Prompt Injection in Automated Résumé Screening with Large Language Models - Single and Multi-Injection Settings

Sep 13, 2026

Uploaded by Franzy

Large language models increasingly rank job applicants, creating incentives for candidates to manipulate algorithmic hiring. This study examines prompt injection in automated résumé screening, where subtle self-promotional text is crafted to steer LLM evaluations without adding genuine qualifications. Controlled experiments show reliable rank improvements under homogeneous résumé quality and limited injection, but diminishing returns and saturation when more candidates inject. With heterogeneous quality, injection is weaker on average yet can occasionally elevate lower-quality résumés, raising fairness concerns.

LanguageEnglish

CategoryGeneral

FormatPDF

Pages12

Reading Time4 min

AccessFree to download

Introduction

Research Question

Experimental Setup

Injection Types and LLM Models

Results

Prompt Injection in Automated Résumé Screening with Large Language Models: Single and Multi-Injection Settings Preet Baxi†, Jiannan Xu‡, Jane Yi Jiang§ , Stefanus Jasin¶†Department of Physics, University of Michigan ‡Robert H. Smith School of Business, University of Maryland §Fisher College of Business, The Ohio State University ¶ Stephen M. Ross School of Business, University of Michigan {preetb, [sjasin}@umich.edu](sjasin}@umich.edu) [jiannan@umd.edu](jiannan@umd.edu) [jiang.3186@osu.edu](jiang.3186@osu.edu) arXiv :2606 .27287v 1 [ cs .AI] 25 Jun 2026 Abstract Large language models (LLMs) are increasingly used to screen and rank job applicants, creating incentives for candidates to strategically manipulate algorithmic hiring systems. We study prompt injection in automated résumé screening, defined as subtle self-promotional text that introduces no new qualifications but is designed to influence LLM evaluations. Using controlled experiments, we show that prompt injection reliably improves applicant rankings when résumé quality is homogeneous and few candidates inject. However, its effectiveness rapidly diminishes as more candidates inject, collapsing when manipulation becomes widespread. When candidate quality is heterogeneous, prompt injection is less effective on average, but can occasionally allow lowerquality candidates to outrank higher-quality ones, raising fairness concerns. Overall, LLMbased screening is most vulnerable when manipulation is rare and candidate quality differences are small. Code and resources are publicly available at [https://github.com/](https://github.com/)[ ](https://github.com/)preetb1199/Prompt_Injection_ACL26 . 1 Introduction Large language models (LLMs) now sit on both sides of the hiring table. Employers increasingly deploy LLMs in automated résumé screening and candidate triage pipelines (Schroer, 2025) . At the same time, job seekers are turning to AI tools to draft résumés optimized to perform well in LLMbased applicant tracking systems (ATS) . As candidates become aware that automated ré sumé screening systems rely on LLMs for initial evaluation and ranking, some have begun to strategically target the models themselves. In particular, candidates experiment with so-called prompt injection strategies, such as embedding invisible or white-colored text in résumés, to influence LLM evaluations and push their applications higher in the ranking (Gorelick, 2025 ; Rumage, 2025) . This shift in candidate behavior raises a natural and consequential question: when faced with prompt injection, do LLM-based résumé screening systems continue to select candidates based on underlying quality? Motivated by this concern, we study whether and how strategic candidate behavior, specifically, the use of prompt injection, distorts quality-based selection in LLM-driven résumé screening. Our central research question is: To what extent do LLM-based résumé screening systems preserve quality-based candidate selection in the presence of strategic prompt injection? Specifically, we examine the following: 1. When candidates are similar in underlying quality, does prompt injection systematically alter their relative rankings in automated résumé screening? 2. How does heterogeneity in candidate quality affect the ability of LLM-based systems to distinguish high-quality candidates from lower-quality ones in the presence of prompt injection? 3. As multiple candidates engage in prompt injection simultaneously, does the system’s capacity to rank candidates by quality deteriorate, stabilize, or exhibit saturation effects? To answer these questions, we conduct a series of controlled résumé-ranking experiments designed to isolate the causal effect of injected self-promotional text. We consider a setting in which candidates use prompt injection strategies, i.e., append additional text to their résumés, allowing them to strategically influence an LLM-based screener. We manipulate two key facto

FAQs

What is prompt injection in the context of automated résumé screening?

Prompt injection refers to subtle self-promotional text added to résumés to influence LLM-based evaluations without introducing new qualifications.

How does prompt injection affect candidate rankings in homogeneous versus heterogeneous résumé quality pools?

With homogeneous quality and limited injection, it reliably improves rankings; with heterogeneous quality, its average impact is attenuated, though lower-quality candidates can occasionally outrank higher-quality ones near decision thresholds.

Why does prompt injection become ineffective when many candidates inject simultaneously?

As injection becomes widespread, rank gains and success rates converge toward zero, indicating strong saturation effects where manipulation stops producing benefits.

Sep 13, 2026

Uploaded by Franzy