Silent Deindustrialization: the Missing Successor Ecosystem
Bernhard Liebl
7 min. read Germany loses economic substance every year without anyone accounting for it. Around 114.000 ...
Read Article
8 min read
Generative AI makes it easy for candidates to produce flawless résumés and to have video interviews delivered in real time. As a result, the early stage of recruiting loses its significance. The consequence is not an HR nuisance but a strategic risk: companies are increasingly selecting candidates who master the selection process best, rather than those who can do the job best.
Key Takeaways
- The funnel breaks at both ends. Resume and first interview, the two classic early filters, have both become manipulable. Gartner expects that by 2028 every fourth candidate profile will be partially or wholly fabricated.
- Polish has become worthless. In an analysis of 6.380 first interviews, suspicion rates for junior developer roles were close to 60 percent. What remains is a thin signal: judgment under pressure.
- The answer is interview design, not surveillance. Adaptive, justification-based conversations outperform résumé filters and are legally more defensible in Germany than the detection tools that vendors are currently selling.
Why the résumé has lost its signal
For decades, recruiting favored candidates who could present a flawless résumé and provide structured answers. Both are now practically free and infinitely scalable. Applicants generate in minutes tailored, keyword‑optimized documents. Worse: the screening tools of HR departments are equally compromised. A 2025 study found a self‑preference effect in AI screeners. Résumés written in the style of the evaluating model were placed on the shortlist with 23 to 60 percent higher probability than equally qualified applications that someone had written themselves.
Thus, the résumé no longer measures suitability, but prompt quality and the question of which AI model the candidate happened to use. Some recruiters take the consequence and stop posting jobs publicly altogether. This reduces noise but simultaneously dismantles diversity: open applications are one of the few ways for candidates without an elite résumé or network to become visible at all.
The first interview is no longer a safe harbor
The remote first interview was long considered the incorruptible test. Real‑time assistance tools now feed candidates answers, bypassing older detection methods. In an operational study, 6.380 recorded first interviews from eight tech companies were analyzed. The result varies sharply by role type: for low‑technical positions the suspicion rate was under 10 percent, for mid‑technical roles close to 40, and for entry‑level software developers almost 60 percent.
The second observation is decisive: not all interview formats are equally vulnerable. The classic behavior‑based interview, which asks predictable questions, can be steered in real time. An adaptive, justification‑based interview works differently. The interviewer probes further, demands defense of trade‑offs, and throws unknown scenarios into the conversation midway. Because the discussion constantly changes direction, the AI assistance lags behind, and the gap between rehearsed polish and genuine thinking becomes visible.
The figure that shifts the calculation
Roughly one in four applicants. That many candidates exceeded the expectations that their résumé had set in unstructured, justification‑based interviews. Anyone who filters only by résumé not only lets the wrong people in, but also screens out the right ones. Gallup estimates the cost of a bad hire at one‑and‑a‑half to two times an annual salary.
The DACH factor: Why the detection solution doesn’t work here
Vendors mainly sell monitoring as the answer: gaze tracking, latency analysis, voice verification. In Germany this approach collides with three realities that a US whitepaper overlooks. First, the deployment of such systems is subject to co‑determination. The works council has genuine say when technology that monitors behavior and performance is introduced. Second, the EU AI Act classifies AI for personnel selection as a high‑risk application, with documentation, transparency and oversight obligations. Third, data‑protection law and the AGG (General Equal Treatment Act) set tight limits on biometric evaluation and on any form of discrimination through automated pre‑selection.
The consultant‑neutral assessment therefore reads: the arms race of detection versus deception sold in the US is not a viable primary route under DACH law. The more robust and legally sound lever is interview design. A conversation that targets judgment rather than rehearsed answers needs no biometric monitoring to be meaningful.
The decision: separating authenticity from competence
The operational consequence has two tiers. First, treat the early round as a genuine assessment step, not as an administrative formality. Instead of predictable questions, dynamic friction points belong in the interview: sudden constraints, a shift in project scope, a request to defend a counter‑intuitive trade‑off. Second, align the later rounds with the actual work. If AI is part of the job, the interview should allow its use and test whether the candidate spots faulty AI outputs, uncovers hallucinations, and corrects the logic. That is precisely hard for an assistance tool to fake.
It’s not about HR romance, but about capital allocation. When the early filters fail, the company wastes senior executives’ valuable time at later stages on candidates who were carried through the first round by an AI proxy. Whoever measures judgment early and honestly conserves the scarcest resource in the selection process.
The First Step in the Next 90 Days
No big-bang overhaul of recruitment-just a pilot for one critical role. Pick a position where bad hires are costly, and shift its initial interview from behavior-based to adaptive, evidence-driven. Measure the impact with two metrics: How many candidates with unremarkable CVs shine in the conversation, and how stable is the hiring accuracy after six months compared to the old process? At the same time, clarify with the works council which assessment methods are acceptable before any tools are introduced. When competence can be faked at will, judgment becomes the rarest hiring signal. Organizations that learn to measure it build stronger teams.
Frequently Asked Questions
Why isn’t a CV enough as a pre-screening filter anymore?
Because generative AI can produce flawless, keyword-optimized applications in minutes-and even the AI screeners reviewing them are compromised. Studies reveal a self-preference effect: CVs styled like the reviewing model are far more likely to make the shortlist. That means CVs now measure prompt quality, not suitability.
Which interview format is more resistant to AI assistance?
The adaptive, evidence-based conversation. The interviewer digs deeper, demands justifications for trade-offs, and introduces unfamiliar scenarios. Since the discussion constantly shifts, real-time AI assistance lags behind, exposing the gap between polish and genuine thinking.
Are AI monitoring tools permissible in German recruitment?
Only to a limited extent. The works council has co-determination rights over behavior and performance monitoring technology, the EU AI Act classifies selection AI as high-risk, and data protection laws like the AGG restrict biometric analysis. Interview design is the more legally sound lever than detection.
Image source: AI-generated (June 2026)