A recent social media experiment, dubbed #WearthePants, has ignited a debate over potential gender bias within LinkedIn's algorithm. Women users reported significant increases in post engagement and impressions after temporarily changing their profiles to male, leading many to suspect the platform's new AI-driven content surfacing system might be implicitly favoring male-coded content.

The controversy emerged after LinkedIn's Vice President of Engineering, Tim Jurka, announced in August that the platform had begun implementing Large Language Models (LLMs) to enhance content relevance for users. Following this, many prominent LinkedIn users observed a noticeable decline in their engagement and impressions, prompting suspicions about the algorithm's fairness.

The #WearthePants Experiment Uncovers Discrepancies

In November, a product strategist, referred to as Michelle (not her real name), decided to test this hypothesis. With over 10,000 followers, she noticed her posts often received similar impressions to those of her husband, who has only about 2,000 followers. Believing gender might be the "only significant variable," she changed her LinkedIn profile name to Michael and switched her gender to male, as she told TechCrunch.

Michelle was part of a broader movement. Marilynn Joyner, a founder who consistently posts on LinkedIn, also saw her visibility drop over several months. Upon changing her profile gender from female to male, Joyner reported a staggering 238% jump in impressions within a single day, she shared with TechCrunch. Similar results were reported by numerous other women, including Megan Cornish, Rosie Taylor, Jessica Doyle Mekkes, Abby Nydam, Felicity Menzies, and Lucy Ferguson, as documented in various posts and reports.

The #WearthePants experiment was initiated by entrepreneurs Cindy Gallop and Jane Evans. They asked two men to post identical content to theirs, curious if gender was behind the dip in engagement many women felt. Despite Gallop and Evans having a combined following of over 150,000 compared to the men's approximately 9,400 at the time, Gallop reported that her post reached only 801 people. In contrast, the man who posted the exact same content reached 10,408 people, exceeding 100% of his followers. These findings fueled concerns among women like Joyner, who stated, "I'd really love to see LinkedIn take accountability for any bias that may exist within its algorithm."

LinkedIn's Official Stance and Expert Analysis

LinkedIn has firmly denied these claims. The company stated that its "algorithm and AI systems do not use demographic information such as age, race, or gender as a signal to determine the visibility of content, profile, or posts in the Feed." LinkedIn's Head of Responsible AI and Governance, Sakshi Jain, reiterated this position in November, explaining that demographic data is only used for internal testing to ensure content from diverse creators competes equally and that the scrolling experience is consistent across audiences.

However, social algorithm experts suggest that while explicit sexism might not be the direct cause, implicit bias could still be at play. Brandeis Marshall, a data ethics consultant, emphasized the intricate nature of platforms, describing them as "an intricate symphony of algorithms that pull specific mathematical and social levers, simultaneously and constantly." She noted that changing a profile photo and name is just one lever, and the algorithm is also influenced by a user's past and current interactions with content. "This is a more complicated problem than people assume," Marshall added, highlighting the lack of transparency around all the factors that prioritize one person's content over another.

The "algorithmic black box" surrounding how companies implement their AI systems makes it difficult to ascertain the exact mechanisms. LinkedIn, like other LLM-dependent platforms, offers scant details on its content-picking model training. Marshall pointed out that many platforms "innately have embedded a white, male, Western-centric viewpoint" due to the demographics of those who train the models. Researchers find evidence of human biases like sexism and racism in popular LLMs because they are trained on human-generated content, and humans are often directly involved in post-training or reinforcement learning.

Beyond Explicit Bias: Tone, Style, and User Behavior

Other factors, such as tone and writing style, might contribute to the observed engagement shifts. Michelle, for instance, noted that during the week she posted as "Michael," she consciously adopted a more simplistic, direct writing style, similar to how she ghostwrites for her husband. This adjustment coincided with her reported 200% jump in impressions and 27% rise in engagements. She concluded that while the system might not be "explicitly sexist," it seemed to devalue communication styles commonly associated with women.

Stereotypical male writing styles are often perceived as more concise, whereas female writing styles are sometimes stereotyped as softer and more emotional. If an LLM is trained to prioritize content aligning with male communication stereotypes, this represents a subtle, implicit bias—a common issue in LLMs, as previous reports have indicated.

Sarah Dean, an assistant professor of computer science at Cornell, explained that platforms like LinkedIn consider entire profiles and user behavior, including job history and engagement patterns, when boosting content. "Someone’s demographics can affect ‘both sides’ of the algorithm – what they see and who sees what they post," Dean stated. LinkedIn confirmed that its AI systems analyze hundreds of signals, including profile insights, network, and activity, to deliver relevant content. The company has also been noted for researching and adjusting its algorithm to try to provide a less biased experience for users.

Chad Johnson, a sales expert active on LinkedIn, observed that the new LLM system appears to deprioritize likes, comments, and reposts, focusing instead on "whether your writing shows understanding, clarity, and value." This shift in algorithmic priorities further complicates the interpretation of #WearthePants results.

Widespread Discontent and the Quest for Transparency

The changes have led to widespread confusion and dissatisfaction among many users, regardless of gender. Data scientist Shailvi Wakhulu, who used to get thousands of impressions daily, now struggles to reach a few hundred. "It’s demotivating for content creators with a large loyal following," she remarked. While some men also reported engagement drops, others saw increases, attributing success to writing on "specific topics for specific audiences," which they believe the new algorithm rewards.

Brandeis Marshall, who is Black, noted another potential area of implicit bias: her posts about personal experiences perform worse than those related to her race. "If Black women only get interactions when they talk about black women but not when they talk about their particular expertise, then that’s a bias," she said. This suggests the algorithm might be amplifying existing signals or historical response patterns across the platform, rather than explicitly targeting demographics.

LinkedIn advises users that increased competition (15% year-over-year post growth, 24% comment growth) means content needs to stand out. Posts offering professional insights, career lessons, industry news, analysis, and educational content related to work, business, and the economy are currently performing well. Despite these insights, many users, including Michelle, yearn for greater transparency regarding the algorithm's workings. However, as content-picking algorithms are closely guarded company secrets, often to prevent them from being "gamed," such transparency remains a significant and likely unfulfilled request.