The 2026 Job Application Paradox

TL;DR. The average office opening now draws about 242 applications, so one application converts at roughly \(1/242 \approx 0.41%\). Ranked against elite institutions, that lands a routine task below Harvard, MIT, Stanford, a McKinsey or SpaceX offer, a Rhodes Scholarship, and a spot on Jeopardy. But the number misleads in both directions. It is a per-application rate inside a pile flooded with low-effort and AI-generated submissions, so a persistent applicant’s cumulative odds across many tries are higher than 0.4%. Yet the popular rebuttal, that 100 applications buy a roughly 34% shot at an offer, is also wrong: it pins the crowd’s average rate onto one person. An individual applying cold to competitive roles usually sits far below that average, where a single offer can take many hundreds or thousands of applications. The one-in-three odds people genuinely feel are real, but they live at the interview stage, not per application. The bar did not rise to astronaut level. The denominator did. Every figure below is sourced and the math is worked through in full.

1. The Ladder, Easiest to Hardest

#

Each row shows the odds, a “1 in X” reading, and the kind of number it is (this last column is the key to reading the table honestly). Source numbers map to the references at the end.

The office job line sits below elite admissions and a tier of legendary jobs. The rest of this note is about whether that ranking means what it appears to mean. It does not, and the reason is more interesting than the headline.

2. The Core Number: Where 0.4% Comes From

#

The average opening now draws about 242 applications and fills roughly one seat [24]. Treating a single application as one independent trial, the per-application probability of success is

$$ p_{\text{app}} = \frac{\text{hires}}{\text{applications}} = \frac{1}{242} \approx 0.00413 = 0.413%. $$

That is the “1 in 250” headline (242 rounds to about 250). Five years ago a typical posting drew closer to 100 applications, so the same calculation would have given roughly 1%.

The coin-flip equivalence. Most people feel coin flips more naturally than tail percentages. How many heads in a row matches this probability? Solve

$$ \left(\tfrac{1}{2}\right)^{k} = \frac{1}{242} \quad\Longrightarrow\quad k = \log_{2} 242 \approx 7.92. $$

So one application converting is about as likely as flipping 8 heads in a row, since

$$ \left(\tfrac{1}{2}\right)^{8} = \frac{1}{256} \approx 0.391%, $$

which sits right beside the 0.413% figure. Good intuition pump. Now the part the intuition pump hides.

3. “Per Application” Is Not Your Real Odds

#

The seductive error is reading 0.4% as your chance of getting a job. It is not. It is the chance that one specific submission lands. A real job hunt is many submissions. If each application is an independent trial with success probability \(p\), the probability of landing at least one offer from \(n\) applications is the complement of failing every time:

$$ P(\text{at least one offer}) = 1 - (1 - p)^{n}, \qquad p = \tfrac{1}{242}. $$

Evaluated across a realistic range of \(n\), assuming a single applicant’s personal rate equalled the pile average \(p = 1/242\):

So odds do compound with volume: the per-application figure understates the cumulative chance across a long search, and that much is sound. The trap is the next step, reading the 33.91% as your chance after 100 applications. Every row here assumes your personal rate equals the pile average, \(1/242\). It almost never does, and the error is not symmetric.

A common wrong shortcut. Even within the pile-average model, people sometimes just multiply, \(n \cdot p\). For \(n = 100\) that gives \(100 \times 0.413% = 41.3%\), which overshoots the model’s own 33.9%. The linear estimate is only accurate when \(np \ll 1\), because the exact expression expands as

$$ 1 - (1 - p)^{n} \approx np - \binom{n}{2} p^{2} + \cdots $$

The quadratic term is what pulls 41.3% down to 33.9%. But both numbers are properties of the pile-average pool, not of any one applicant, which is the larger correction and the subject of the next section.

The Flaw of Averages: Your Rate Is Not the Crowd’s

#

The pile of 242 is not 242 interchangeable people. It is a handful of strong, referred, or precisely matched candidates who convert at high rates, sitting on top of a large mass of low-effort and AI-blasted submissions that convert at essentially zero. The one hire comes off the top of that distribution. Averaging across the whole pile produces \(1/242\), but that average describes the pile, not any person in it. Reusing it as an individual’s rate is the flaw of averages, and two independent facts make it overstate a real applicant’s odds, both in the same direction.

First, the curve is concave. The at-least-one function \(1 - (1-p)^n\) bends over as \(p\) grows, so by Jensen’s inequality its value at the average rate exceeds the average of its values across a spread of rates. The 33.91% is therefore a ceiling on the average applicant. Because the pile is heavily right-skewed, most applicants sit below even that average rate, so the typical person is already under 34% before anything else is counted.

Second, the trials are not independent. Your 100 applications share one resume, one skill set, one band of target roles. A rejection is highly informative about the next, because the cause is correlated rather than redrawn each time. The binomial model assumes 100 independent coin flips; a real search is closer to one weighted coin flipped 100 times. Positively correlated failures cluster, which pushes \(P(\text{at least one})\) further below the independent number.

Put your own rate in and the picture matches lived experience. If you are applying cold to competitive roles and your true rate is nearer \(1/1000\), then 100 applications give about a 9.5% chance of an offer, and reaching 90% confidence takes roughly 2,300 applications, not 557. Searches that run into the hundreds or thousands of applications per offer are not anomalies; they are what the curve looks like below the mean, which is where most people applying cold actually live.

This is also why the headline and the rebuttal can both feel true. The brutal \(1/242\) is the per-application rate, the product of every stage of the funnel: surviving the automated screen, reaching a human, landing the interview, winning the offer. The comforting one-in-three is a conditional rate that only exists deep in that funnel, at the interview, among the few applications that got there. Sending more applications multiplies your shots at the very top of the funnel, where almost everything is filtered out, not at the interview, where the one-in-three lives. The two numbers are not competing estimates of one quantity; they are rates at different stages, and the mistake is gluing a late stage’s odds onto the first stage’s denominator.

The Spread Hidden in “242 Tries”

#

“Average” is the wrong word for it. The number of applications until the first offer follows a geometric distribution, which is heavily right-skewed, so the mean overstates the typical search. At the random rate the mean is 242, but the median is only 168, and the standard deviation is about 241, nearly as large as the mean itself.

$$ \mathbb{E}[X] = \frac{1}{p}, \qquad \text{median} = \left\lceil \frac{\ln 2}{-\ln(1-p)} \right\rceil, \qquad \mathrm{SD}(X) = \frac{\sqrt{1-p}}{p}. $$

The gap between a median of 168 and a mean of 242 is the signature of the skew: most successful searches finish before the “average,” while a long tail of unlucky ones drags the mean up. The qualified pool has the same shape in miniature, with a mean of 25, a median of 17, and a standard deviation near 24. So “expect 242 tries” is not a forecast of your search, it is the center of a very wide, lopsided distribution.

A Cleaner Mental Model: The Poisson Approximation

#

For small \(p\) and moderate \(n\), the at-least-one curve is almost exactly an exponential, because \((1-p)^n \approx e^{-np}\) when \(p\) is small:

$$ 1 - (1-p)^{n} \approx 1 - e^{-np}. $$

This is not just tidy, it is accurate. Across the realistic range it sits within a fraction of a percentage point of the exact value: at \(n = 100\) it gives 33.85% against the exact 33.91%. The practical payoff is a mental shortcut far better than the linear one: your chance of at least one offer is about \(1 - e^{-n/242}\), the same curve that governs radioactive decay and waiting times. It climbs fast, then bends, which is the diminishing-returns shape plotted above.

How Many Applications for a Given Confidence

#

The most useful version of the formula answers the question people actually ask. Not “what are my odds at \(n\),” but “how many applications until I am reasonably sure.” Solving \(1-(1-p)^n \geq P\) for \(n\):

$$ n = \left\lceil \frac{\ln(1-P)}{\ln(1-p)} \right\rceil. $$

At the random rate, coin-flip confidence (50%) takes 168 applications, 90% takes 557, and 95% takes 724. In the qualified pool those same targets collapse to 17, 57, and 74. That contrast is the whole argument in one line: being in the signal turns a 557-application slog into 57. As a check, the \(1 - 1/e\) point (about 63.2% confidence) lands exactly at the mean, 242 applications in the random pile and 25 in the qualified pool, which is what the expected-value math predicts. Both columns describe the two pools, not any one applicant; substitute your own rate, which for cold applications usually sits below even the random pile, and every target grows in proportion.

4. Noise Versus Difficulty: The Qualified-Pool Model

#

Why is the per-application number so punishing in the first place? Because the denominator is padded with submissions that were never real contenders. Model the pile with three quantities:

• \(N\) = total applications per opening, about 242 [24]
• \(q\) = genuinely qualified, serious applicants, about 20 to 30 (call it 25)
• one hire, which goes to a qualified applicant

A random application’s odds are \(1/N\). A qualified applicant’s odds are roughly \(1/q\), because the seat is contested among the serious few:

$$ p_{\text{app}} = \frac{1}{N} = \frac{1}{242} \approx 0.41%, \qquad p_{\text{qualified}} \approx \frac{1}{q} = \frac{1}{25} = 4.0%. $$

The ratio is the noise penalty baked into the headline:

$$ \frac{p_{\text{qualified}}}{p_{\text{app}}} = \frac{N}{q} = \frac{242}{25} \approx 9.7. $$

A qualified candidate is roughly 10 times more likely to convert than the headline implies. Define the signal fraction \(f = q/N\), the share of the pile that is real. Then the whole relationship is one clean equation:

$$ p_{\text{qualified}} = \frac{1}{fN} = \frac{p_{\text{app}}}{f}, \qquad \frac{p_{\text{qualified}}}{p_{\text{app}}} = \frac{1}{f}. $$

Expected effort (geometric distribution). If each application converts with probability \(p\), the number of applications until the first offer follows a geometric distribution with mean \(1/p\):

$$ \mathbb{E}[\text{applications to first offer}] = \frac{1}{p}. $$

For a random applicant that is \(1/p_{\text{app}} = 242\) tries. For a qualified applicant it is \(1/p_{\text{qualified}} \approx 25\). Same labor market, an order of magnitude apart, entirely from being in the signal rather than the noise.

5. The Big Caveat: Did Office Jobs Really Get Harder Than SpaceX?

#

Not in the way it sounds, and the math says exactly why. This is the part to internalize before sharing the ladder.

First, the obvious point: SpaceX, Google, and McKinsey are themselves white collar office jobs. So “an office job is harder than SpaceX” is partly comparing office jobs to other office jobs. Fun line, not the deep truth.

Second, the subtle point. SpaceX’s own court filing states that about 1% of applications result in a hire [16]. That is a per-application number, exactly like the office job’s 0.413%. Comparing like to like:

$$ \frac{p_{\text{SpaceX}}}{p_{\text{office}}} = \frac{0.01}{0.00413} \approx 2.42. $$

A SpaceX application converts about 2.4 times better than a generic office application. Not because SpaceX is easier, but because its applicant pool is cleaner. Under the signal-fraction model, if both fill one seat and the underlying bar is comparable, the per-application rate is proportional to the signal fraction, so

$$ p_{\text{app}} \propto f \quad\Longrightarrow\quad \frac{f_{\text{SpaceX}}}{f_{\text{office}}} \approx \frac{p_{\text{SpaceX}}}{p_{\text{office}}} \approx 2.4. $$

In plain terms, SpaceX’s pile is roughly 2.4 times less diluted by noise. Nobody one-clicks into a rocket company on a whim. Applicants self-select, and the firm filters hard with referrals, work samples, and many interview rounds. A generic posting gets blasted by hundreds of low-effort and AI-mass-applied submissions, so its \(f\) collapses and its per-application \(p\) collapses with it.

The honest takeaway is not that office jobs became more elite than astronaut selection. It is that the per-application math for ordinary jobs has been broken by sheer volume, until a single click converts about as rarely as a shot at the world’s most selective employers. The bar did not rise to SpaceX’s level. The denominator did.

6. Reading the Ladder Honestly: Ratios and Log Scale

#

It is tempting to say “an office application is 10x rarer than getting into Harvard.” The arithmetic is real,

$$ \frac{p_{\text{Harvard}}}{p_{\text{office}}} = \frac{0.042}{0.00413} \approx 10.2, $$

but Harvard’s 4.2% is per applicant and the office 0.413% is per application. The comparison is dramatic precisely because it crosses metric types. Here is the full ladder as a ratio to the office line and on a base-10 log scale, which is the honest way to see the spread.

The whole ladder spans

$$ \log_{10} \frac{0.73}{0.0003} \approx 3.4 \text{ orders of magnitude}, $$

from “3 in 4” down to “1 in 3,300.” The office job sits at \(-2.38\), deep in elite-selection territory and only about half an order of magnitude above NASA. Read it as a striking coincidence of magnitudes, not as proof that office jobs out-filter astronaut programs.

7. What Actually Changed Since 2021: The Volume Math

#

The collapse in per-application odds is almost entirely a denominator story, and it decomposes cleanly.

Applications per opening rose from about 100 to about 242 [24]:

$$ \frac{242}{100} = 2.42 \quad (\text{a } 142% \text{ increase}). $$

If hires per opening \(H\) are unchanged, then \(p = H/N\) and per-application odds fall in exact proportion to the growth in \(N\):

$$ \frac{p_{2026}}{p_{2021}} = \frac{N_{2021}}{N_{2026}} = \frac{100}{242} \approx 0.41. $$

So per-application odds dropped by about 59% with no change whatsoever in the hiring bar. That is the core argument, quantified.

Recruiter load. Total annual application volume rose 411% (a multiplier of \(1 + 4.11 = 5.11\)) while recruiting teams shrank about 55% (a multiplier of \(1 - 0.55 = 0.45\)) [30] [31]. Applications per recruiter therefore scaled by

$$ \text{load multiplier} = \frac{5.11}{0.45} \approx 11.4. $$

Each recruiter now faces roughly 11 times the applications in the same hours. That forces aggressive automated filtering, which rewards more mass-applying, which feeds the loop. The per-person fuel is AI: job seekers who use AI complete about 41% more applications than those who do not, per Capterra’s global survey of roughly 3,000 job seekers across 12 countries [32] [33].

Sensitivity: How Much Does the 242 Drive This?

#

The 242 figure is the softest, most load-bearing input in this note, so it is worth asking how much the conclusions move if it is off. Because the per-application odds are exactly \(1/N\), the headline scales inversely with the denominator, and the confidence targets scale with it. Varying only \(N\):

Even at the optimistic end (\(N = 100\), roughly the pre-2021 norm), reaching 90% confidence still takes 230 applications at the random rate. The exact numbers move with \(N\); the conclusion, that per-application odds are punishing and only volume or signal rescues them, does not.

8. The Macro Backdrop: The Market Did Cool

#

The per-application math explains most of the pain, but it is not the whole story. White collar hiring genuinely softened. The US job-openings-to-unemployed ratio dipped below 1.0 in March 2025 and bottomed near 0.87 in December 2025, and information-sector openings fell about 33% year over year, the steepest of any private sector, per Bureau of Labor Statistics JOLTS data and Indeed Hiring Lab analysis [34] [35]. So the picture is two effects stacked: a modestly tighter market, and a per-application denominator that exploded. The second effect is the larger and less understood one, which is why this note focuses on it.

9. What This Means for You

#

The model points to a precise strategy, not vague encouragement. Your real odds are \(p_{\text{qualified}} = p_{\text{app}} / f\), so you win by raising \(f\) for your own application or by raising \(n\) while keeping quality intact.

• Get a referral. A referred application effectively jumps the noise pile, pushing your personal \(f\) toward 1. This is the single largest lever in the model.
• Apply with intent to roles you fit. Each application where you are genuinely qualified carries odds near \(1/q\) (a few percent), not \(1/N\) (a fraction of a percent).
• Choose quality \(n\) over spam \(n\). A hundred targeted applications, each one raising your own \(f\), move you up the rate curve; a hundred copies of one blast keep you at the bottom of it, in exactly the pile being filtered out. Volume only pays at a decent personal rate.
• Use any warm human contact. A single human who actually reads your file changes the trial from “1 of 242” to something near “1 of \(q\).”

The volume game is rigged against volume. The way out is signal.

10. Methodology and Data Quality

#

Not all of these figures are equal, and an honest analysis says so. They fall into three tiers.

A few specific notes on the trickier numbers:

• Yale Law is the most selective US law school. Yale’s own ABA Standard 509 disclosure reports 226 offers from 5,562 completed applications, a 4.06% rate for the fall 2025 class [6], and US News reported 5.25% (229 of 4,358) for fall 2024 [7]. The “5%” in the ladder is the conservative reading. A commonly repeated “8 to 9%” figure is stale or a confusion with Stanford Law’s recent rate.
• The 242 applications figure is the softest input and the most load-bearing, so it deserves scrutiny. It traces to Business Insider data cited in a February 2026 analysis [24]. Greenhouse’s own benchmark data is in the same range (about 222 applications per opening in early 2024) and independently confirms the 411% volume surge [30] [31].
• The 41% AI figure is from Capterra’s 2024 Job Seeker AI Survey of about 3,000 respondents across 12 countries [32] [33]. It is a global, not regional, survey, and it is a distinct metric from the 411% volume surge.
• Metric types are not interchangeable. The single biggest analytical caveat in this piece is that the ladder mixes per-applicant rates (most schools, NASA, Rhodes) with per-application rates (the office job, SpaceX, Google, Y Combinator), a completion rate (Navy SEAL), and a pure-chance lottery (green card). The comparison is illustrative, not a strict like-for-like ranking. Section 5 works through why that distinction matters most for the SpaceX comparison.

11. Limitations and Assumptions

#

• The independence assumption in Section 3 (\(1 - (1-p)^n\)) overstates real odds, because applications from one person are correlated. Read those numbers as an upper bound on a persistent applicant’s odds.
• The qualified-pool model in Section 4 treats one hire per opening and equal odds among qualified applicants. Real hiring is messier, but the order-of-magnitude gap between per-application and per-qualified odds is robust to reasonable choices of \(q\).
• “Office job” is a broad category. Per-application odds vary widely by role, seniority, sector, and geography. The 242 figure is an average, and tails in both directions exist.
• Several elite-employer rates are derived from total-applications-to-total-hires, which blends roles and regions. They are directionally sound but not precise per-posting figures.

Note to self: this was put together with help from Claude (Anthropic), which did the data gathering and the probability math. Irony fully acknowledged: using AI to explain why landing a job got so brutal in a market that AI itself flooded.

References

#

All links were live as of May 2026. Verify before citing.

1. National Association for College Admission Counseling (NACAC). “Selectivity: Acceptance Rates at 4-Year Colleges.” https://www.nacacnet.org/selectivity-acceptance-rates-at-4-year-colleges/
2. Sandboxx News. “New Navy report reveals rare SEAL training attrition data” (2024). https://www.sandboxx.us/news/special-ops/new-navy-report-reveals-rare-seal-training-attrition-data/
3. SOF Prep Coach. “How hard is Navy SEAL training and how to pass BUD/S?” https://sofprepcoach.com/how-hard-is-navy-seal-training/
4. CollegeSimply. “The Juilliard School Admission” (181 admitted of 2,020). https://www.collegesimply.com/colleges/new-york/the-juilliard-school/admission/
5. U.S. News & World Report. “Juilliard School Applying.” https://www.usnews.com/best-colleges/juilliard-school-2742/applying
6. Yale Law School. ABA Standard 509 Information Report (fall 2025: 226 offers of 5,562 applications, 4.06%). https://law.yale.edu/sites/default/files/documents/pdf/std509inforeport.pdf
7. U.S. News & World Report. “20 Law Schools That Are Hardest to Get Into” (Yale 5.25%, fall 2024). https://www.usnews.com/education/best-graduate-schools/the-short-list-grad-school/articles/law-schools-that-are-hardest-to-get-into
8. Federal Bureau of Investigation. “What I See: A Message from the Assistant Director of the FBI’s Training Division” (48,000+ applications, ~1,900 graduates over two years). https://www.fbi.gov/news/press-releases/what-i-see-a-message-from-the-assistant-director-of-the-fbi-s-training-division
9. MIT Admissions. “Admissions Statistics” (Class of 2029: 1,334 of 29,281, 4.6%). https://mitadmissions.org/apply/process/admissions-statistics/
10. Harvard Magazine. “Harvard’s Class of 2029 Admissions Data” (2,003 of 47,893, 4.2%). https://www.harvardmagazine.com/university-news/harvard-admissions-class-2029-admissions-data-ethnicity
11. Harvard Office of Institutional Research. Admissions data series. https://oira.harvard.edu/
12. The Stanford Daily. “Class of ‘29 admitted to the Farm” (recent rate ~3.6%). https://stanforddaily.com/2025/04/03/stanford-admits-class-of-2029/
13. Bloomberg. “Y Combinator’s Latest Batch Is 35% AI Startups” (record 24,000 applications, ~1% accepted). https://www.bloomberg.com/news/articles/2023-06-27/y-combinator-s-latest-batch-is-35-ai-startups
14. Hacking the Case Interview. “McKinsey Acceptance Rate” (~200,000 applications, ~2,000 hires, ~1%). https://www.hackingthecaseinterview.com/pages/mckinsey-acceptance-rate
15. CaseCoach. “How Selective are Bain, BCG and McKinsey Through the Application Process?” https://casecoach.com/b/how-selective-are-bain-bcg-and-mckinsey-through-the-application-process/
16. Michael Sheetz (CNBC), reporting SpaceX’s legal complaint: “only about 1% of applications result in a hire.” https://x.com/thesheetztweetz/status/1704501132133912830
17. U.S. Courts. “Supreme Court Procedures” (accepts 100 to 150 of more than 7,000 cases per year). https://www.uscourts.gov/about-federal-courts/educational-resources/about-educational-outreach/activity-resources/supreme-court-procedures
18. D.C. Bar. “Getting Heard at the High Court” (~80 of 7,000 to 8,000 petitions). https://www.dcbar.org/news-events/publications/d-c-bar-blog/getting-heard-at-the-high-court-what-you-need-to-k
19. Scholarships360. “Rhodes Scholarship” (~0.7%; ~100 scholars from thousands of applicants). https://scholarships360.org/scholarships/rhodes-scholarship/
20. Fox Business. “Goldman Sachs internships have 0.7% acceptance rate with 360,000 applicants” (2,600 selected). https://www.foxbusiness.com/lifestyle/landing-goldman-sachs-internship-difficult-nasa-astronaut-acceptance
21. eFinancialCareers. “A brief history of the Goldman Sachs internship acceptance rate.” https://www.efinancialcareers.com/news/goldman-sachs-internship-acceptance-rate
22. Ken Jennings. “Frequently Asked Questions” (tens of thousands of applicants for ~400 spots). https://www.ken-jennings.com/faq
23. Jeopardy.com. “Contestant FAQs.” https://www.jeopardy.com/be-on-j/faqs
24. The Interview Guys. “The Average Job Opening Now Gets 242 Applications (And a 0.4% Chance of Landing It)” (Business Insider data via Novorésumé, Feb 2026). https://blog.theinterviewguys.com/the-average-job-opening-now-gets-242-applications/
25. U.S. Department of State. “DV-2025 Selected Entrants” (~55,000 visas from 19,927,656 qualified entries). https://travel.state.gov/content/travel/en/us-visas/immigrate/diversity-visa-program-entry/dv-2025-selected-entrants.html
26. Quartz. “Here’s why you only have a 0.2% chance of getting hired at Google” (~7,000 hires from ~3M applications). https://qz.com/285001/heres-why-you-only-have-a-0-2-chance-of-getting-hired-at-google
27. Fortune. “NASA just picked its newest astronauts, only 0.1% made the cut” (10 of 8,000+, ~0.125%). https://fortune.com/2025/09/24/nasa-astronaut-class-2025-candidates-six-figure-salaries-150k-future-missions-space-moon-mars
28. NASA. “NASA Selects All-American 2025 Class of Astronaut Candidates.” https://www.nasa.gov/news-release/nasa-selects-all-american-2025-class-of-astronaut-candidates
29. NCAA Research. “Estimated Probability of Competing in Professional Athletics.” https://www.ncaa.org/sports/2015/3/6/estimated-probability-of-competing-in-professional-athletics.aspx
30. RecTech Media. “Greenhouse Report: More Applications, Fewer Recruiters” (411% application surge, ~55% smaller recruiting teams since 2022). https://www.rectechmedia.com/blog/2026/5/9/greenhouse-report-more-applications-fewer-recruiters
31. Greenhouse. “AI has doubled recruiters’ workloads” (~222 applications per opening, Q1 2024). https://www.greenhouse.com/blog/ai-has-doubled-recruiters-workloads
32. BusinessWire / Capterra. “AI-Driven Job Applications Are Taking Over Job Market” (AI users complete 41% more applications). https://www.businesswire.com/news/home/20240829618028/en/AI-Driven-Job-Applications-Are-Taking-Over-Job-Market-and-Creating-New-Challenges-for-Recruiters
33. Capterra. “How To Use AI in Recruitment” (survey methodology: ~3,000 respondents, 12 countries, July 2024). https://www.capterra.com/resources/ai-in-recruitment/
34. U.S. Bureau of Labor Statistics. “Job Openings and Labor Turnover Survey (JOLTS).” https://www.bls.gov/news.release/jolts.htm
35. Indeed Hiring Lab. “March 2026 JOLTS Report: Stable, Depending on What You Do.” https://www.hiringlab.org/2026/05/05/march-2026-jolts-report-stable-depending-on-what-you-do/

Appendix: Notation

#