Currently, the U.S. job market has experienced a profound technological transformation. Getting employment in this country no longer depends exclusively on possessing solid work experience or outstanding skills. On the contrary, before a human recruiter reads a professional history, it is very likely that an automated system will decide its fate in just a few seconds. For the Latino community in the United States, this change represents a monumental challenge that definitively alters the rules of the game.
Many Hispanic professionals, both citizens and residents, remember the long hours dedicated to meticulously writing their professional trajectories. Frequently, community advisors at employment agencies hear stories of qualified applicants who send hundreds of applications without receiving a single response. The explanation behind this massive silence is usually not a lack of work capacity, but the existence of an invisible algorithmic barrier. Small details in the document’s structure determine whether an application advances or is permanently lost in a digital database.
Researchers and labor sector specialists warn that filters powered by artificial intelligence (AI) tend to systematically favor documents optimized with these same digital tools. In some cases, the disparity in results is so evident that two candidates with identical profiles receive completely opposite responses. This concerning trend already impacts real hiring processes throughout the country. Therefore, it forces millions of workers to urgently reconsider the way they present their credentials to the corporate market.
What does scientific research reveal about artificial intelligence bias?
The algorithms’ preference for machine-generated texts is not a simple suspicion, but a fact supported by recent scientific data. A rigorous joint academic study from 3 U.S. universities compared candidate profiles with identical qualifications using advanced models such as GPT-4o, DeepSeek-V3, LLaMA, and Qwen. The research, titled AI Self-Preferencing in Algorithmic Hiring, demonstrated that the preference for algorithm-generated applications applies in between 67% and 82% of the evaluated cases. Consequently, software tends to validate with higher scores what was written by another machine.
The most alarming finding of the technical report detailed that the GPT-4o model showed a self-preference bias of up to 81.9%. This percentage remained extremely high even after controlling for critical variables such as text length, vocabulary complexity, and semantic content similarity. According to an analysis published by the prestigious news organization CNN en Español, this technological behavior creates a clear disadvantage for traditional applicants. Candidates who draft their profiles in conventional ways are discarded before a real person evaluates their competencies.
| Artificial Intelligence Model | Level of Self-Preference Bias Detected | Probability of Human Resume Rejection |
| GPT-4o | 81.9% | Very High |
| Alternative Models (LLaMA / Qwen) | Between 67.0% and 82.0% | High |
As a direct result of this phenomenon, an applicant who uses the same technological tool that the evaluating company employs possesses evident numerical advantages. This candidate registers between 23% and 60% more probability of advancing to the in-person interview phase.
