The 2018 Ladders eye-tracking study measured 30 recruiters reading resumes with calibrated eye-tracking equipment. The number that everyone cites is 7.4 seconds. The number nobody cites is the gaze pattern. Recruiters fixed their eyes on six places on the page, in a near-identical order: name, current title and employer, dates of current role, previous title and employer, dates of previous role, education. Everything else was peripheral. If you are a founder reviewing resumes, your brain is doing the same thing whether you mean to or not.
What recruiters read in the first 7 seconds
Recruiters do not read resumes. They scan them, and the scan is structured by visual cognition more than by content strategy. The Ladders study found that experienced recruiters spend an average of 7.4 seconds on the initial pass, fixating on six anchor points on the page. The signals they extract from that scan are not keywords. They are: relevance of the most recent role to the open job, tenure pattern (did this person stay long enough to deliver?), trajectory (does each role represent more responsibility than the last?), and credential cleanliness (does the resume look like the person took it seriously?). Keywords are the fuel that gets a resume past automated screening, but the actual hiring decision happens at the layer above keywords. This is the gap most ATS platforms fail to close. They optimize for keyword recall and surface 200 “matches.” The recruiter still has to do the 7-second scan, 200 times. The point of intelligent screening is to move that signal extraction upstream so the founder reads the top 8 candidates on their actual merits, not their token density.
What the Ladders gaze map actually shows
Six fixation points: name (top-left), current title-employer-date block, previous title-employer-date block, education, and the bottom-right corner where bullet points sometimes contain a closing accomplishment. The middle of the page got almost no fixations. Long bullet lists were almost entirely skipped. This explains why a founder who reads 30 resumes back-to-back ends up with strong opinions despite reading less than half of each one: the brain has compressed the resume to its skeleton.
Why keyword matching is the wrong layer to optimize
Most legacy ATS platforms run keyword matching as the first filter. The job description has “Python, AWS, distributed systems.” The system scans every resume for those tokens and ranks by frequency. This catches people who do have the skills, and it also catches people who learned to game the filter by stuffing the document. It misses people who described the same skills in different words. A resume that says “reduced inference latency by 40% on the production ranking service” is materially stronger than one that lists “Python, AWS, ML” as a skill bullet, but a keyword filter scores them in the wrong order. The Jobscan-popularized statistic that 75% of resumes are rejected by ATS is widely repeated, but as ERE’s industry analysis notes, the methodology behind that number is thin. The real failure mode is more specific: keyword filters reject context-rich resumes that lack the exact tokens, and surface context-poor resumes that have them. The fix is not to remove the filter. It is to evaluate beneath it.
Signals that actually predict performance
Four matter, in order: quantified achievements (numbers, percentages, dollar amounts attached to outcomes), experience relevance (did this person solve problems near the one we are hiring for), career trajectory (is responsibility growing role-over-role), and skills alignment (do the named skills match the work). Each of these can be extracted from a resume by a model that reads the document the way a human reads it, not the way a keyword scanner does. This is what CurriculoATS Impact Scoring measures.
What we learned from Amazon search that applies here
Before CurriculoATS, our founder Dev worked on Amazon’s search and recommendation systems. The lesson that translated most directly to hiring: recall is cheap, ranking is the hard part. Amazon’s search index can find a million products that match a query. The product team’s job is making sure the top three are correct. Resume screening has the same shape. Pulling 300 resumes that contain “product manager” is a five-line script. Returning the eight that a senior PM would actually want to phone-screen is the real engineering problem. Multi-signal scoring is how Amazon ranks products, and the same approach, with different signals, ranks candidates. We weight quantified achievements heavily because they are the strongest predictor that someone has actually shipped work, not just held a title. We weight career trajectory because a person whose responsibility doubled twice in five years is statistically a stronger hire than a person whose title stayed the same. We expose the reasoning so the founder can audit it.
Why “signal” matters more than “keyword”
A keyword is a token. A signal is a pattern that correlates with outcomes. “Python” is a keyword. “Owned the ranking service from prototype to 50M daily queries” is a signal. Both can appear in the same resume. Only one tells you whether to interview the person. A founder reviewing the top 8 candidates from a signal-based scoring system spends about 30 seconds per candidate and arrives at a decision they can defend. A founder reviewing the top 8 from a keyword-rank system spends 5 minutes per candidate and still has to do the work the screener should have done.
How a founder can read a resume in 30 seconds
The same gaze pattern the Ladders study measured can be turned into a deliberate framework. Three sweeps, in order, each one a few seconds.
- The trajectory sweep. Read only titles and dates. Are the roles getting more senior? Are the tenures long enough that the person delivered? Is the most recent role aligned with what you are hiring for?
- The achievement sweep. Read only the bullet points that contain numbers. Did this person produce measurable outcomes? Are the outcomes plausibly attributable to them?
- The disqualifier sweep. Look for the things that would rule the person out: gaps you cannot explain, role-hopping under 9 months, claimed skills that the resume does not back up.
Three sweeps, 30 seconds. If the candidate clears all three, they advance to a phone screen. If they fail any one, you can move on. This is exactly the framework CurriculoATS automates for the first pass, so you only do it manually for the top 10 candidates.
The single best resume signal
If we had to keep one signal and throw out all the rest, we would keep quantified achievements. A resume with five bullets that include numbers is materially more likely to belong to someone who delivers than a resume with fifteen bullets of pure responsibilities. The reason is selection: people who track outcomes write resumes that track outcomes. People who do not track outcomes do not.
Frequently asked questions
Do recruiters really only look at a resume for 7 seconds?
The 7.4-second figure is from the 2018 Ladders eye-tracking study and refers to the initial scan, not the full review. Resumes that pass the initial scan get a more careful read, often 90 to 180 seconds. The 7-second window is what determines whether your resume gets a second look at all, which is why the structural anchors (current title, dates, trajectory) carry so much weight.
What’s the difference between keyword matching and AI resume screening?
Keyword matching counts tokens. AI resume screening reads the resume for context, extracting signals like quantified outcomes, role relevance, and career progression. The output of keyword matching is a frequency count. The output of AI screening, in CurriculoATS specifically, is a 0-100 composite score and a written reasoning paragraph that a founder can audit. We compare the two approaches in detail on the AI resume screening page.
What resume signals should a startup founder care about most?
Quantified achievements first, role relevance second, trajectory third. Skills should be a filter, not a ranking signal. The reason is that named skills are easy to claim and hard to verify; outcomes are harder to claim and easier to verify by interview. If you only have time to look for one signal, look for numbers attached to outcomes.
How does CurriculoATS handle non-traditional career paths?
The model evaluates achievement and impact independently of credential pedigree. A self-taught engineer who shipped a product used by 100K people scores high on impact even without a CS degree. A founder who pivoted from a non-technical role into product management scores on trajectory and outcome rather than on the linearity of the path. The reasoning paragraph makes the score explainable, so a hiring manager can see exactly why the model surfaced the candidate.
Are resume formatting tricks still effective?
Less than they used to be. Modern parsers handle reasonable formatting well. Multi-column layouts and image-based names still cause problems. The single most common parsing failure we see is contact information embedded in headers or text boxes, which older parsers skip. Plain, single-column resumes parse cleanly across all major systems including ours.
Take the next step
If you are a founder reviewing resumes by hand, the highest-leverage move is not learning to read faster. It is moving the first 90% of the screen out of your inbox and into a system that does the gaze-pattern work for you, then surfaces the top 8 with reasoning. That is what we built. Start with the free Starter plan for one active job, or read the Impact Scoring breakdown to see exactly how each signal is weighted. The 7-second scan is real. Make it work for the candidates who deserve a second look.