A recruiter's first look at your resume lasts about seven seconds, which is enough time to read roughly 30 words of a page that holds well over 300. This post works out where that number comes from and which 30 words they usually are on a software engineer's resume. By the end you will know which few lines to rewrite first, and why the second word of your bullets matters more than it should.
The estimate starts with one small calculation. Ladders reported an average first screen of 7.4 seconds in its 2018 eye-tracking study of recruiters. A 2019 meta-analysis of 190 studies puts adult silent reading speed at 238 words per minute. Multiply them and you get 29 words. Nobody designed those numbers to be multiplied together, and both have caveats we get to below, but a second method based on eye movements lands in the same place.
For scale, the sample content in Jake's resume, a free LaTeX template popular with software engineers, has 366 words. Thirty words is about 8% of that page, and most of it goes to your job titles, company names and dates before a single bullet gets read.
Where the 7.4 seconds comes from
Most resume advice repeats "6 seconds" or "7.4 seconds" without saying where the numbers came from, so here is the short version.
In 2012, the job site TheLadders put eye trackers on 30 professional recruiters over a 10-week period and recorded how they reviewed resumes. Recruiters had previously estimated that they spent 4 to 5 minutes per resume. The eye tracker said about 6 seconds for the initial fit or no-fit decision. Almost 80% of that time went to six data points: the candidate's name, current title and company, current start and end dates, previous title and company, previous start and end dates, and education. Beyond those, the report says recruiters did little more than scan for keywords.
Ladders updated the study in 2018 and reported an average initial screen of 7.4 seconds. The press release does not give a sample size. It does say which resumes did well: simple layouts, clear section and title headings, bold job titles with bulleted accomplishments. The weakest had cluttered layouts, multiple columns, long sentences and little white space.
Treat these numbers as rough. Ladders sells resume rewriting, and the 2012 report closes with a pitch for that service. Neither study was peer reviewed. Both measure triage, the quick decision about whether to keep reading, not how long a promising resume eventually gets.
A third study gives a useful upper bound. In April 2018, the job board StepStone ran eye tracking with 33 HR professionals at a conference in Austria. They looked at four sample CVs with no queue of applicants behind them. Their mean viewing time was 43 seconds, while their own median estimate was 2 minutes. So the gap between how long recruiters think they read and how long they actually look shows up in both companies' data, even though the absolute numbers differ a lot.
Two ways to count to 30
Reading speed is the obvious route. Marc Brysbaert's 2019 meta-analysis puts adult silent reading of English non-fiction at 238 words per minute, with most adults between 175 and 300. Over 7.4 seconds that is 22 to 37 words, and 29 at the average.
The objection is that nobody reads a resume like an essay. They skim. Carver (1992) proposed a skimming speed of 450 words per minute and a scanning speed of 650. Brysbaert's review found little evidence for separate reading "gears" apart from the difference between reading and scanning for a specific word. Even at Carver's scanning speed, 7.4 seconds covers about 80 words.
The second route ignores words and counts eye movements. Your eyes do not glide across text. They stop (a fixation), jump (a saccade), and stop again. In Keith Rayner's 2009 review of eye-movement research, the average fixation lasts 225 to 250 ms in silent reading and 180 to 275 ms in visual search tasks. Divide 7.4 seconds by those durations and you get about 30 to 33 stops for reading and 27 to 41 for search. That is a slight overestimate, because the jumps between stops take time too.
Both methods use the 7.4-second figure from Ladders (2018). Reading rates: Brysbaert (2019). Fixation durations: Rayner (2009), Table 1. Dots mark the mean or midpoint.
Each stop also has a limited reach. Rayner's review describes the useful window during a fixation as roughly 3 to 4 characters to the left and 14 to 15 characters to the right. That is about 18 characters per stop, or two or three short words.
Two different methods, both landing near 30. We would not argue for 30 over 35. We would argue that the right unit is tens of words, not hundreds.
Even the generous number leaves most of the page unread
Here are the three measured times, converted to words at the average reading rate and set against the 366 words in Jake's sample.
Words = seconds × 238 ÷ 60. Times: Ladders eye-tracking studies (2012, 2018) and StepStone eye-tracking study (2018, mean viewing time). Page length: visible words in the sample content of Jake's resume template.
Even 43 unhurried seconds covers less than half the page. Of course nobody reads the first 171 words and stops. Brysbaert suggests that skimming probably alternates between skipping large parts of a text and short bursts of normal reading, so you find information if it sits in a part that gets read and miss it if it does not. That fits what the eye-tracking heatmaps show.
So the useful question is which parts get read.
Most of the 30 words are already spoken for
The 2012 Ladders study already answered most of it. Almost 80% of the time went to name, two titles and companies, two sets of dates, and education. We counted those six data points on Jake's sample: they come to 31 words.
Time share: Ladders (2012), "almost 80%" of initial review time. Word share: our count on Jake's template sample (name; the first two experience entries' title, organization and dates; the first education entry's school and degree).
That is the whole budget, more or less. The six fields take about 8% of the words and get about 80% of the attention. Your job title, your company name and your dates are probably the most-read words you will ever put on a resume, and they are usually the words people spend the least time editing.
Now do the remaining arithmetic. If 20% of a 6-second look goes to everything else, that is 1.2 seconds, or about five eye stops at 225 to 250 ms each. Five stops for 335 words of bullets, skills and projects.
Where the last five stops land
Nielsen Norman Group's eye-tracking work on web pages found what it calls the F-shaped pattern: when text has no strong visual cues and the reader is in a hurry, people read across the top, read a shorter line further down, and then run their eyes down the left edge. The first words of each line get the most fixations. In a separate 2009 study of link text, Jakob Nielsen found that users typically see about 2 words of each list item.
A resume is a stack of list items with a ragged right edge, so the leftover stops tend to land on the start of a line: the first words of a bullet, the start of the skills line, a project name.
One stop at the start of a bullet picks up roughly 15 characters to its right. On the most common bullet opening in Jake's sample, those 15 characters are "Developed a new". Five of the template's 20 sample bullets start with "Developed a".
Before blaming the template, we checked our own work. We publish 18 example resumes for different stacks (Python, Go, Kubernetes and others), with 108 bullets between them. In 74 of those 108 bullets, 69%, the second word is "a", "an" or "the". Five two-word openings account for more than half of all bullets.
First two words of every experience and project bullet in the 18 example resumes published on sweresume.app (October 2026). 55 of 108 bullets start with one of these five openings.
Starting with a verb is good advice, and we are not arguing against it. Laszlo Bock's XYZ formula starts with an active verb too. The waste is in the word after the verb. "Developed a" tells a skimming reader that you did something. "Migrated billing" tells them what.
The same bullets, written for the first stop
Here is one made-up experience entry before and after. The bold text marks the first 15 characters of each bullet, about what a single eye stop at the start of the line takes in. Read only the bold parts and the header line, which is roughly what the first pass gets.
Before
Software Engineer II · Mar 2022 to Present
Acme Corp
- Developed a new service to retry failed card payments using Go and Kafka, reducing failed charges by 18%.
- Worked on the migration of the billing database from MySQL to PostgreSQL with zero downtime.
- Responsible for improving on-call alerting, which cut pages by 40%.
After
Software Engineer II · Mar 2022 to Present
Acme Corp, Payments team
- Cut failed card charges 18% with a Go and Kafka service that retries declined payments.
- Migrated billing from MySQL to PostgreSQL (2.1 TB, 140 tables) with zero downtime.
- Reduced on-call pages 40% by rewriting 60 alert rules around SLOs.
An illustration of the model, not eye-tracking data. The 15-character window follows Rayner (2009): useful information extends about 14 to 15 characters to the right of a fixation. The employer, team and numbers are invented.
From the bold parts of the "before" version, a skimming reader learns that you developed something, worked on something, and were responsible for something. From the "after" version, they learn that you cut failed card charges, migrated billing, and reduced on-call pages. Every bullet still starts with a verb. None of them got longer. The official title is unchanged; the team name went on the company line, where it costs nothing and gives the reader context for everything below it.
For more rewrites of this kind, our guide to software engineer resume bullet points has examples across levels and stacks.
What to change, starting with the most-read words
The title, company and date lines. Keep your official title, since employment verification can check it. Use the company line for what the title hides: "Acme Corp, Payments team" or "Acme Corp, internal developer platform". Put dates in the same place and the same format on every entry so the reader finds them without searching.
The order of roles. "Current" and "previous" are two of the six data points. Reverse chronological order puts them where the eye already goes. If your most relevant job is your second most recent, its title line is still read, so make that line carry the relevance.
The first bullet in each role. The top line of each block gets a horizontal pass that lower lines may not get. Lead with the bullet you would keep if you could keep only one.
The first two words of every bullet. Verb, then the thing: "Migrated billing", "Cut p99 latency", "Replaced Jenkins". Drop "a" and "the" from the opening wherever the sentence still reads naturally. If the number is the point, move it into the first 15 characters.
The start of the skills line. Only the first item or two in each category is likely to be seen on a skim. Order them by the job description, not alphabetically and not by what you learned first.
One column. Ladders found multi-column layouts among the weakest performers, and a left-edge scan works best when there is one left edge. Machines have a version of the same problem: in our resume format benchmark, both PDF text extractors we tested scrambled the section order of a two-column layout.
None of this needs a rewrite of the whole resume. On a typical page it is the header lines, a few dozen bullet openings and one skills line. If you use SWE Resume to tailor bullets to a job description, check the first two words of what it writes the same way you would check your own.
What this estimate does not tell you
The 30-word figure describes triage, not the decision. A peer-reviewed 2023 study by Pina and colleagues recorded recruiters' eye movements while they screened resumes and trained a model to predict which resumes would move forward. Total viewing time and time spent on the Experience and Education sections were among the useful predictors. Resumes that survive the first pass get read more closely, and that is where the rest of your bullets do their work.
The reading rates come from studies of continuous prose, and a resume is not prose. The fixation route avoids that problem but assumes that resume skimming looks like reading or visual search in a lab, which nobody has measured directly.
The recruiter studies are small, run by job sites, not peer reviewed, and now 8 to 14 years old. The 2018 Ladders sample size is not public. Screening habits also differ by company, role and applicant volume.
The estimate also says nothing about applicant tracking systems or recruiter search, which decide whether a person sees your resume at all. That is a separate problem with separate fixes.
A seven-second test you can run tonight
Print your resume. With a marker, cover everything except your name, the title, company and date lines, your education line, and the first 15 characters of each bullet and of the skills line. Give it to a friend, preferably one who is not an engineer, for seven seconds. Then take it back and ask them two questions: what kind of engineer is this, and what have they done?
If the answers are "a software engineer" and "developed things", you know which words to change.
How we got these numbers
Screen times. 6 seconds and "almost 80%" on six data points: TheLadders, "Keeping an eye on recruiter behavior" (2012; 30 recruiters, 10 weeks). 7.4 seconds: Ladders, press release, November 6, 2018, also covered by HR Dive. 43 seconds measured and 2 minutes estimated: StepStone Austria, Eyetracking-Studie 2018 (33 HR professionals, in German).
Reading rates. 238 words per minute, typical range 175 to 300, for silent reading of English non-fiction: Brysbaert, M. (2019), "How many words do we read per minute? A review and meta-analysis of reading rate", Journal of Memory and Language 109, 104047. Carver's skimming (450) and scanning (650) rates are as cited in that review.
Eye movements. Fixation durations of 225 to 250 ms (silent reading) and 180 to 275 ms (visual search), and the perceptual span of 3 to 4 characters left and 14 to 15 right: Rayner, K. (2009), "Eye movements and attention in reading, scene perception, and visual search", Quarterly Journal of Experimental Psychology 62(8), 1457 to 1506.
Scanning patterns. Pernice, K. (2017), "F-Shaped Pattern of Reading on the Web: Misunderstood, But Still Relevant", Nielsen Norman Group. Nielsen, J. (2009), "First 2 Words: A Signal for the Scanning Eye". Screening predictors: Pina et al. (2023), "Using Machine Learning with Eye-Tracking Data to Predict if a Recruiter Will Approve a Resume", Machine Learning and Knowledge Extraction 5(3).
Calculations. Words covered = seconds × words per minute ÷ 60, so 7.4 × 238 ÷ 60 = 29.4, 6 × 238 ÷ 60 = 23.8, and 43 × 238 ÷ 60 = 170.6. Eye stops = seconds ÷ fixation duration, so 7.4 ÷ 0.250 = 29.6 and 7.4 ÷ 0.225 = 32.9; 7.4 ÷ 0.275 = 26.9 and 7.4 ÷ 0.180 = 41.1. These ignore the time spent moving between stops, so they run slightly high. The "five stops" figure is 20% of 6 seconds (1.2 s) divided by 0.225 to 0.250 s, which gives 4.8 to 5.3.
Word counts. We took the sample content of Jake's resume template (MIT license), dropped LaTeX commands, comments and link targets, and counted whitespace-separated tokens of visible text, including headings and contact details: 366. The six data points (name; title, organization and dates of the first two experience entries; school and degree of the first education entry) come to 31 words. Date ranges count as three words ("June 2020 Present").
Bullet openings. We took every experience and project bullet from the 18 example resumes published under /examples/ on sweresume.app (108 bullets), stripped the highlight markup, and counted the first two whitespace-separated words, treating "an" as "a". In Jake's template we counted the 20 uncommented sample bullets. Our own examples were written as models of strong bullets, which is why we thought the result was worth showing.
Originally published on sweresume.app.
