Jodi Beggs experienced a setback in her job application process when her curriculum vitae was penalised for exceeding two pages. The use of her middle name’s initial in one document but not in another also contributed to a lower score. Interestingly, had she replaced “per cent” with the symbol “%”, her score would have improved. Beggs, a data scientist seeking employment, received this feedback from an online tool designed to compare job descriptions with candidates’ documents to assess their compatibility. One of the more renowned platforms is Jobscan, although Beggs did not specify which one she used. She remarked, “If my job is to please robots, then, despite my reservations about AI-generated content, I will do so.”
Jobscan and similar applications operate on the premise that hiring committees employ AI to evaluate and rank candidates within their Applicant Tracking Systems (ATS). It is a common belief that only 10-20% of the highest-rated candidates are considered, leaving the remaining 80% in the dark. To make it onto the shortlist, candidates must tailor their CVs and cover letters to include the relevant keywords that the ATS seeks.
Understanding Applicant Tracking Systems
At its core, an ATS collects job applications, assists hiring committees in reviewing them, and tracks candidates throughout the interview and hiring process. Some systems offer additional features, such as new employee onboarding and automated selection interviews via AI.
Daniel Chait, CEO of ATS firm Greenhouse, claims there are numerous myths surrounding these tools, especially among job seekers. Not all ATS are alike; technology evolves rapidly, and an AI integrated into one ATS may function differently today than it will tomorrow. Whether an ATS employs AI depends on the specific product and the features for which the team has paid and is using.
Regardless, as long as candidates believe that AI is a factor, they will continue to attempt to outsmart it using their own AI tools. Relying on a generative tool to optimise one’s profile for an ATS is a precarious gamble. A recruiter demonstrated that despite receiving poor Jobscan scores, she managed to secure a dozen interviews and an employment offer. Jobscan subscriptions range from $30 to $50 per month, and the company has little incentive to see candidates exit the job market.
The current economic climate, characterised by low activity and minimal hiring, exacerbates these challenges. Job vacancies are scarce, and job scams are so prevalent that some states are considering new legislation to combat them. Both job seekers and employers report a crisis of trust; candidates invest considerable time in their searches only to receive no feedback, while employers contend with hundreds of almost identical applications. Consequently, many submit them through their ATS AI evaluation systems in a bid to differentiate them quickly.
The Vicious Cycle of AI in Recruitment
Chait posits that we find ourselves in a vicious cycle with AI: “We are in a complicated situation where both parties have a problem and try to solve it with AI, but they end up exacerbating it. The more they rely on it, the more its use increases, without anyone truly benefiting. With each repetition of this cycle, the situation worsens.”
In discussions with numerous recruiters, HR professionals, and small business owners who handle their own hiring, a division emerged regarding the use of automated candidate ranking. This division does not correlate with the size of the organisation or the number of applications received; instead, it seems determined by the company’s philosophy and culture.
Kim Jones, VP of HR at Toshiba, stated, “We have people review all applications.” She does not mind candidates using AI to refine their documents, but “it won’t actually help them get past the applicant tracking system.” The selection ultimately hinges on job requirements, salary expectations, and other factors, such as whether it is a rehire.
Jones has noted an undesirable use of AI during the interview phase: “You hear a pause, perhaps the sound of someone typing, and then they give a ‘perfect answer.’
Real-World Experiences with AI in Hiring
Doist, a smaller company that operates entirely remotely and hires internationally, places greater trust in human judgement over technology in its hiring processes. Nadia Vatalidis, the company’s HR director, shared that her team conducted an experiment to determine whether ATS rankings genuinely aided in the hiring process. They took positions that had already been filled and uploaded the job descriptions along with the documents of all candidates who had participated in those processes. “We wanted to see what would happen if AI were to pre-select the same group of people and whether those we interviewed would also appear among the top-ranked candidates. In two cases, the person we hired did not make it onto the shortlist,” Vatalidis explained. While there were overlaps in terms of who was invited for interviews, the new hires, who performed exceptionally well after six months, would not have passed the selection process.
James Jacobsen, a designer, embarked on his job search five months ago, dedicating as much time to it as a full-time job. Acknowledging that AI cannot match his creativity, he experimented with it and recognised its potential for certain tasks. He utilised Claude and ChatGPT to refine his application documents to fit job descriptions but found limited success.
Innovative Approaches to Job Searching
Rather than concentrating solely on his application documents, Jacobsen began employing Claude to streamline and track his job search from start to finish. He instructed the AI to review job listings, analyse descriptions, and log them. He devised a scoring system based on job type, seniority, and salary expectations, which varied depending on whether the roles were in-person, hybrid, or remote. His AI assistant highlighted the best opportunities for application. When he rejected a position, he would add a note so that if the same job were reposted weeks later, the AI could remind him of his initial disinterest. This approach allowed him to spend less time searching while uncovering higher-quality positions. Although he received some responses, he did not secure any offers.
In essence, Jacobsen created the antithesis of an ATS: a job seeker tracking system. He did not stop there; he also requested evaluations of his portfolio from Claude and ChatGPT, resulting in a six-hour review. Two days later, he received a call from a potential employer, but again, it did not lead to an offer.
Chait suggests that candidates like Jacobsen are left wondering how they can further enhance their efforts: “Continuing to do the same is not the solution. This is the first time I can recall both sides feeling dissatisfied. It’s simply not working.” For job seekers, he advises against randomly sending applications.
Chait further recommends investing time in researching companies of interest, even if they are not the most obvious choices. Kim Jones noted that she rarely sees cover letters anymore and asserts that candidates stand out when they do submit one. Networking is another undervalued skill in today’s job market.
Chait concludes with a reassuring message for job seekers: “It’s not your fault; it’s the system, and it’s a mess.”