A fake résumé that lists "Spread Herpes STD to 60 percent of intern team" as a bullet point got a 90 percent callback rate. It’s not a hypothetical — someone built it on Reddit as a joke, gave it a recognizable university, a couple of well-known company names, and job titles that read like real software engineering, and then watched it sail through screens that were supposedly evaluating substance. Aline Lerner, who ran a recruiting firm before founding interviewing.io, uses it to make one point: recruiters are skimming for brand names on a page they never really read.
That’s a funny story with an uncomfortable implication. If a fake accomplishment nobody could have missed still got a callback, the résumé wasn’t being read closely enough to catch it — which means a lot of the advice about résumé wording is solving a problem nobody downstream is checking for.
The parser is not the villain people think it is
Ask anyone who’s job-hunted since roughly 2015 what happens to a résumé after they hit submit, and you’ll get some version of the same story: a robot reads it, greps for keywords, and silently discards anything that doesn’t match closely enough, no human ever seeing the difference. It’s a tidy explanation for a system that gives no feedback. It’s also not what Greenhouse — the applicant tracking system behind a large share of tech job postings — says its own product does.
Greenhouse’s technical documentation on its AI matching feature, Talent Matching, is unambiguous: the system "does not automatically advance or reject candidates," and "the recruiter must make the decision to advance, reject, or add notes about a candidate" on every single application. When the feature can’t parse a résumé at all — a scanned image, a layout its extractor chokes on — the candidate doesn’t vanish into a silent pile. They get surfaced to a recruiter for manual review instead.
Greenhouse SupportTalent Matching – Data Processing FAQThe vendor documentation itself: no automated advance-or-reject decision, and an unparseable résumé routes to a human rather than a silent rejection.This is one vendor’s documented behavior, not a law of physics. Greenhouse doesn’t run every job board in existence, and plenty of smaller or older systems really do have a harder keyword gate somewhere in the pipeline that a résumé can fail outright. But the popular image — a merciless bot summarily trashing qualified people over a missing synonym — describes something the market leader’s own documentation doesn’t actually support.
What the parser rewards is boring. Greenhouse’s own advice to candidates is to upload a PDF "for best results," and to re-save and shrink the file if an upload fails — no mention of fonts, keyword density, or any of the tricks resume-writing services charge for. A résumé built as an image, or with text boxes and columns a parser can’t reliably read top to bottom, is the actual failure mode, and it has nothing to do with which words you chose. Greenhouse spells this out in its candidate-facing FAQ too, alongside a small, honest admission: parsing "is designed to save time, but can make mistakes," so check what it pulled before you submit anything.
The human reading it isn’t much more reliable
Once a résumé clears whatever mechanical gate exists, a person looks at it, briefly. Ladders’ eye-tracking research — first run in 2012, updated in 2018 — puts the average initial screening time at 7.4 seconds. interviewing.io’s 2024 study, measured differently, found a median of 31 seconds per résumé across a full evaluation. Different methodology, same order of magnitude: almost nobody is reading closely, and "the six-second scan" undersells it about as often as it oversells it.
What that person does in those seconds is the part worth understanding, and Lerner has now measured it twice, a decade apart, with the same result both times.
The first version, in the mid-2010s, asked recruiters, engineers, and hiring managers to match six résumés to the right candidates. Average accuracy across everyone: 53 percent. Agency recruiters did best, at 56 percent. Eng hiring managers did worst, at 48 percent — worse than chance. Statistically, agreement between raters (Fleiss’ kappa) came out to 0.13, functionally indistinguishable from coin flips agreeing with each other.
Ten years later, with an actual research lab attached and nearly 2,200 evaluations from 76 technical recruiters across more than 1,000 real resumes, she ran it again. Recruiters correctly predicted whether a candidate would pass a technical interview 55 percent of the time. Two recruiters shown the identical résumé disagreed on pass probability by an average of 41 percentage points — disagreement between two people looking at the same page that turned out to be statistically indistinguishable from the disagreement between two people looking at two different pages entirely. And the calibration ran backwards in a way that should bother anyone whose résumé just got a form rejection: candidates recruiters rated a 0–5 percent chance of passing actually passed the technical interview 47 percent of the time. Candidates rated 95–100 percent passed at 64 percent — barely better.
Aline Lerner and Peter Bergman, interviewing.ioAre recruiters better than a coin flip at judging resumes?The 2024 replication, run with Learning Collider. The calibration numbers matter more than the headline accuracy figure.The median time a recruiter in that study spent per résumé was 31 seconds — 25 seconds for anyone advanced, 44 for anyone rejected, which on its own says something about how much of a screen a rejection actually got. Recruiters who spent more than 45 seconds were measurably more accurate: roughly 15 extra seconds on a page produced about a 34 percent jump in prediction accuracy. Slowing down works. Almost nobody in a real hiring pipeline is incentivized to do it.
What actually gets you rejected
A recruiter asked why they passed on a candidate will almost always answer "missing skill." It was the runaway top reason recruiters gave for their own rejections in the 2024 study. Measured against what those same recruiters actually did, rather than what they said, it isn’t close to the real reason.
The strongest predictor of an actual rejection wasn’t a skills gap. It was the absence of a recognizable company name. The next predictor — smaller effect, wide error bars, but present — was having an MBA. Third and fourth were no top-tier school and holding a master’s degree at all. The top four rejection reasons the data actually found were all about a candidate’s background, which is close to the opposite of what recruiters told researchers they were screening for.
An MBA hurting your odds is the kind of finding nobody would guess going in, and the sample behind it is thin enough — eleven observations — that it shouldn’t be treated as settled. It’s in the data anyway.
Lerner opens the piece that includes the fake-résumé anecdote by asking ChatGPT to size the global résumé-writing industry, mostly as a bit — the model answers with a suspiciously precise $1.37 billion for 2024, growing to $1.59 billion by 2033, complete with a tidy explanation about AI-driven demand. I have no way to check that number, and neither, as far as I can tell, did she; it reads like a figure a language model invents on request, without anything backing it in a filing or a market report anywhere. I’m including it anyway, because it’s a good demonstration of exactly the failure this whole piece is about — a confident, specific-sounding number nobody checked, sitting two paragraphs above a statistic that was checked and holds up. The résumé-writing industry might really be worth $1.37 billion. It might not be. Either way, it’s not the number in this piece worth remembering.
Aline Lerner, interviewing.ioWhy resume writing is snake oilThe herpes-résumé example and the case against paying for a rewrite, from someone who used to sell recruiting services for a living.There’s a specific case in the 2024 study worth sitting with if your own background doesn’t match the pattern above. One candidate — résumé shared with permission, name changed to "John" — studied chemical engineering, worked his way into software through a job doing penetration testing, completed a coding bootcamp, then attended Bradfield School of Computer Science, and reached a senior title within three years. Recruiters in the study rated him consistently poorly. On interviewing.io’s own mock interviews, he’s a top performer, outperforming candidates with FAANG résumés on the same platform.
The first pass doesn’t see distance traveled.
It sees a résumé that doesn’t pattern-match the template it’s used to.
What’s actually worth fixing
- Put your two or three best facts in the summary line, in plain English, and skip filler like "passionate self-starter" or "detail-oriented team player" — interviewing.io calls out exactly that phrasing, and it tells a skimming reader nothing they can act on.
- Drop your GPA if it’s under 3.8. A middling number reads as academic mediocrity to the one recruiter glancing at it for half a second.
- Save the file as a PDF. It is, unglamorously, the single format change most likely to matter, and it costs nothing.
- Don’t pay someone to reword your bullet points for the fifth time. The data on how those bullets get read doesn’t support the idea that better phrasing moves the needle much, and the money is better spent elsewhere.
None of this is a case for perfecting one résumé. If two recruiters reading the identical page land 41 percentage points apart, and the recruiters most confident you’ll fail are wrong more than half the time, the sane response to that much noise is more attempts, not one flawless draft — which turns the actual problem into remembering which version went to which of a dozen companies, and which one owes you an answer by when. That bookkeeping is closer to what Vigil does than anything about wording.
Send the version that’s already good enough. Save the polishing time for the ones that write back.