What should replace the resume in an AI-first hiring process?

Gartner projects that by 2028, about one in four job candidates worldwide could be partly or fully fake, built from stolen identities, AI generated profiles, or some mix of both. The resume was never a perfect signal of ability. Now it's becoming an actively unreliable one, and most hiring teams are still using tools that were never built to catch the difference.

Why the Resume Is Losing Its Value as a Signal

The traditional resume worked, imperfectly, on a basic assumption: that the words on the page reflected genuine effort and honest self-reporting. That assumption is breaking down quickly.

Forty-four percent of job seekers admit to lying somewhere in the hiring process, and 24 percent specifically admit to falsifying their resumes, according to a January 2025 survey. Thirty-three percent misrepresent their educational qualifications outright, including incomplete degrees and inflated grades.

Generative AI has not created this problem. It has industrialized it. A candidate no longer needs to be a skilled writer to produce a polished, tailored, professionally worded application. They need access to a chatbot. The result is an application pipeline where fabrication is not just easier to produce, it is harder to distinguish from legitimate work at the exact stage, initial screening, where hiring decisions are cheapest and fastest to make incorrectly. Only 19 percent of hiring managers say they are confident in their ability to detect a fraudulent applicant at all.

Why Credentials and Self-Reported Skills Were Always a Weaker Signal Than They Looked

The deeper issue predates generative AI, which has simply made an existing weakness impossible to ignore. A resume, a degree listing, and a self-described skill set are all, fundamentally, unverified claims. An applicant tracking system built to rank resumes by keyword density was never designed to confirm that a listed job actually existed, that a claimed skill was ever genuinely demonstrated, or that a credential was earned rather than embellished. That verification work was always meant to happen somewhere later in the process, usually a background check, if it happened at all.

By the time that check runs, a fabricated application has often already cleared the initial screen, impressed a hiring manager, and advanced through multiple interview rounds. Eighty-six percent of recruiters say they've caught or suspected candidate fraud in the past year alone, and it's no longer just a resume problem. Flagged AI assisted cheating during live interviews jumped from 9 percent to 45 percent in just three months in late 2025, as real time AI interview tools became widely accessible. Even the fallback many companies leaned on, take home assessments, saw cheating rates climb from 15 percent to 35 percent over the same six months. Each new verification layer has been met almost immediately by a new way to defeat it.

What Actually Predicts Job Performance, and Why It Rarely Involves a Resume

This is precisely where the research on hiring validity becomes useful, because it points toward a solution that predates the current fraud wave entirely. Work sample tests, structured exercises where a candidate performs an actual piece of representative work, carry a predictive validity for job performance in the range of 0.33 to 0.54, according to Schmidt and Oh's 2016 update to the foundational Schmidt-Hunter meta-analysis. Resume-based screening does not even register among the top five predictors of job performance in that same body of research. Combining a work sample with a structured measure of cognitive ability pushes predictive validity as high as 0.63, among the strongest combinations known in the entire field of personnel selection.

The implication is direct. A resume tells an employer what a candidate claims to have done. A work sample shows an employer what a candidate can actually do, in a format that is considerably harder to fabricate convincingly than a paragraph of text.

Where Structured Assessment Brings Consistency AI-Generated Applications Cannot Fake

Job simulations extend this same logic further. Instead of questioning a candidate about their skills, a simulation places them inside a realistic, job-relevant scenario, a customer complaint to resolve, a dataset to analyze, a piece of code to debug, and evaluates the actual output against a consistent, predefined rubric applied to every candidate equally. This does two things a resume-driven process cannot. It produces evidence rather than a claim, and it standardizes what is being measured, reducing the room for both fabrication and unconscious reviewer bias to shape the outcome.

Where Human Judgement Still Has to Sit

None of this argues for removing people from hiring decisions. It argues for relocating human judgment to the part of the process where it is genuinely irreplaceable: interpreting a candidate's completed work, weighing cultural and team fit, and making the final call on ambiguous cases a structured score cannot fully resolve on its own. What should be minimized is human judgment's role in the early screening stage, where it is currently being asked to evaluate a document that an increasing share of the time was not honestly produced by the person submitting it.

What an Evidence-Based Hiring Process Could Actually Look Like

A more resilient hiring pipeline in 2026 probably starts small: a short intake form instead of an open ended resume. Then a quick, job relevant task before anyone sits through a long interview. Identity and work history get confirmed live, not just taken from a document. And interview time goes to candidates who've already shown they can actually do the work. This sequence does not eliminate the resume outright. It demotes it from primary evidence to background context, replacing it as the central hiring signal with something considerably harder to counterfeit: proof of execution.

The resume was always a proxy for a harder, more direct question. Generative AI just made the resume too easy to fake to still lean on it. The employers who adapt fastest will be the ones who stop needing it to answer the question it was never really good at answering: can this person actually do the job? The ones who get better at spotting a fake one won't be.