Getting frustrated after applying to thousands of jobs on LinkedIn and never hearing back?

USC graduate student Siyi Zhou was one of them, struggling to land internships and jobs.

She turned to the online job platform Chinese in LA, where she applied for an administrative role and was finally invited to an interview.

When she arrived at the address listed in the job posting, she discovered it was a residential apartment instead of a professional office.

Feeling suspicious, she walked away from the interview. But the unsettling experience didn’t end there.

The recruiter persistently contacted her every day by text and email, asking intrusive personal questions about her gender and age and insisting that she return for the interview.

After the encounter, Zhou decided to investigate with her PhD advisor, Emilio Ferrara. Together, they discovered the recruiter’s phone number was listed on websites associated with human trafficking, or “escort sites,” confirming the job posting was a front for potential sex trafficking.

According to the International Labour Organization (ILO), more than 27 million people worldwide are victims of human trafficking, with fraudulent recruitment serving as a major pathway into exploitation.

Zhou set out to investigate online job postings linked to human trafficking across multiple job platforms. Specifically, she wanted to determine whether these deceptive job postings shared identifiable patterns and how they could be detected.

Working with Ferrara, Zhou and her team developed an automated AI system that can flag job ads linked to human trafficking before potential victims are exposed to harm.

Using the system, the researchers also identified the online job platforms with the highest concentrations of suspicious postings, the industries most frequently targeted by traffickers, common recruiter contact methods, preferred victim demographics described in the postings, and other recurring patterns.

Launched in 2023, the project resulted in the paper, “The Trafficker’s Pitch: Detecting Deceptive Recruitment in Online Job Boards,” which was published in the proceedings of the 2026 International Conference on Web and Social Media (ICWSM).

Ferrara is a professor in the USC Viterbi School of Engineering and the USC Mark and Mary Stevens School of Computing and AI‘s Thomas Lord Department of Computer Science and a research team leader at the Information Sciences Institute (ISI).

Collecting Data on Risky Job Ads and Training an AI System to Identify Them

The team leveraged large language models (LLMs) to identify risky job ads because many indicators are too subtle for humans to detect.

Human researchers often struggle to identify the hidden linguistic patterns traffickers use across multiple languages. In contrast, LLMs can embed thousands of features into a unified system to uncover multilingual associations.

AI models also excel at processing noisy, cross-lingual data across languages such as English, Chinese and Russian, while identifying the complex relationships traffickers use to mask illicit intentions with seemingly legitimate language.

Zhou emphasized that training these models enables the detection of dangerous job ads based on their distinct “linguistic style,” even before phone numbers are linked to illicit websites.

To build the system, the researchers collected more than 85,000 job advertisements from online job platforms and extracted the contact phone numbers listed in each posting. They then searched for those same phone numbers across the web and labeled ads as risky if the numbers also appeared on escort websites, creating a verified dataset for model training. Job ads from trusted platforms such as LinkedIn and Indeed served as the primary examples of safe postings, allowing the AI system to learn the linguistic and behavioral patterns that distinguish legitimate recruitment from trafficking-linked advertisements.

Researchers Identify Job Boards With High-Risk Listings

Chinese in LA was among the job boards with the highest total number of trafficking-linked job ads. However, because it hosts a large volume of overall listings, Zhou noted that it had a relatively low proportion of risky postings, with 885 suspicious ads among more than 45,000 listings.

The researchers cautioned that their classification method identifies signals warranting further investigation, not confirmed trafficking cases. 

Traffickers Prefer to Contact Job Seekers Through Phone Numbers

The researchers found that phone calls and SMS text messages are the primary communication channels used by human traffickers, as nearly all risky job ads required applicants to provide a phone number.

In contrast, job ads that asked applicants to provide an email address or visit a website had a 0% risk rate. These more formal communication channels were consistently free of trafficking-linked job postings.

Jobs Seeking Female Applicants Are More Likely to Be Linked to Human Trafficking

The study found that job ads seeking female applicants were at a much higher risk of being linked to human trafficking, with nearly 50% of those postings labeled as risky.

To Zhou’s surprise, her team also found a trend of traffickers specifically targeting couples, with about 30% of job ads seeking couples labeled as risky.

In contrast, job ads seeking only male applicants or expressing no gender preference posed a much lower risk, with fewer than 1% of those postings associated with human trafficking activity.

Wellness, Modeling, Nursing Among Top Industries at Risk

The study found that the wellness industry, including jobs such as massage therapy and physical therapy, had the highest concentration of trafficking-linked job ads, with more than 60% of postings labeled as risky.

Entertainment jobs, including modeling and photography assistant positions, ranked second, with about 20% of job ads linked to human trafficking.

Healthcare, including nursing positions, ranked third, with approximately 10% of job ads labeled as risky.

Job Listings With Unknown Website Domain Locations Carry Higher Trafficking Risk

The team found that job ads posted on platforms with an unspecified website domain location carried a much higher risk of being linked to human trafficking, with a risk rate of more than 50%.

When examining the geographic patterns of job platform domain locations, Zhou’s team found that New York (2,779 ads) and California (2,065 ads) had the highest number of trafficking-linked job postings.

In terms of concentration, Pennsylvania and Ohio had the highest proportions of risky job ads relative to their total number of postings, with more than 60% of listings in both states labeled as risky.

Traffickers Use Visa Promises and Multilingual Job Ads to Target Immigrants

The study found that trafficking-linked job ads shared hidden linguistic patterns across English, Chinese and Russian, allowing AI models to distinguish them from legitimate postings. Recruiters often used vague promises such as “easy visa,” “migration” or unusually high pay to lure job seekers while masking illicit intent. Zhou said these multilingual job boards frequently target immigrants, international students and other vulnerable workers searching for employment.

Published on September 22nd, 2026

Last updated on September 22nd, 2026