Writing Your Own CV? AI Could Give Other Applications an Advantage

Artificial intelligence may favour CVs written using the same AI model that evaluates them, raising questions about how impartial automated recruitment processes really are.

This issue is highlighted in a column by Pilita Clark in The Irish Times, drawing on the experience of the law firm Addleshaw Goddard and research findings on algorithmic candidate selection.

Why Did Humans and AI Choose Different Candidates?

The firm received 178 applications for an IT position in its human resources department, compared with the few dozen it would previously have expected, prompting it to test an agent based on its internal Microsoft Copilot system.

After employees shortlisted approximately a dozen applications, the AI ranked them according to the job requirements. However, its list of preferred candidates did not match the human assessment.

The agent gave high scores to polished applications containing claims of exceptional achievements despite limited experience. Employees found these claims unreliable and suspected that the applications had been written with AI assistance.

Is Using AI Becoming an Advantage in Its Own Right?

Katherine Rathbone, the firm's head of human resources, offered one possible explanation: since AI proficiency was a job requirement, the agent may have interpreted signs of AI use in an application as evidence of a candidate's skills.

Another possibility is that the model favours text that resembles its own responses.

What Did the Research Show?

Researchers simulated recruitment processes across 24 occupations and found that equally qualified candidates who used the same language model as the evaluator were between 23% and 60% more likely to make the shortlist than candidates with human-written CVs.

The greatest disadvantage for candidates with human-written CVs was observed in business fields such as sales and accounting. The findings describe simulated recruitment processes and do not establish a guaranteed advantage in real-world hiring.

How Can Bias Be Reduced?

The research suggests that bias can be reduced by more than half by directing the model to focus on the content of a CV rather than its origin, and by combining multiple models for evaluation.

At Addleshaw Goddard, the final decision was made by humans, and the successful candidate was someone the agent had not ranked first.

“Human judgement still matters,” Rathbone said, according to The Irish Times.