Forty applications, two replies, and the advice that arrives next is always the same one. It is a numbers game. Send more.
The number is the part of a job search you fully control, which is probably why every piece of advice reaches for it. It is also the part with the least evidence behind it. The studies that have measured application volume directly keep finding that it does not identify whose search ends soonest, and if anything that the heaviest appliers are the people still looking. The one randomised trial that raised the number of interviews people were invited to did it without raising the number of applications they sent.
Applications per job doubled, and hiring got slower
Greenhouse, which sells applicant tracking software, publishes benchmarks drawn from its own customers. Its March 2026 report, The Hire Standard, describes the dataset as "over 6,000 companies and over 640M applications between 2022 to 2025".1
Applications per job opening went 116, 189, 223, 244 across those four years, which the page labels "More applications per job" and puts at up 111 per cent. Over the same period the average recruiting team went from 10.43 people to 4.62, a 56 per cent cut, and each recruiter's annual application load went from 146 to 746.1
One more number, and it is the one that matters here. Time to fill went from 43.64 days to 59.67, up 37 per cent.1 Twice as many applications arrived per job, and the jobs took about a third longer to close.
Two honest caveats. This is one vendor's view of its own customer base, not a measurement of the labour market, and Greenhouse's own reading of it is that its recruiters coped, the report saying they hired at a higher rate over the period. But from where an applicant stands, the queue roughly doubled in three years and is not moving faster.
That does not make an application worthless. It makes the marginal one worth less than it was, and it means a strategy built entirely on volume is now being run by everybody at once.
The people sending the most applications are the ones still searching
R. Jason Faberman and Marianna Kudlyak had something rare: every application sent on a US job board over a year, with details of both the applicants and the vacancies, which let them control for who was searching and for how many jobs were open as a search went on.2
Their two findings, from the abstract, are that "the number of applications sent by a job seeker declines over the duration of search" and that "longer-duration job seekers send relatively more applications per week throughout their entire search".2
The working-paper version puts numbers on both.3 The average job seeker sent about 2.3 applications in their first week. By the fourth week that had fallen to roughly 1.2 a week, and after six months of searching it had crept back only to about 1.5. Sorted by how long their search eventually lasted, people whose spell ran 13 weeks averaged 1.3 applications a week; people whose spell ran ten months or more averaged 2.3. The slopes are almost identical; the whole difference is one of level. The people who ended up searching for the best part of a year were sending nearly twice as many applications a week as the people who finished in three months, and they were doing it from the first week, before either group knew how it would go.
It would be easy and wrong to read that as applying more making a search longer. The authors' own explanation runs the other way, and they state it plainly: a "dominant income effect in search effort, whereby job seekers exert more effort when their job-finding prospects are poorer".3 People with weaker prospects work harder at it. Effort is the response, not the cause.
Which is why the correlation is worth knowing, and also why it cannot be turned around. This was not an experiment. Nobody was assigned a number of applications, so nothing here estimates what would happen if one particular person sent more next week, and the association is exactly what the authors' own explanation predicts even if volume changed nothing. What it does establish is narrower and still worth having: the weekly count does not identify who will finish sooner. Volume is what a difficult search looks like from the inside, and a search running longer than you expected is the condition under which the advice to send more arrives.
The limits here are real. This is one website, dominated by hourly-paid jobs, between September 2010 and September 2011, with young and less-educated job seekers somewhat over-represented compared with the Current Population Survey — a comparison the authors run themselves.3 It counts applications sent on that site and cannot see the ones sent elsewhere. But it does dispose of the comfortable story in which the people who get hired quickly are the ones grinding out the most applications. In this data, they were sending fewer.
The trial that raised interviews did not raise applications
Michèle Belot, Philipp Kircher and Paul Muller recruited 300 job seekers from Job Centres in Edinburgh and asked them to come in once a week for twelve weeks to search for real jobs on a platform the researchers built over Universal Jobmatch, the UK Department for Work and Pensions site that at the time carried over 80 per cent of official UK vacancies.4
For the first three weeks everyone used a standard interface: type in your own keywords and occupations, as on any job board. That established a baseline for how each person searched when left alone. After three weeks, half of them were offered a different interface. It asked which occupation they were targeting, then returned lists of related occupations — computed from labour market data on where comparable job seekers had actually found work and where skills transferred well, along with the jobs currently open in them.
The set of jobs people looked at got broader, by 0.2 of a standard deviation, which the paper glosses as "the broadening that would occur naturally over 3 months of unemployment". And then: "the total number of job interviews increases by 44%". Among the people who had been searching most narrowly to begin with, the paper reports "a 2-fold increase in total job interviews" against comparable narrow searchers in the control group, concentrated in those who had already been unemployed more than 80 days.4
Then the finding this article turns on. On applications, the paper says: "We find no overall treatment effect on applications, except for a decrease in their geographical breadth."4 Its table of application effects puts the estimate on the total number of applications at 0.01 with a standard error of 0.09 — indistinguishable from zero — and it stays indistinguishable from zero when the sample is split, at 0.08 for the initially narrow searchers and −0.05 for the broad ones.4
The same people. Roughly the same number of applications. Forty-four per cent more interviews.
Where those interviews came from is the part the trial pins down less firmly, and it is worth being exact about. The 0.2 standard deviations of extra breadth is a measure of the jobs participants were shown and looked at, not of the jobs they applied to. On the breadth of the applications themselves the overall estimate is 0.03 with a standard error of 0.20, which is nothing; the paper's introduction still summarises the result as "Job applications become broader", a sentence its own table supports only once the sample is split.4
Split, it holds for the group that mattered. Among those who had been searching narrowly, applications did become occupationally broader, by 0.49 at the 10 per cent significance level, while the already-broad searchers moved the other way at −0.43. The narrow searchers are the ones whose interviews doubled.4 So the defensible claim is this: being shown adjacent occupations raised interviews by 44 per cent without raising the amount of applying, and in the group that drove that increase, the applications moved to different occupations too. That the redirection is what produced the interviews is consistent with the data rather than demonstrated by it.
It also travelled beyond the researchers' own site. Because participants were surveyed weekly about search done elsewhere, the authors could check for spillovers, and report that "the statistical significant impact on job interviews is driven by significantly larger reported interviews due to search outside the lab".4 The occupational information changed how people searched everywhere, not just on the platform in front of them.
The authors are candid about the size of it: "our sample is limited to 300 participants… limited relative to usual labour market studies, with associated limits in terms of power".4 These were job seekers recruited through Job Centres in one Scottish city in the mid-2010s, and the study measures interviews rather than jobs — the paper explicitly asks readers to be cautious about its job-finding results for that reason.
What a job title is carrying
If direction is the thing that moved, the next question is what a direction is made of. One study puts a number on how much work the job title alone is doing.
Ioana Marinescu and Ronald Wolthoff took every vacancy posted on CareerBuilder.com in Chicago and Washington, DC at the beginning of 2011 — a board carrying about 35 per cent of all US vacancies at the time, by comparison with the official JOLTS count — together with the applicant pool each advert attracted.5
Only 20 per cent of those vacancies posted a wage. Among the ones that did, "job titles explain more than 90% of the wage variance". Job titles also "explain more than 80% of the across-vacancies variance in the education and experience of applicants".5
The reason is granularity. In their full sample there were 20,447 distinct job titles against 762 distinct occupation codes from the Bureau of Labor Statistics classification, more than 25 titles per code. Employers distinguish "inside sales representative" from "outside sales representative", and "executive assistant" from "administrative assistant", and the official classification does not.5
Those are two specific outcomes and the claim is worth keeping to them. Among adverts that name a wage, the title predicts nearly all of the variation in what that wage is. Across adverts, it predicts most of the variation in the education and experience of the people who apply. The paper measures neither how a job board retrieves results nor what job seekers do, and this article does not extend it to either.
What it does establish is that two titles you might treat as interchangeable are not interchangeable to the market. They carry different pay and they draw different applicants, and which of them you type is a choice most people make once, at the start, using whatever the last job was called.
The evidence that changing which occupations you look at raises interviews is the Edinburgh trial in the section above, not this paper. What this paper adds is why the label is load-bearing rather than cosmetic: the titles either side of the one you have been searching are attached to different money and different competition.
Who this does not apply to
The broadening result has a group it demonstrably did not help. In the Edinburgh trial the people who were already searching across a wide range of occupations got narrower, not broader, and the interview effect was driven by the narrow searchers.4 If you are already applying across several occupations and geographies, the finding is not about you, and nothing here says what would help instead.
Nor does any of this say the right number is low. Faberman and Kudlyak's one-to-two applications a week is a description of what people did, not a recommendation. If you are sending two a week and could comfortably send five well-aimed ones, send five. The argument is against volume as the only adjustment, not against volume.
And two of the three studies predate the flood the first section described: 2010–11 and the mid-2010s. The current numbers come from one software vendor's customers. Nobody has run the Edinburgh experiment in a market where the average job gets 244 applications, and until somebody does, the honest position is that broadening was worth 44 per cent more interviews then.
What to change before you change the number
Count job titles, not applications. Go back through the last month and write down the distinct titles you applied under. If the list has one or two entries on it, that is the finding pointed directly at you, and it is a much cheaper thing to fix than sending another forty applications.
Get the adjacent titles from data rather than from your own imagination, which is the part the trial automated: the occupations closest to yours have already been worked out and published, along with how the nearness was calculated.
Then run a search under each neighbouring title and read what comes back, because a title you have never used sits on different pay and a different applicant pool, and you cannot tell which from the label. Some will turn out not to be real options, though fewer than the requirements list suggests.
Keep the count roughly where it is while you do it. The people in the trial did not send more applications, and the ones whose interviews doubled were applying under different occupations by the end.
The number of applications is the part of a job search that feels like effort, and it is the part an employer can see least. Before you double it, spend an hour on the list of words you have been typing into the box.
