Skills-based hiring when the skills in the application are written by AI
A new survey of 159 US talent acquisition leaders shows more applications, more AI-written content and more candidates overstating their skills. What does that mean for a skills-based process that starts with the candidate's own words?
The news in brief
On 3 November 2026, the talent acquisition platform Recruitics announced through GlobeNewswire that its chief technology officer, Austin Anderson, will lead a new Jobs and Hiring special interest group within the Agentic AI Foundation, which is hosted by the Linux Foundation. The group is meant to examine how agentic AI is changing hiring, with an emphasis on the structure of job data, consent and verification. According to the release, AI agents already search job postings, compare roles and submit applications on behalf of candidates, while no agreed framework yet exists for how employers and agents should work together.
The announcement follows a survey Recruitics published on 20 October 2026, also through GlobeNewswire, under the title The State of AI in Hiring for 2027. Its figures are what make the news relevant to anyone running a skills-based process.
What the survey found
The survey was conducted online by Centiment in September 2026 among 159 talent acquisition decision-makers at US organisations with at least 5,000 employees. All respondents were director level or above.
71 percent say the number of applications per role has risen over the past 12 months. 74 percent believe that more than a quarter of applications contain AI-generated or AI-assisted content. And 77 percent report an increase in candidates overstating their skills or experience.
At the same time, 97 percent use AI somewhere in their hiring process. The most common uses are resume screening and ranking (73 percent) and interview scheduling (61 percent). 70 percent say a recruiter loses six or more hours a week to unqualified, duplicate or misrepresenting applicants. Teams using AI in four or more parts of the process lose 10.4 hours per recruiter each week, compared with 7 hours for teams using it in three or fewer.
Asked about AI agents applying on a candidate's behalf, 35 percent would allow them but flag them. 27 percent would block them outright. 18 percent would build an experience specifically for agents, and 18 percent would treat them the same as human applicants.
Why this matters for skills-based hiring
Skills-based hiring rests on a simple idea. Instead of sorting candidates by degrees and job titles, you look at what they can actually do. In practice, many skills-based processes still start with the candidate's own words. A CV with a list of skills, a questionnaire about experience or a free-text field, which is then read and ranked by an AI tool.
That is exactly the link the survey calls into question. When three in four leaders believe a large share of applications are written with AI, and when 77 percent see more candidates overstating their skills, a self-reported skill becomes a weak signal. An AI tool that ranks CVs is then ranking text that another AI tool produced.
It is striking that 85 percent of leaders are nevertheless somewhat or completely confident that their process identifies real, qualified candidates before recruiter time is spent. Those who are completely confident report the highest loss of time, at 9.7 hours a week. The release does not explain the pattern, but it is worth noting when reviewing your own screening.
Which signals leaders trust more
The survey also asked which signals leaders trust more than they did two years ago. Live interviews were cited most often, followed by internal referrals and verified employment data. The cover letter was the only signal to lose trust, because AI-assisted content makes written material harder to read as a signal.
Over the next 12 months, 70 percent plan to increase spending on AI screening and matching, and 53 percent plan to spend more on identity and fraud verification.
From an assessment perspective, this points in a clear direction. The signals gaining trust are those where the candidate does something under observation, or where a third party confirms the information. A standardised test, a work sample or a structured interview belongs to the same category. What matters is not whether a tool is labelled skills-based, but whether the skill has been measured or merely claimed.
Three questions for your own setup
First, which steps in the process rely on the candidate's own descriptions, and which rely on a measurement? Second, where does the measurement sit, before or after the AI-based ranking of applications? Third, how do you make sure that the person taking the test is the same person who applied? That last point is precisely the identity and verification question the new interest group will work on.
Caveats
The survey was commissioned by a vendor that sells solutions to the problem it describes. The sample is small, at 159 respondents, and covers only large US companies. Percentages are rounded. The release contains no data on the accuracy or bias of the AI screening tools respondents use, and it does not describe how identity is verified. Read the figures as a temperature check among large US employers, not as a picture of Danish conditions.
Kilde: https://www.globenewswire.com/news-release/2026/11/03/3394323/0/en/recruitics-cto-austin-anderson-to-lead-new-agentic-ai-foundation-group-on-jobs-and-hiring.html and https://www.globenewswire.com/news-release/2026/10/20/3384492/0/en/more-than-1-in-4-talent-acquisition-leaders-would-block-ai-agents-from-applying-to-jobs-recruitics-survey-finds.html