That is the question that will arise in the future. Everyone, from mothers and working professionals to students and everyone in between, uses AI. Today, it is more present than ever. So let me ask the question again: If AI creates the CV and also scans the CV, who is actually in control at the end of the day?

To anticipate the question above, I do not want to generalize its impact yet. However, as the findings so far show, the trend continues to grow. A Greenhouse study based on candidates from the United Kingdom, Ireland, and Germany found that a remarkable 78% of candidates used AI to tailor their CVs or job applications to specific positions. To manage the growing volume of applications, recruiters are also turning to AI; otherwise, they would be overwhelmed by an ever-expanding pool of applications. Within this dichotomy, clarity and transparency remain blurred and obscured.

The candidate asks: “Did a recruiter actually see my CV?” after being rejected.

The recruiter asks: “Are these people really who they claim to be?”

This downward spiral only continues; there is no escape. Galdin and Silbert’s Making Talk Cheap: Generative AI and Labor Market Signaling shows that employers on Freelancer.com were more likely to reward applicants whose applications were strongly tailored to the advertised position. Sending a personalized application used to mean something. Today, it is the absolute minimum, and even then, there is no guarantee.

Suppose a candidate writes their application entirely from scratch. They find the job advertisement appealing. They put in the effort and write as well as they can. Historically, this would have provided insight into who that person was: a personalized cover letter, their education, their biography, their professional experience, what they had done before applying. It reflected the person and provided a sense of who they were. Translated into today’s job market? There is barely any difference between a carefully self-written application and a highly polished, AI-generated one.

The screening process had to keep up. Ghost rejections are not uncommon. The 2025 Greenhouse survey found that 56% of UK job seekers and 66% of Irish job seekers reported being ghosted after an interview. Being rejected is one thing; not knowing whether you were rejected by a human, a rule, a ranking system, or simply never reviewed is another.

Before AI, we had ATS: Applicant Tracking Systems. Filtering by keywords, qualifications, and company-defined criteria has existed for a long time. After AI, companies continue to use the same systems. However, the misconception of a magical machine that simply rejects CVs should be cleared up immediately. Automation is not the same as artificial intelligence. ATS parameters can be configured and defined by humans to filter candidates without an AI model independently evaluating them. Modern ATS platforms may now include AI capabilities, but the terms are not interchangeable.

After AI, this raises a very interesting question. LLMs are not magical beings living in the cloud. No — although that description might draw parallels with an ethereal being, similar to a divine entity, but that is a topic for another day. LLMs are trained on data, and no dataset is truly perfect. In a study by Iso et al., Evaluating Bias in LLMs for Job-Resume Matching: Gender, Race, and Education, NAACL 2025, researchers processed 40 job descriptions from different professions and systematically evaluated 12 LLMs. Each model assessed how well a CV matched a job, while the researchers systematically changed characteristics such as gender, ethnicity, and educational background.

The results? They are nuanced. Newer models showed lower levels of bias for certain characteristics, particularly ethnicity and gender. However, bias related to educational background remains. The study is limited to an English-language and US context, but it still raises an interesting question. If LLMs can exhibit bias, how do we know what they consider important? What have they learned to consider important?

There is no denying that AI is part of our world. To answer the question, “Did a recruiter see my CV?”, I refer to Vaishampayan et al.’s Human and LLM-Based Resume Matching, Findings of NAACL 2025, in which 736 real CVs were compared. It showed only limited agreement between human and AI-based assessments. Humans and AI are not interchangeable, even when evaluating the same candidate against the same requirements. The results and reasoning can differ. If there is only limited agreement, where should the human stand in the process?

An AI-powered recruitment process could look like this:

CV → ATS → Filtering → AI Matching/Ranking → Recruiter

Imagine that 76 applications are received and, after the process and AI ranking, only 12 are made visible. The recruiter may choose to review these highlighted candidates. The question of who makes the final decision is not simply whether it was AI or a human; the question is whether my CV ever reached a human at all.

So where do we stand in 2026?

Honestly, it is difficult to say. The scale is enormous, and implementation differs from company to company, country to country, and culture to culture. We are entering a new era. LLMs are everywhere, and we have to adapt. The European Commission classifies AI tools for employment, worker management and access to self-employment (e.g. CV-sorting software in recruitment) as high-risk AI use cases under the EU AI Act. The regulatory framework requires high-risk systems to provide traceability, documentation, accuracy, and appropriate human oversight, with these requirements applying from 2 December 2027.

Therefore, I come to the following conclusion: The more we automate a fundamentally human process, the more important the boundary becomes. Where does human interaction begin, and where should AI end?