Too Many Applications, Signal vs Noise
Too Many Applications, Signal vs Noise
By Calin Muresan, co-founder of Wise Step Recruiting. Calin spent about 20 years as a software engineer and cloud architect before he started recruiting engineers.
Hiring used to be a search problem. In 2026, it’s a filtering problem. Between January and September 2026, candidates sent 10.4 million applications through eJobs, one of Romania’s main job boards, up 27% on the same period of 2025 and the highest level in ten years. Employers posted about 160,000 new jobs over the same months, at a slightly slower pace than in 2025. That works out to roughly 65 applications per role, and more than 90 in IT and telecom (StartupCafe, reporting eJobs data, October 15, 2026).
The volume is only half the problem. The other half is that the things employers have always used to tell a strong candidate from a weak one, the CV, the GitHub history, the portfolio site, can now be produced by a model in minutes. You can no longer check, from the artifact alone, whether the work is the candidate’s own.
We’re former software engineers who now run technical searches across Romania and CEE, so we see both ends of this every week. This piece covers how big the flood is, what caused it, why the usual signals broke, how employers are reacting, and what still tells you something about a candidate.
Key Takeaways
- Application volume is outrunning job openings. eJobs logged 10.4 million applications in January to September 2026, up 27%, while new job postings slowed. Greenhouse reports applications per recruiter up 412%.
- Two forces feed the flood: fewer entry-level roles in AI-exposed jobs (clearest in US payroll data), and AI that makes each extra application nearly free. Gartner found 4 in 10 candidates use AI in the application process.
- The classic artifacts lost their value as evidence. A polished CV, an active GitHub profile, and a portfolio site no longer prove who did the work. Gartner predicts 1 in 4 candidate profiles worldwide could be fake by 2028.
- Employers are reacting with short-lived job ads, quieter direct search, and AI screening. Each one cuts volume, and each one has a cost.
- The signal that survives is expensive to fake: live work, a walkthrough of real decisions, and references from people who saw the work.
How big is the application flood in 2026?
Every large dataset we found points the same way: applications are rising far faster than jobs. The Romanian numbers above come from a single platform, but the pattern matches what Greenhouse and LinkedIn report globally.
Greenhouse, an applicant tracking system, has 2,539 applicants for every 10 open roles on its platform, about 254 per job ad across roughly 175,000 live listings. Its CEO, Daniel Chait, told Fortune that applications per recruiter are up 412%, though the article doesn’t give the comparison period (Fortune, August 7, 2026). A year earlier, LinkedIn was already reporting an average of 11,000 applications a minute, up 45% in a year, which a recruiter quoted by The New York Times called an “applicant tsunami” (eWeek, reporting The New York Times, June 25, 2025).
In Romania, the gap between supply and demand is easiest to see in the first half of 2026. eJobs published just over 110,000 new jobs, 15% fewer than in the first half of 2025, while candidates sent more than 6.7 million applications, 27% more (StartupCafe, reporting eJobs data, July 20, 2026).
One caveat before going further. These figures come from the platforms themselves, and each counts differently. Use them for the direction and the rough size of the gap, not for precise ratios.
What is driving the surge: fewer junior roles, cheaper applications
Two things are happening at once. There are fewer entry-level jobs to go around, and each application now costs the candidate almost nothing to produce.
The first force is the shrinking entry level. Stanford’s Digital Economy Lab, working from ADP payroll records for millions of US workers through June 2026, finds that employment of 22- to 25-year-olds in AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers (Brynjolfsson, Chandar and Chen, “Canaries in the Coal Mine,” revised August 12, 2026). The authors find the losses concentrated in jobs where AI substitutes for human work rather than adding to it, and the adjustment shows up in employment rather than wages. Whether AI is the main cause is still debated. This is US data, and eJobs doesn’t break its numbers out by seniority, so treat it as the clearest evidence of the mechanism rather than a Romanian measurement.
The second force is the cost of applying. In a Gartner survey of 3,000 job candidates, 4 in 10 said they use AI during the application process, mainly to draft CV and cover letter text, writing samples, and assessment answers (HR Dive, reporting Gartner, August 8, 2025). Once an application takes minutes instead of an evening, the rational move for a candidate in a tight market is to send many more of them.
Put the two together and you get the shape in the chart above: fewer roles posted and far more applications sent.
Why the CV, GitHub and portfolio stopped being signal
A signal is only useful if it’s expensive to fake. Most of what employers used to screen on now costs a model a few minutes to produce.
For years, hiring teams leaned on three artifacts to filter engineers before any conversation:
| Artifact | What it used to prove | Why it no longer proves it |
|---|---|---|
| CV | The candidate can describe their own work clearly and in context | A model writes a clean, keyword-matched CV for any job ad in seconds |
| GitHub contributions and pull requests | Sustained, real engineering effort over time | Commits can be backdated and repositories generated; small AI-written PRs can be sent to open-source projects to dress up a profile |
| Portfolio site or work samples | Taste, finish, and the ability to ship something end to end | The site and the sample can both be generated, and you can’t see who did the work |
The GitHub point isn’t hypothetical. At least one open-source maintainer has written publicly about a rise in AI-assisted pull requests, issues, and security reports, and suspects some are sent to pad contributors’ GitHub profiles rather than help the project. He closed three typo-fix PRs without comment and asked people to “contribute because you care” (Neil Alexander, June 30, 2026).
At the far end, some candidates are fabricated outright. In the same Gartner survey, 6% of candidates admitted to interview fraud, either posing as someone else or having someone else pose as them. Gartner predicts that by 2028, 1 in 4 candidate profiles worldwide could be fake, and its senior research director Jamie Kohn warns that “candidate fraud creates cybersecurity risks that can be far more serious than making a bad hire” (HR Dive, reporting Gartner).
What we see in our own searches: the CVs we receive for engineering roles have converged. The formatting is cleaner, the keyword match to the job ad is closer, and the claims are harder to tell apart. The gap between a strong CV and a strong engineer is the widest we’ve seen in our own searches. Two CVs that look nearly identical on paper can belong to someone who has run production systems for years and someone who has read about them. You find out which in the first fifteen minutes of a technical conversation, not on page one of the CV.
The application DDoS: how employers are responding
For a hiring team, the flood behaves like a denial-of-service attack: legitimate requests get lost in a volume the system was never sized for. Recruiting teams didn’t grow to match. Greenhouse’s 412% rise in applications per recruiter is the clearest sign of that. So employers are improvising, and three responses keep coming up.
1. Peek-a-boo job ads. The ad goes live, collects a few hundred applications in a day or two, and comes down. Sometimes it’s taken down within hours. The team reviews what arrived early and ignores the rest. It caps the pile, but it rewards speed over fit, and the earliest applicants are likely to include the ones running automated tools. The strong engineer who saw the ad on Thursday and meant to apply over the weekend never gets read.
2. Quieter search. More roles never get advertised at all. They’re filled through referrals, former colleagues, and direct approaches to people who aren’t looking. This skips the flood entirely, but it narrows the pool to whoever is already in someone’s network, and it takes real sourcing time. It’s the approach we’ve written about in our guide to headhunting in Romania.
3. AI on the employer side. The volume answer to AI-written applications is AI-read applications. It works for triage, but it brings its own reliability problem. One open-sourced screening agent from HackerRank scored the same CV anywhere from 66 to 99 across 100 runs. Under the EU AI Act, AI used to filter candidates is a high-risk system, so it needs human oversight, not just a threshold (our EU AI Act recruitment guide).
| Response | What it fixes | What it breaks |
|---|---|---|
| Peek-a-boo ads | Caps the volume a team must read | Rewards the fastest applicants, which can favour the automated ones, over the best |
| Quiet, direct search | Bypasses the flood entirely | Takes sourcing time and skill; limited to whoever you can find and reach |
| AI screening | Reads the whole pile in hours | Inconsistent scores; AI judging AI-written CVs; EU AI Act obligations |
None of these fixes the underlying problem. They all reduce volume. None of them gives you back a trustworthy signal.
Where the hiring signal lives now
The signal that survives is the one that’s still expensive to fake: anything that has to happen live, in context, with a person who knows the work.
- Live work, not submitted work. A take-home can be generated. A session where the candidate extends their own code, debugs a failing test, or talks through a design while you change the constraints is much harder to fake, especially paired with the identity check below. In Fabric’s data on 19,368 AI-led interviews, 61% of candidates flagged for AI-assisted cheating still cleared the passing score (our breakdown of AI technical screening), which is why the follow-up conversation matters more than the number.
- Decisions, not descriptions. Ask about a trade-off they made and later regretted, an incident they handled, or a model output they refused to ship. People who did the work answer with specifics, side effects, and names of tools. People who read about it answer with principles.
- Verified history. References from people who saw the work, contribution history spread over years rather than weeks, and real back-and-forth with maintainers in issues and reviews are all much harder to fabricate than a green commit graph.
- Identity checks, early. Given the Gartner fraud numbers, a short camera-on conversation and an ID check before the technical round are basic hygiene, especially for remote roles with access to production systems.
A five-step playbook for hiring when applications are mostly noise
- Write the ad to filter, not to attract. Name the actual work, the stack, the seniority, and the salary range. A precise ad gets fewer applications, and more of them are relevant. A vague one invites a pile of auto-applications.
- Add a knock-out question a model can’t answer for them. One short question about a real decision in their past work, read by a person. It doubles as your first signal of who wrote what.
- Run a fixed application window instead of a peek-a-boo ad. Say up front that the ad closes in five working days, and then read everything that arrived. You cap volume without penalising the people who don’t apply in the first hour.
- Source in parallel. For senior and specialist roles, assume the best candidate never applies. Map the companies doing the work and approach the engineers directly while the ad runs.
- Move the expensive checks earlier. A 20-minute live technical conversation and an identity check before the full loop in our experience filter better than a CV score, and they save your senior engineers from interviewing people who shouldn’t have made it through.
Hiring for a role where the applications are mostly noise? Brief your search and we’ll tell you what the market for it looks like and how we’d reach the people who aren’t applying.
Frequently asked questions
Why are there so many job applications in 2026?
There are fewer openings (and, in US payroll data, notably fewer entry-level roles in AI-exposed jobs), and AI makes each application nearly free to produce. In Romania, eJobs recorded 10.4 million applications in January to September 2026, up 27%, while new job postings slowed. Greenhouse reports applications per recruiter up 412%.
Can employers tell if a CV was written by AI?
Not reliably, and it matters less than it seems. Many candidates now use AI somewhere in the process: Gartner found 4 in 10 do. The useful question isn’t who wrote the CV, it’s whether the person can do the work, and that has to be tested live.
Is GitHub still a good signal when hiring engineers?
It’s weaker than it was. Commit history can be backdated and repositories generated, and at least one open-source maintainer reports AI-assisted contributions he suspects are sent to pad profiles. Long-running history, real code review discussions, and contributions maintainers actually accepted still carry weight. A busy contribution graph alone doesn’t.
How common is candidate fraud?
In a Gartner survey of 3,000 candidates, 6% admitted to interview fraud, either posing as someone else or having someone else pose as them. Gartner predicts that by 2028, 1 in 4 candidate profiles worldwide could be fake.
The bottom line
Nothing in the 2026 numbers suggests the application flood is a temporary spike. Fewer junior roles (clearest in US data), cheap AI-written applications, and recruiting teams that didn’t grow have together broken the old funnel. Posting an ad and reading the pile used to work. Now it costs more time than it saves, and the CV, the GitHub profile, and the portfolio no longer tell you who did the work.
The fix isn’t more volume filtering. It’s moving to signals that are still expensive to fake: live work, real decisions, verified history, and people who can judge them. That last part is where being former engineers matters. AI handles the scale; we handle the judgement.
If you’re hiring engineers in Romania or CEE and the pile has stopped telling you anything, tell us who you’re looking for. We’ll send a curated shortlist, not a CV flood.
Author: Calin Muresan. Last updated: October 20, 2026. Data sources: eJobs (via StartupCafe), Greenhouse (via Fortune), LinkedIn (via The New York Times and eWeek), Stanford Digital Economy Lab, Gartner (via HR Dive), Fabric and HackerRank/Pinggy (via our earlier screening posts), Neil Alexander.