Nobody Is Auto-Rejecting Your Resume

Job search advice has a sourcing problem. Almost every article cites another article, which cites a third, and somewhere near the bottom of the stack there is usually nothing at all. So I did the boring thing: I collected twenty sources — the primary research papers, the recruiter surveys, the practitioner playbooks, and the people who think the whole premise is wrong — and worked out which claims survive contact with their own citations.

Several didn't. Including the one you've heard most.

Audio deep dive · 42 minutes

Why the resume supercomputer is a ghost

The statistic that turned out to be a sales pitch

You have read this sentence somewhere: 75% of resumes are rejected by an applicant tracking system before a human ever sees them. It appears in Forbes, on university career pages, in the marketing copy of every resume-optimisation tool, and in the answers chatbots give when you ask them how hiring works.

It comes from a 2012 sales pitch by Preptel, a small job-search software startup. There was no study attached to it. No survey, no methodology, no data. Preptel shut down in 2013, and the number outlived the company by way of a citation chain that anyone can walk: Forbes picked it up in 2014 without checking, CIO.com cited Forbes in 2018 as though Forbes were the source, CNBC cited CIO.com in 2019, and from about 2020 the rest of the internet cited CNBC. The researcher Christine Assaf went looking for the underlying study in Google Scholar and found nothing — no academic research of any kind supporting the figure.

What recruiters actually do with the software is rank and sort, not silently discard. A 2025 study by Enhancv found that 92% of recruiters do not configure content-based auto-rejection rules at all. That study is small — 25 recruiters across more than ten platforms — and I'd treat the precise number as indicative rather than settled. But it points the same direction as everything else: what genuinely gates an application before a human sees it are knockout questions, the hard eligibility requirements an employer sets by hand. Work authorisation. A required licence. A minimum number of years. Those are decisions a person made, not an algorithm's opinion of your bullet points.

This is not good news, and it took me a while to notice why.

The bottleneck is human attention, and there is much less of it

If a machine were rejecting you, the fix would be mechanical: change the format, add the keywords, clear the filter. The evidence says something worse. Your resume reaches a person, and that person cannot give it any real time.

The average corporate posting draws around 250 applications, and four to six people get interviewed. A 2018 eye-tracking study by TheLadders put the median initial review at 7.4 seconds. That is not laziness; it's triage by a recruiter carrying twenty to forty open roles at once. And the filtering that does happen upstream is genuinely costly: Harvard Business School and Accenture surveyed 2,250 employers across the US, UK and Germany in 2021 and found 88% of them acknowledged that qualified, high-skilled candidates get screened out for not matching automated search criteria exactly. They estimated 27 million Americans are "hidden workers" on that basis.

So the case for going directly to a hiring manager doesn't rest on beating a robot. It rests on the fact that seven seconds of a stranger's attention, contested by 249 other people, is a bad channel — and that a short, specific, well-aimed message is a better one.

What the strongest evidence actually says about networking

The one finding in all of this with real experimental weight behind it is Rajkumar et al., A causal test of the strength of weak ties, published in Science in 2022. It is worth being precise about, because the popular summary of it is wrong in an important way.

The researchers used five years of randomised experiments on LinkedIn's "People You May Know" algorithm, covering more than 20 million people. Across two waves in 2015 and 2019 the experiments produced roughly two billion new connections, 70 million job applications and 600,000 new jobs. Because assignment to the weak-tie or strong-tie variant was random, they could use it as an instrument and estimate a genuine causal effect rather than the correlation earlier studies were stuck with — which matters, because the correlational answer had been the opposite one.

The result: weaker ties do cause more job mobility, but not without limit. The relationship is an inverted U. Adding connections with a few mutual friends in common raises your probability of a job transmission; adding connections with more than about ten mutual friends in common lowers it again. As one of the authors, Iavor Bojinov, put it: it is not a matter of the weaker the better. The peak sits at moderately weak ties — people somewhere between total strangers and your actual friends.

The paper measures tie strength two different ways, and they do not behave the same. Both matter, and so does your industry:

Rajkumar et al., Science (2022). The middle column is the paper’s result; the right column is my reading of it.

The practical translation is unglamorous. The person most likely to move your career is not your close friend and not a cold stranger. It's the acquaintance — the ex-colleague two jobs back, the person you shared a project with once, the one you'd both recognise but haven't spoken to in two years.

The mechanics, briefly

Find the likely manager rather than the recruiter. Verify the email rather than guessing at the pattern. Then write something short enough that reading it is not a favour.

The best-known template here is Liz Ryan's Pain Letter, from her company Human Workplace, and its value is the reframe more than the format: you write about the manager's problem instead of your own history. Four parts — a hook tied to something publicly true and recent about their team, a researched guess at the operational pain that follows from it, one short story about solving something similar, and a low-pressure close.

Everything else worth borrowing is about lowering the cost of replying. Austin Belcak's version ends with an either/or question and the line "if you can, feel free to just respond A or B." Kyle Asay's ends with "what's the best way to ensure my application is reviewed by a member of your team?" — a question that takes ten seconds to answer, rather than a request for a meeting that shifts all the work onto the recipient. The LinkedIn outreach data lines up: messages under 400 characters get around 22% better response rates than average, while ones over 1,200 characters run about 11% below it. I'd keep a cold note well under 125 words.

Send on a Tuesday, Wednesday or Thursday morning. Monday is backlog, Friday is checkout.

The numbers underneath these tactics are much weaker than the numbers in the section above, and it would be dishonest to present them the same way. Treat this table as directional:

Not one of these is a controlled study. They are consistent with each other, which is the most that can be said for them.

The case against all of this, taken seriously

I want to give this real space, because the sources that disagree are better than the ones that agree.

Farah Sharghi, a former Google recruiter, argues the "hidden job market" is largely a trap sold by career coaches. Her points are hard to wave away. The genuinely confidential, unposted search is real but effectively reserved for VP and C-suite roles — at early and mid-career level, she says, that market does not operate at your level. The window between a budget being approved and a role going public is real too, but, as she puts it, the relationship has to exist before the opening does; cold-messaging a stranger inside that window rarely does anything. And she makes an argument I found genuinely uncomfortable: recruiters are actively searching for people using LinkedIn Recruiter, and the candidates who surface aren't the ones with the biggest networks, they're the ones who are easiest to understand and find. If your profile reads as a generic list of duties, networking is a distraction from a positioning problem. Her advice is to apply to posted jobs and stop treating that as the consolation prize.

Then there are the hiring managers and job seekers arguing in the comments on Nick Corcodilos's Ask The Headhunter, who describe what happens when direct outreach meets a real company. One commenter reports managers telling him outright that they cannot deal with him, that he has to go through HR and apply to a posted role, because compliance rules and ISO standards have pushed hiring into a process designed to prevent any appearance of favouritism. Another describes a firm posting jobs with explicit instructions not to contact the company, HR, or the manager. A third had a senior internal contact who was blocked from referring him except by handing the application to HR — where it was forced into a mismatched posted role and auto-rejected. And one is simply about the risk: a director who called an executive search firm directly got told down the phone that they had no time for losers who don't send a CV first.

A hiring manager in the same thread makes the mildest and most persuasive version of the objection: an uninvited phone call in response to an ad can easily make you look like an aggressive jerk rather than an eager one.

All of that is true, and none of it is a reason to send 300 applications instead. It's a reason to be selective, to research properly, and to accept that some companies have made themselves unreachable — which is information about the company.

The whole evidence base, graded

Here is every load-bearing number in this post in one place, with what I think it is worth. The top of the table is where the argument actually rests. The bottom is what gets quoted most.

Built from the notebook’s own generated data tables, then re-checked against the underlying sources. The grades are mine.

Where that leaves the actual cadence

The honest synthesis across twenty sources is narrower than any single one of them promises:

• Fix positioning first. Sharghi is right that outreach can't rescue a profile nobody can parse. This is the cheapest thing on the list and the one most people skip.
• Apply to posted roles. That's where the majority of the market actually is. The "70–80% of jobs are never advertised" line traces back to a study from 1974, when employers paid by the line to print listings in newspapers, and got requoted in the New York Times in 1980. It has been recycled for half a century past the conditions that made it true.
• Work the middle of your network, not the edges. Moderately weak ties, per the Science paper — and lean on this harder if you're in a digital industry.
• Add a small number of researched, direct notes per week. Not dozens. Five specific messages that lead with someone else's problem will out-perform three hundred applications, and they cost about the same amount of an evening.
• Respect an explicit "don't contact us." Not on etiquette grounds — it just doesn't work, and the sources include people it visibly cost.

How this was made

The notebook behind this post is a Google NotebookLM project with twenty sources: the Science paper itself, MIT Sloan's write-up of it, an original-data ATS statistics page, the LinkedIn and cold-email practitioner guides, the ex-recruiter video arguing the opposite case, and the Ask The Headhunter comment thread. Source selection was the entire job. A notebook is only as good as what you put in it, and the deliberate inclusion of two sources that contradict the thesis is what makes the output worth reading — without them you get a confident, one-sided briefing.

The audio came from a long custom instruction rather than a one-line prompt. I gave it a six-beat arc: open on application volume, debunk the 75% figure and explain why that strengthens the case for direct outreach, present the evidence that holds up, cover the mechanics, give the counterargument real airtime and engage the ex-Google recruiter honestly rather than dismissing her, and close on a cadence. I also told it explicitly to distinguish between statistics that are well-evidenced and claims that are merely popular, to prefer the primary paper over secondary summaries, and to use the arguing voices in the comment thread.

The notebook also generated structured data tables across all twenty sources, one of which graded every statistic by reliability. That column is where the graded table above came from — it did in minutes the tedious part I would otherwise have skipped, and it is the single most useful thing the tool produced. I re-checked each row against its source and kept the grades my own.

Most of that landed. The debunk is accurate, the counterargument gets genuine time instead of a token sentence, and the hosts do reach for the Science paper rather than the press release about it.

Two things didn't. I asked for roughly twenty minutes and got forty-two — length is a request, not a setting, and the extra time is mostly conversational padding rather than more content. And the format flattens quantitative nuance: the inverted U is the single most interesting finding in the whole notebook, and in audio it becomes "weak ties are good," which is exactly the oversimplification the paper was written to correct. That's the general shape of the tool. It is very good at synthesis and structure across more reading than you'd do yourself, and it is not a substitute for reading the one source that actually matters.

The sources doing the work

The claims above come from these, in roughly the order I'd trust them:

• Rajkumar, Saint-Jacques, Bojinov, Brynjolfsson and Aral, "A causal test of the strength of weak ties," Science 377:6612 (2022) — doi:10.1126/science.abl4476
• MIT Sloan's summary of the same study, for the authors' own plain-language framing
• Harvard Business School and Accenture, "Hidden Workers: Untapped Talent" (2021)
• ResumeAdapter's ATS statistics page, which traces the 75% citation chain and publishes its own pipeline data
• Enhancv's 2025 recruiter survey, and TheLadders' 2018 eye-tracking study
• Liz Ryan / Human Workplace on the Pain Letter
• Cultivated Culture and other practitioner guides on cold email structure
• Farah Sharghi on why the hidden job market is a trap
• Nick Corcodilos, Ask The Headhunter — including the comment thread, which is the most useful part

If you only read one, read the Science paper. It's the only thing here that was designed to be able to prove itself wrong.