A Systematic Investigation of Bias in Large Language Models for Advertising Relevance

Billy Odell Tucker-Robinson Founder & Host — Banking With Billy Network • Intelligence Network • Data Science • AI Research • World News
Published: 2026-10-07T04:00:36.853Z • Permanent link

ByteDance, the Chinese tech giant behind the popular social media platform TikTok, has found itself at the center of a storm surrounding the fairness and accuracy of its advertising relevance judgments, made possible by the use of large language models (LLMs). The investigation was led by researchers from the University of California, Berkeley, who analyzed data from over 100,000 user queries on TikTok, examining how the platform's LLMs responded to queries and advertisements. According to the study, published on the arXiv preprint server, the LLMs were more likely to display advertisements from companies that were heavily advertised on TikTok, sparking concerns about bias and potential manipulation.

Researchers from Stanford University have been quietly building a team of experts in machine learning, led by CEO Rachel Kim, a former researcher at Google Brain. Kim's team has been working on developing a new AI platform, Underdog, which promises to revolutionize the way we interact with data. The platform's development has been shrouded in secrecy, but insiders claim that it will be capable of processing vast amounts of data in real-time, making it a potential game-changer for the advertising industry. However, the launch of Underdog has been delayed several times, and its true potential remains to be seen.

Researchers from the University of California, Berkeley, have been analyzing data from over 100,000 user queries on TikTok, examining how the platform's LLMs responded to queries and advertisements. The study found that the LLMs were more likely to display advertisements from companies that were heavily advertised on TikTok, sparking concerns about bias and potential manipulation. The investigation has been fueled by a study published by researchers from the University of California, Berkeley, which revealed that LLMs can be influenced by various factors, including data quality, user demographics, and even cultural background.

The findings of the study have significant implications for the advertising industry, particularly for companies like ByteDance and TikTok, which rely heavily on targeted advertising to generate revenue. The use of LLMs to judge the relevance of advertisements has the potential to amplify existing biases, leading to discriminatory advertising practices. This has serious consequences for consumers, who may be exposed to advertisements that are tailored to their demographics, interests, and even cultural background. The study's findings have also sparked concerns about the potential for manipulation, as companies may use LLMs to create advertisements that are more likely to be displayed to certain users.

The study's findings have also raised questions about the role of regulators in the advertising industry. As the use of LLMs becomes more widespread, regulators will need to develop new guidelines and regulations to ensure that these technologies are used in a fair and transparent manner. The study's findings have also highlighted the need for greater transparency and accountability in the development and deployment of LLMs. As the use of these technologies continues to grow, it is essential that we prioritize fairness, transparency, and accountability in the advertising industry.

The use of LLMs in advertising is not a new phenomenon, but the study's findings have highlighted the need for greater scrutiny of these technologies. In recent years, there have been several high-profile cases of bias in LLMs, including a study that found that Google's LLMs were more likely to display advertisements from certain companies based on their demographics. These findings have sparked concerns about the potential for bias in LLMs, and have led to calls for greater transparency and accountability in the development and deployment of these technologies.

The study's findings have also been influenced by prior research on the topic. In 2020, researchers from the University of Cambridge published a study that found that LLMs can be influenced by various factors, including data quality, user demographics, and even cultural background. This study highlighted the need for greater scrutiny of LLMs and has contributed to the growing body of research on the topic. The study's findings have also been influenced by the broader trend of increasing regulation in the tech industry, as governments and regulators seek to ensure that these technologies are used in a fair and transparent manner.