LLMs show bias toward their own generated resumes over human-written ones

AI is now picking its own CVs over human ones, and if you're using AI tools to screen job applications, you've just walked into a hall of mirrors.

New research shows that when large language models evaluate resumes, they consistently favour CVs generated by AI over those written by humans or competing AI systems. This isn't about quality, it's about AI recognising its own patterns and rewarding them, even when human recruiters can't tell the difference.

**The Mirror Effect Takes Hold**

This research confirms what we've suspected: AI systems have developed a preference for their own output. When given a stack of resumes to evaluate, LLMs consistently ranked AI-generated CVs higher, regardless of the actual qualifications or experience listed. The models weren't explicitly told which resumes were AI-generated, they simply recognised the linguistic patterns they've been trained to produce.

The implications stretch far beyond hiring. If you're using AI to evaluate anything, from grant applications to marketing copy to business proposals, the system may be unconsciously favouring content that follows AI-generated patterns over genuinely human work.

**What This Means If You Run a Business**

If you're a small business owner using AI screening tools to handle job applications, you're potentially creating a feedback loop that rewards candidates who know how to game AI systems rather than those with the best qualifications. The person who prompts ChatGPT to write their CV might get through your initial screening, while the genuinely experienced candidate who writes their own gets filtered out.

For freelancers, this cuts both ways. If you're applying for work through platforms that use AI screening, understanding how to structure your proposals and profiles to appeal to these systems becomes as important as your actual skills. We're moving toward a world where knowing how AI "thinks" becomes a meta-skill for getting work.

The broader concern is authenticity erosion. When AI systems favour AI-generated content, we risk creating environments where genuine human voice and experience get systematically filtered out. Your most creative, unconventional candidates, often your best hires, might be the first to get eliminated by systems that prefer predictable, AI-friendly formatting.

**What To Do About It**

1. 1.Audit your screening process immediately. If you're using any AI-powered recruitment tools, test them with a mix of obviously AI-generated and clearly human-written applications. See if patterns emerge in how they're being ranked.

1. 1.Implement human oversight at the right stage. Don't let AI make the final cut on applications. Use it for initial volume reduction if you must, but ensure humans review the top candidates before making decisions. The AI can handle the obvious nos, but shouldn't pick your shortlist.

1. 1.Focus on work samples over polished applications. Ask candidates to complete a small, relevant task or provide portfolio examples. AI can write a perfect cover letter, but it can't fake three years of actual design work or coding experience.

1. 1.Train your team to spot AI-generated content. Learn the telltale signs: overly formal language, perfect grammar with no personality, buzzword-heavy descriptions that say nothing specific. Human applications have quirks, personality, and occasional typos, these might now be features, not bugs.

1. 1.Consider going analog for key hires. For crucial positions, skip the AI screening entirely. Yes, it means more manual work, but you'll avoid the risk of filtering out your best candidates because they didn't optimise their CV for an algorithm.

The machines are starting to favour their own kind. Make sure your hiring process still has room for actual humans.