Two equally qualified applicants should not be ranked differently because they chose different writing assistants. Yet this can happen when candidates use AI for applications and employers use AI to screen them. HR may check a resume for skills while missing the screening system’s preference for a particular writing style.
The Writing Model Can Change the Ranking
A September 2026 controlled AI hiring simulation tested eight models that wrote applications against six that evaluated them. The simulation covered ten jobs. They found that evaluators favored output from some writing models over others. In the worst pairing, roughly two of every 25 qualified applicants fell out of consideration because of model mismatch rather than credentials. These were simulated decisions, not actual employer rejection rates.
The pattern extends further. In separate University of Maryland hiring research, investigators assessed more than 2,200 resumes and simulated selection across 24 occupations. Models often preferred resumes produced by themselves. Candidates whose writing tool matched the evaluator were more likely to be shortlisted than equally qualified applicants using human-written materials. A consistent algorithm can still make biased decisions.
A Resume Score Is Not a Skills Assessment
A July 2026 comparison of automated skill-extraction methods found substantial differences in what systems identified as skills in the same resumes. Systems produced different skill descriptions. Candidates may describe equivalent work in different terms, and scores can reflect vocabulary rather than ability.
HR should stop treating software rankings as self-explanatory. Ask vendors which information determines a score, how the system handles unconventional experience, and whether changing harmless wording changes a candidate’s position. Review a sample of rejected candidates as well as those that were advanced. Just accepting the machine’s shortlist hides who was excluded.
Test Equivalent Candidates Before Trusting the Shortlist
Start with matched application profiles showing the same verified qualifications. Vary only the writing model, prose style, or order of equivalent skills. Compare selection rates and investigate changes that cannot be explained by job requirements. Repeat across roles and system updates. Record when recruiters overturn recommendations.
The EEOC’s guidance on AI hiring tools and disability discrimination adds another reason to preserve alternative assessments and accommodation processes. A system can exclude qualified candidates even when nobody deliberately designs it to discriminate. The recent research identifies a different risk, preference for machine-generated prose, that deserves its own tests alongside established fairness checks.
Successful workplace AI adoption requires HR to measure whether technology improves decisions, not merely how quickly it produces them. The crucial question is whether every qualified applicant receives a fair evaluation regardless of how the application was written.













