'Self-preference' | The new recruitment bias: How AI hiring tools may favour AI-written CVs

For years, the narrative around AI in recruitment has been relatively similar. Employers worry about candidates using tools like ChatGPT to polish their applications, fearing it reduces authenticity and makes it harder to assess genuine skills. Jobseekers, meanwhile, have been warned that relying too heavily on AI could backfire if recruiters spot formulaic and polished language or, infamously, the notorious em-dash. 

At the same time, organisations have been rapidly embedding AI into their own hiring processes. From CV screening to candidate ranking, large language models (LLMs) are increasingly being used to sift through applications at scale, promising efficiency and consistency in what has traditionally been a time-intensive process.

This has created a quiet tension at the heart of modern hiring. The assumption underpinning much of the debate is that a human recruiter sits on the other side of the process, capable of spotting when AI has been used, or simply running a CV through AI detection software. But what happens when that assumption no longer holds true?

Recent research suggests the answer may fundamentally reshape how employers think about fairness in hiring.