The Hiring Process Got Longer. ATS Data Shows How Much of It Is AI

Greenhouse counted 244 applications per job in 2025, up from 116 in 2022. Its time-to-fill measure rose 37% over the same years. The hiring process did get longer, and AI job applications are part of the reason. They are not most of it. Roughly half of that doubling is explained by employers posting fewer jobs, not by candidates sending more applications.

Short answer: the hiring process in 2026 takes longer than it did in 2022, mostly because each hire now involves more interviews. Applications per job doubled, but about half of that rise comes from a 40% fall in job openings. AI application tools add volume on top, and no vendor has yet measured exactly how much.

What ATS recruiting benchmarks actually measure

An applicant tracking system, or ATS, is the software a company uses to post jobs, receive applications and move candidates through interviews. Because every step is logged, the vendors sit on the best hiring process data that exists. They publish slices of it as recruiting benchmarks.

That makes these reports valuable and partial at the same time. They describe the customers of one vendor, measured on that vendor's definitions. The figures below are compiled with their sample sizes and windows stated, so you can judge each one on its own terms.

Applications per job is not applications per hire

Greenhouse reports applications per job, which divides applications by postings. Ashby reports applications per hire, which divides applications by people actually hired. A job can close without a hire, so the second ratio runs higher and moves differently. Treat the two as separate series that point the same way, never as one series.

The samples differ too. Greenhouse's March 2026 report covers over 6,000 companies and 640 million applications from 2022 to 2025. Ashby's April 2026 productivity report covers 109 million applications and 247,000 jobs from January 2021 to March 2026. Both customer bases lean toward technology companies and venture-backed firms. Neither is a random sample of employers, and neither claims to be.

The first four rows are vendor customer data. The last row is an official statistic covering the whole US economy, included because it supplies the denominator the vendor reports leave out.

Applications per job doubled between 2022 and 2025

Both vendors agree on direction and roughly on size. The volume reaching a single requisition about doubled in three years.

Greenhouse: 116 to 244 applications per job

Greenhouse's yearly averages run 116 in 2022, 189 in 2023, 223 in 2024 and 244 in 2025. Greenhouse calls that a 111% increase. Computed from the rounded figures it is 110%, which is the kind of gap you should expect when a headline is calculated from unrounded data.

The shape matters more than the endpoint. The biggest single jump came between 2022 and 2023, which is before AI auto-apply tools were in wide use. Growth has slowed every year since.

Ashby: a peak in late 2024, then a plateau

Ashby's series moves the same way on a different denominator. Applications per hire rose from 121 in Q1 2021 to a peak of 319 in Q4 2024, and Ashby reports 291 today. Ashby's own text says every 2025 hire needed more than 300 applications on average, so the current figure is a modest easing rather than a reversal.

My read is that the plateau is the more important finding. If AI tools were still multiplying volume, the curve would be steepening in 2025. It is not.

Half the application flood is a missing denominator

Applications per job is a ratio. It rises when applications go up, and it rises just as fast when jobs go down. Most coverage of these benchmarks reads only the numerator.

The denominator moved a lot. US job openings peaked at 12.3 million in March 2022, fell to 10.9 million by December 2022 and reached 6.55 million in December 2025. The BLS release for August 2026 puts openings at 7.1 million, with a hires rate of 3.3%.

Do the arithmetic once. From December 2022 to December 2025, openings fell by about 40%. Hold total applications constant and divide by 40% fewer openings. Applications per opening rise by about 66%, with no change in how anyone applies.

Greenhouse measured a rise of about 110%. So a falling job count can account for roughly 66 points of the 110. What is left over is about 26% more applications per opening than the job market alone predicts. That residual is where AI tools, layoffs and remote job reach all have to fit together.

This is illustrative arithmetic, not a decomposition. JOLTS covers the whole US economy, and Greenhouse's customers are a tech-heavy global subset. The point survives the imprecision: a large share of the hiring process strain comes from fewer jobs, not only from AI tools.

The same problem appears in other labour data. Headline figures about AI and jobs often turn out to be a ratio with a moving denominator. That is the pattern in the breakdown of why entry-level hiring data conflicts. It also runs through the comparison of AI layoff claims against payroll data.

Where the hiring process got longer, and where it did not

The timing data is less uniform than the volume data. Whether the hiring process got longer depends on which vendor you read and which clock you use.

Time to fill and time to hire are different clocks

Time to fill usually starts when a job opens and ends when an offer is accepted. Time to hire usually starts when a specific candidate applies. A role that sits open for weeks before the right person applies shows a long time to fill and a short time to hire.

Greenhouse reports the first clock rising sharply. Its time-to-fill measure went from 43.64 in 2022 to 59.67 in 2025, up 37%. The public page does not label the unit, though the values read as days.

Ashby's clocks look steadier. Its April 2026 report gives medians rather than averages. Business roles take a median of 30 days to hire and technical roles 40 days. Median time to first fill is 56 days for business roles and 76 for technical. Ashby's EMEA report, covering more than 180,000 jobs, finds time to hire has stayed in a 37 to 43 day range since 2021.

Ashby's text describes time to first fill as about 8 and 10 weeks in one place and gives medians of 56 and 76 days in another. The medians are used here because they are the more specific figures.

Read together, the table says something narrower than "hiring takes longer." Once a good candidate applies, the process runs about as fast as it did in 2021. The extra time is spent waiting for that candidate to surface, and then putting them through more interviews.

The interview loop is where the time went

This is the finding I would put in front of a hiring manager. More applications did not mean more screening time per hire as much as it meant more interviews per hire.

Fewer interviews per applicant, more per hire

Ashby reports that in 2021 about 7 to 8% of applications led to an interview. Today it is between 3.6% and 4.7%, depending on role type. A single applicant's odds roughly halved. Yet each hire now takes more interviews, because the pool is larger and teams are sampling more of it before deciding.

Business roles average 11.7 interviews per hire, up from about 8 in 2021. Technical roles average 17.6, up from about 11. Ashby puts total interview time per technical hire at 23.3 hours, against 12.2 for business roles.

There is good news at the back of the funnel. Once a candidate reaches interview, the odds of an offer are higher than in 2021. Ashby reports offer conversion from interview at 10.4% for business roles and 7.3% for technical roles, and says both have passed 2021 levels. Screening is getting more selective at the front, so the people who reach a panel are more likely to be right.

That cuts against the popular story that AI applications have broken hiring. The data says the front of the funnel is noisier and the back of it is working better than before. That is a capacity problem, and capacity problems have process fixes. How to keep the back end honest when candidates use AI in interviews too is covered in the analysis of what technical interviews still measure.

Recruiter workload: fewer recruiters, more hires each

The quiet story in the Greenhouse data is headcount. Recruiting teams shrank while volume doubled, and the remaining recruiters closed more roles than before.

Greenhouse reports recruiters per organisation falling from 10.43 in 2022 to 4.62 in 2025, down 56%. Annual applications per recruiter rose from 1,610 to 4,890. Monthly hires per recruiter rose from 2.2 to 4.9. Greenhouse also reports an 8% improvement in jobs closed with a hire.

Two vendors, two definitions of productivity

Ashby measures the same idea per quarter. Hires per recruiter bottomed at 4.5 per quarter in early 2023 and reached 7.3 by Q1 2026. Its EMEA report puts the regional figure between 3.8 and 4.8 per quarter.

Greenhouse's 4.9 per month and Ashby's 7.3 per quarter cannot both describe the same kind of recruiter. The vendors almost certainly count different people as recruiters and attribute hires differently. Use each figure to track your own trend against the same vendor's benchmark. Do not use them to argue about whether your team is productive in absolute terms.

Ashby's capacity analysis adds one more detail. The busiest recruiting coordinators handle around 27 jobs a week, up from about 22 a year earlier, across data from November 2024 to August 2025. Combine that with the interview counts above and the bottleneck is clear. Scheduling and interviewer time, not résumé reading, is what stretches the calendar.

What AI job applications did, and what nobody has measured

Here is the uncomfortable part for anyone writing about AI job applications. The ATS vendors have the logs, and none of their public benchmarks puts a number on the share of applications written or submitted by AI.

The attribution is qualitative

Ashby's startup report covers 1,200 venture-backed startups and 11 million applications. It says the volume increase "has only grown through the ease of applying with AI and the flood of fraudulent candidates." That is a fair description. It is also one sentence with no measured share attached.

Greenhouse's March 2026 page says AI "continues to serve as a key lever for both candidates and recruiters." Again, no share. A candidate AI-use percentage attributed to Greenhouse circulates in secondary coverage, and I could not trace it to a public Greenhouse page, so it is not used here.

The number that drifted

In a July 2026 Fortune interview, Greenhouse CEO Daniel Chait was reported as citing around 254 applicants per job ad and applications per recruiter up 412%. Neither figure appears on Greenhouse's benchmark page, which gives 244 and 203.7%. The interview may draw on newer or different data. It gives no window or sample, so the benchmark figures are the ones to cite.

This is the common failure the standard warns about. A striking number repeats across coverage and starts to feel corroborated, when it traces to one unsourced remark. The same dynamic is behind much of the volume problem in publishing, examined in the economics of the content flood.

Where this benchmark argument is weakest

The denominator argument above is the spine of this post. It is also the part most open to challenge, and here is where it bends.

Customer base drift can fake a trend

A vendor that adds larger or more popular employers each year will see applications per job rise. That holds even if no candidate changes behaviour. Neither Greenhouse nor Ashby publishes a fixed-cohort version of its volume series. Ashby did use a fixed cohort for its 2025 hiring analysis, which shows the method exists. Until the volume series gets the same treatment, part of the rise could be sample drift.

The openings comparison also crosses datasets. JOLTS measures US openings across every industry. Greenhouse measures its customers' postings worldwide. Tech and venture-backed hiring fell harder than the economy in 2022 and 2023, which would make the true denominator effect larger than the 66% used here. Remote roles reaching global applicant pools would push the other way. The direction of the argument holds. The precise split does not, and the figure above says so.

The reporting incentive also deserves a sentence. Both vendors sell tools to manage application volume, so a report showing volume rising is also a sales asset. That does not make the data wrong. It does mean the framing of "AI flooded hiring" suits the publisher, and you should read the ratios rather than the press releases.

Frequently asked questions

How to read your own hiring numbers this week

Pull applications per opening and interviews per hire for your last four quarters out of your own ATS. Then divide your application count by your open roles yourself, rather than trusting the dashboard ratio. If your openings fell, you now know how much of your volume problem is really a demand problem.

Then look at interviews per hire by role. If a technical role is running well above 17.6, the fix is a tighter scorecard and an earlier structured assessment, not another screening tool. The case for testing skills earlier is laid out in the review of skills-based hiring assessments.

Related reading