Artificial intelligence is now working both sides of the hiring table. Employers use it to parse, rank, and deselect applications at volume. Candidates use it to rewrite bullet points, mirror the job description, and generate a tailored cover letter in under a minute. What arrives in the applicant tracking system is a stack of well-written, well-targeted, algorithm-friendly documents that are nearly impossible to tell apart. Hiring teams are reading more applications than ever and learning less from each one.
That is not a recruiting problem. It is a measurement problem. When the signal disappears from the application, the only way to recover it is to measure the candidate directly, with an instrument that has a published relationship to job performance. Everything else is sorting.
The Résumé Stopped Being a Signal and Became a Formatting Exercise
For most of the last forty years, the résumé worked as a proxy. It was not a precise one, but a clear, specific, well-organized application carried information. Writing one well took effort, and effort correlated loosely with ability, diligence, and interest in the role.
Generative AI removed the effort. Economists Anaïs Galdin of Dartmouth and Jesse Silbert of Princeton modeled exactly what happens when it does. Using roughly 960,000 applications from 92,000 workers on Freelancer.com, their November 2025 working paper “Making Talk Cheap” simulated an equilibrium in which writing quality no longer carries information about ability. In that scenario, workers in the top quintile of the ability distribution are hired 19% less often than before, and workers in the bottom quintile are hired 14% more often. The study is a structural model built on freelance platform data rather than a corporate hiring audit, and it should be read in that context. The direction, however, is consistent with what every talent acquisition leader is reporting, the best candidates are getting harder to see, not easier.
Volume made it worse. The World Economic Forum reported in August 2026 that applications on LinkedIn rose 45.5% in late 2024 while job postings fell 10.6%. A 2025 NACE survey found that students using AI submitted roughly 60 applications on average, double the overall average of 30. LinkedIn’s own 2026 research, drawn from 19,113 consumers and 6,554 HR professionals, found that the number of applicants per open role has doubled since spring 2022, and that two-thirds of recruiters say finding qualified talent has become harder, not easier.
More applications. Less differentiation. Worse odds of finding the person you are actually looking for.
The Screen Got Faster. The Decision Did Not Get Better.
The reflex has been to answer volume with more automation. Applicant tracking system adoption nearly doubled between 2024 and 2025, from 26% of organizations to 43%, according to figures cited by the World Economic Forum. Faster screening, more filters, more keyword logic.
Consider what that actually buys. A Greenhouse survey of 4,100 recruiters in 2025 found that 34% spend up to half their working week filtering spam applications, and 91% had caught candidates being dishonest during the hiring process. Automated screening has not returned that week. It has accelerated the sorting of documents that no longer distinguish anyone.
It has also inherited every bias in the data it learned from. A University of Washington study of three large language models across more than 500 résumé comparisons found the models favored white-associated names 85.1% of the time, compared with 8.6% for Black-associated names. An algorithm trained on who you hired before will reliably recommend more people who look like who you hired before. That is not intelligence. That is precedent, running faster.
Candidates have noticed. Greenhouse found that only 8% of job seekers believe AI screening makes hiring fairer, and half report lower trust in application processes than a year earlier.
Here is the uncomfortable truth, and it belongs to our profession before it belongs to any vendor. We spent two decades telling candidates to tailor every résumé to the job description, load it with the right keywords, and quantify every accomplishment. They now have a tool that does all three perfectly, for free, in seconds. We asked for optimization. We received it. The screening method we built cannot survive its own success.
Predictive Validity Is Not a Marketing Word. It Is a Number.
Every selection method makes an implicit claim: people who score higher on this will perform better on the job. Predictive validity is the measured strength of that claim, expressed as a correlation coefficient. It can be calculated, published, challenged, and replicated. Most of what happens in hiring has never been subjected to that test.
The most current authority is the 2022 meta-analytic reassessment by Sackett, Zhang, Berry, and Lievens, published in the Journal of Applied Psychology, which corrected decades of overstated estimates. Their revised operational validities rank structured interviews at .42, job knowledge tests at .40, empirically keyed biodata at .38, work sample tests at .33, and general cognitive ability at .31.
Now ask the questions that matter for your process. What is the published validity coefficient of your résumé screen? What is the coefficient of the unstructured conversation your hiring manager runs after it? What is the coefficient of the AI ranking engine sitting between them? Who validated any of it against actual performance in your roles? What would you find if you ran the analysis this quarter?
Assessment is the part of the hiring process that can answer those questions with a number. That is the entire argument.
This is not personality typing, and it is not a soft add-on to a hard process. A properly built assessment is an instrument with a known error rate, established reliability, and documented fairness analysis. LAK Goup publishes how we establish validity and reliability for exactly that reason: test-retest and internal consistency reliability, face, content, construct, and criterion validity, and equal opportunity testing that includes Differential Item Functioning analysis and regression modeling across demographic groups. Instruments that cannot show that work are not measurement. They are opinion with a scoring rubric.
What a Candidate Cannot Have AI Write for Them
The reason assessment cuts through AI-saturated applicant pools is structural, not clever. A résumé is a document produced outside your process, at the candidate’s convenience, with unlimited time and unlimited assistance. A validated assessment is an event that happens inside your process, under controlled conditions, measuring constructs the candidate cannot outsource.
The LAK Group Talassure® suite measures three things a polished application cannot demonstrate. Cognitive ability, which predicts how quickly a person learns the work and how well they reason through ambiguity. Core behavioral traits, which predict how they will operate under pressure, with a team, and with a customer. Occupational interests, which predict whether they will still want the job in eighteen months.
Talassure®MX then does the thing that keyword matching only pretends to do. It compares a candidate against a Success Pattern built from the traits of your top performers in that specific role, and returns a job match percentage. Not a match to the job posting. A match to the people who are actually good at the job. The suite runs the same logic by role family, including sales selection, leadership potential, entry-level screening, 360 development feedback, and an automotive-specific instrument for dealership sales and service.
Gartner reached the same conclusion from the other direction. In October 2025 the firm predicted that by 2027, 75% of hiring processes will include certifications and tests of workplace AI proficiency, and named the underlying requirement plainly. As Gartner’s Jamie Kohn put it, “Maintaining candidate quality requires organizations to assess candidates’ true abilities without GenAI while also integrating AI into assessments for roles that have to use it on the job.”
Assess the person, not the paperwork. Here some of the recommendations LAK Group give our customers as we work with them to improve quality or hire and retention metrics.
- What This Looks Like in Practice: Do not start by buying a test. Start by defining what good looks like.
- Build the Success Profile From Your Own Top Performers: Assess the people currently succeeding in the role and identify the traits that separate them from the people who are not. That benchmark is specific to your business and cannot be copied off a job board.
- Move Assessment Earlier: Most organizations assess at the finalist stage, which means the résumé screen has already thrown away the people they most needed to see. If the application no longer differentiates, it should not be the gate that eliminates.
- Structure the Interview Off the Assessment Output: A structured interview carries the highest validity of any common method at .42. An assessment report tells the interviewer exactly which two or three areas to probe, which is what converts a conversation into a measurement.
- Validate Locally, and Keep Validating: Run the correlation between assessment scores and actual performance in your roles after twelve months. If it does not hold, change the instrument or the profile.
- Price the Process You Run Today: Take your annual hires in one critical role, your twelve-month attrition rate in that role, and your fully loaded replacement cost. Multiply it out. Then compare that number to what objective measurement would cost across the same population.
Run the math on your own numbers before you accept another year of the status quo. In most organizations the calculation is not close.
The Signal Did Not Disappear. It Moved.
AI did not make candidates less distinguishable from one another. It made the document itself less distinguishable. The differences in reasoning ability, behavioral fit, and genuine interest are exactly where they have always been, sitting in the candidate. What changed is that the cheap proxy stopped pointing at them.
Organizations that respond by adding more filters will keep sorting identical paper faster. Organizations that respond by measuring the person will be selecting on the things that actually predict performance while their competitors are still ranking keywords.
Hiring is not a guessing game. It is a science, and the science does have and should have a number attached to it.
See What Objective Measurement Reveals About Your Next Hire
At LAK Group, we have built our Talassure® assessments that are workplace-focused and psychometrically defensible, developed under psychometrician Dr. Ron Trego and reviewed continuously against real outcomes. Our clients report results we track rather than claim: more than $2.24 million in added annual revenue, 9% lower turnover, and a 7.8x increase in monthly performance. Those figures come from our own client base and should be read in that context. Bring your own numbers and test them against ours.
If you are filling a critical role right now and the applicant pool all reads the same, stop guessing which one is real. Request a complimentary talent audit and we will assess your next new hire at no cost, so you can see precisely what objective data tells you that the résumé cannot. If you would rather start with the evidence, review our validity and reliability methodology or our research library first.
The applications will keep getting better. Your ability to tell them apart will not, unless you change what you are measuring.
References
Galdin, A. and Silbert, J.: Making Talk Cheap: Generative AI and Labor Market Signaling (November 2025). https://jesse-silbert.github.io/website/silbert_jmp.pdf
World Economic Forum: How ‘Trace Hiring’ Can Reclaim Human Authenticity in the Age of AI (August 2026). https://www.weforum.org/stories/jobs-and-the-future-of-work/how-trace-hiring-can-reclaim-human-authenticity-in-the-age-of-ai/
LinkedIn 2026 Workforce Research, reported in Fortune: How Hard It Is to Get Hired Right Now (January 2026). https://fortune.com/2026/01/07/how-hard-is-it-to-get-hired-mismatch-recruiters-job-seekers-linkedin-ai-skills/
Greenhouse: 2025 Recruiter and Job Seeker Survey, cited in World Economic Forum and Fortune reporting
National Association of Colleges and Employers (NACE): 2025 Student Survey on AI and Job Applications, cited in World Economic Forum reporting
University of Washington: Study of Large Language Model Résumé Screening and Name-Associated Bias, cited in World Economic Forum reporting
Sackett, P., Zhang, C., Berry, C., and Lievens, F.: Revisiting Meta-Analytic Estimates of Validity in Personnel Selection, Journal of Applied Psychology (2022); summarized by the Society for Industrial and Organizational Psychology. https://www.siop.org/tip-article/is-cognitive-ability-the-best-predictor-of-job-performance-new-research-says-its-time-to-think-again/
Gartner: AI Revolution and Cost Pressures Drive the Top Four Trends for Talent Acquisition in 2026 (October 2025). https://www.gartner.com/en/newsroom/press-releases/2025-10-07-gartner-says-ai-revolution-and-cost-pressures-are-two-forces-driving-the-top-four-trends-for-talent-acquisition-in-2026
Talexes: Assessment Validity and Reliability. https://talexes.com/validity/