not sampled |

The shares come from 25-profile samples, so read them as direction, not precise splits. Even so, the direction matters. A UK reviewer and a US reviewer picture different duties when they read the same title, so run the title-inferable test using the title wording your target market uses. If the posting's market calls the role something different from what's on your resume, imagine the stranger reading the posting's title, not yours.

**122,927 - US profiles titled Recruiter, Technical Recruiter, or Talent Acquisition Partner in Refolk's index**

That is 9.7 recruiter-titled profiles for every resume writer or career coach profile, so far more people grade bullets than help write them.

The coach count includes some off-target titles picked up by keyword noise, which inflates it, so the real ratio of graders to helpers is probably higher than 9.7. Most people who read your bullets have never been told what a good one looks like. A binary standard gives every reader the same bar.

## The grading procedure, step by step

Grade the whole resume in one sitting: extract every bullet, run the seven line tests in order, then grade each role block. Two graders working independently, then reconciling every split, is the strongest version.

#### Grade every bullet, then every role block

1. **Extract every bullet into one list** - Copy every bullet into a single list tagged by role ID, about 10 minutes. Done means each bullet has its own row and a role ID.
2. **Run the title-inferable test** - Cover everything except the job title and ask whether a stranger could have written the bullet, using the target market's title wording. Done means each bullet is marked pass or fail.
3. **Check for an owned action** - Confirm each bullet opens with a verb you personally performed. Done means no Responsible for, Helped, Worked on, or Duties include.
4. **Name the result or substitute evidence** - Find a metric or a named substitute: scope with a change, frequency, audience, before-and-after, adoption, speed, or quality. Done means an evidence type is written next to every bullet.
5. **Demand a baseline for every number** - Find the from value, denominator, or period for each number. Done means no bare percentage or count remains.
6. **Confirm a specific noun** - Check that each bullet names a system, artifact, audience, or count AI could not have invented. Done means the noun is recorded and it changes what the reader believes.
7. **Check length and source each number** - Confirm one line, two at most, and name the artifact behind every figure. Done means you could explain how each number was reached.
8. **Grade each role block on lead and count** - Put the strongest relevant passing bullet first and count only passing bullets against the band, about 5 minutes. Done means every role passes both rules.

Use this sheet for each bullet, then once for each role. The two-grader column is what makes the result reproducible.

**Bullet grading sheet**

```
ROLE ID: [R1]   COMPANY / TITLE / DATES: [ ]   TARGET MARKET: [US / UK / other]
BULLET (verbatim): [ ]

T1 Title-inferable - a stranger could NOT write it from the title:   PASS / FAIL
T2 Owned action - opening verb is mine personally:                  PASS / FAIL
T3 Evidence type: [metric / scope+change / frequency / audience / before-after / adoption / speed / quality / none]   PASS / FAIL
T4 Baseline for every number (from / denominator / period):         PASS / FAIL / no numbers
T5 Specific noun: [name it]                                          PASS / FAIL
T6 Length: [1 / 2 / 3+ lines]                                        PASS / FAIL
T7 Source artifact for every number: [name it]                       PASS / FAIL / no numbers

VERDICT: SHIP / REWRITE (tests failed: [ ]) / CUT / CONTEXT LINE

ROLE BLOCK [R1]
Lead bullet is the strongest relevant passing bullet:   PASS / FAIL
Passing bullets: [n]   Band: [4-6 most recent / 3-4 previous / 1-3 older]   PASS / FAIL
Context lines: [0 / 1]   Not in lead position:   PASS / FAIL

RECONCILIATION
Grader A vs Grader B splits: [bullet, test, A verdict, B verdict, agreed verdict, reason]
```

*One block per bullet, one role block per role; each grader fills it in alone before comparing.*

## How bullet grading goes wrong

Most grading errors are false passes: lines that look like they pass but don't. Each test has a typical false pass and a check that catches it. Running the check is what makes two graders agree.

### Title-inferable: fancy words, ordinary duty

- **False pass:** "Orchestrated stakeholder alignment." It sounds specific, but it's a duty in a costume.
- **Check:** rewrite it in plain verbs. If the plain version is something everyone in the role does, it fails.

### Owned action: team credit behind a strong verb

- **False pass:** "Launched" when the team launched and you attended the planning meetings.
- **Check:** ask "what did you personally do?" If the answer is "attended," "reviewed," or "supported," it fails. Rewrite around what you actually did.

### Substitute evidence: scope mistaken for a result

- **False pass:** "Managed 30 accounts." That's real scope, but nothing changed.
- **Check:** ask what was different at the end. If nothing was, it becomes a context line at most and doesn't count toward the role band.

### Baseline: a big number on a tiny base

- **False pass:** "Increased conversion 40%" from a base of 5 users. Jobscan lists missing context behind impressive numbers as an AI tell, so this line hurts you twice.
- **Check:** demand the "from" value or the denominator. If it's embarrassing, the claim was weaker than it looked. Restate it honestly or find a different bullet.

### Sourced number: an estimate presented as fact

- **False pass:** a figure narrowed down by guesswork and written as if it were measured. One career coach argues "Nobody's going to check these data points." Yet 49% of hiring managers in one survey say candidates frequently can't back up their claims, so the interview is exactly where the checking happens.
- **Check:** name the artifact the number came from. A reconstruction from an accurate frequency passes. A number with no artifact and no method fails.

### Specific noun: a tool dropped in for show

- **False pass:** "...using Excel." It's a noun, but it says nothing about the result.
- **Check:** delete the noun and reread. If the reader believes the same thing about the result, the noun wasn't doing work. Replace it with the system, artifact, or audience that matters.

### Count: the band met with padding

- **False pass:** six bullets under the current role, three of them duties.
- **Check:** count only bullets that passed all seven tests. If the passing count falls short of the band, the fix is better bullets or a shorter block, never filler.

### Rubric drift: graders quietly loosen the tests

- **False pass:** a grader brings "when applicable" or "concrete enough" thinking back in and passes borderline lines. ULM's and WPI's rubrics are built this way, which is why graders using them diverge.
- **Check:** two graders score independently, then reconcile every split by pointing to the check that decides it. If a split can't be settled by a check, write the ruling down so the next resume gets the same answer.

## Before the resume ships, and keeping it current

Ship when every bullet on the page passes all seven tests and every role block passes the lead and count rules. Re-grade whenever a bullet changes, which happens every time you tailor.

#### Ship check for experience bullets

- [ ] Every bullet has a row on the grading sheet with a role ID
- [ ] No bullet could be written by a stranger from the job title in the target market's wording
- [ ] No bullet opens with Responsible for, Helped, Worked on, or Duties include
- [ ] Every bullet has a named evidence type, and no scope-only line is counted as a result
- [ ] Every number on the page has a from value, denominator, or period
- [ ] Every bullet names a system, artifact, audience, or count that changes what the reader believes
- [ ] No bullet runs past two lines
- [ ] Every number has a named source artifact you could walk an interviewer through
- [ ] Each role leads with its strongest relevant passing bullet
- [ ] Each role's passing-bullet count falls inside its band, with at most one context line
- [ ] A second grader scored the resume independently and every split was reconciled

### When to re-grade

- **Every tailored version.** The lead rule depends on the posting, so the right lead bullet changes from one application to the next. Refolk tailors the resume to each posting and scores how well you actually fit, which turns re-choosing the lead into a check rather than a rewrite. Re-run the lead rule and test 1 on every version anyway.
- **When you switch target markets.** A move between US and UK postings changes the title wording a stranger would infer from, so test 1 has to be re-run.
- **When a number's source goes stale.** If you can no longer get to the artifact behind a figure, test 7 fails until you can.
- **When the research moves.** The review-time studies above differ in method and sample, and none measures bullet-level attention. If a study publishes read time by bullet position or length, revisit the lead and length rules. Until then, treat them as convention backed by verification logic.

## Frequently asked questions

### How do I tell if a resume bullet is a duty or an accomplishment?

Cover everything except the job title and ask whether a stranger could have written the bullet from the title alone. If they could, it is a duty, however polished the wording. An accomplishment names something only you did: an owned verb, a specific system or audience, and evidence that something changed. Rewrite fancy phrasing in plain verbs first; if the plain version reads like a job description, it fails.

### How many bullet points per job should a resume have?

Ranking pages converge on 4-6 bullets for the most recent role, 3-4 for the previous one, and 1-3 for older roles. Indiana State's career center suggests 2-4 per job. These bands are convention, not measured reader behavior. The rule that matters more: count only bullets that pass every test. A role with three strong bullets beats one with six where three are padding.

### Can I use estimated numbers on my resume?

Yes, if you can say exactly how you got the estimate. Rebuilding a count from an accurate frequency, such as tickets per week across a known period, is reconstruction, not invention. A figure you cannot walk an interviewer through should not appear. In one survey, 49% of hiring managers said candidates frequently cannot back up claims on AI-assisted resumes, so expect the question.

### How do I make resume bullets sound less AI-written?

Add what a model could not know: the named system, the artifact you built, the specific audience, the real count with its baseline. Hiring managers in Resume Genius data flagged unnatural phrasing, generic language, vague or inflated claims, and buzzwords as giveaways. Jobscan adds identical sentence structure across bullets. Varying structure helps, but specificity is what actually changes the verdict.

### What if my job genuinely has no metrics?

Use substitute evidence: scope paired with a change, frequency, audience, before-and-after without a percentage, adoption, speed, or quality. A bullet stating who used what you built, or what process ran differently afterward, passes the evidence test. Scope on its own, like the number of accounts you managed, is context rather than a result, so it should not lead a role.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/candidates/guides/experience-bullet-standard-line-grading*