Lots of great stuff in here, Justin. I am confused by one thing, however. If the algorithm doesn't include gender in its analysis, why did changing a marker from male to female have such a powerful effect? Presumably, all of the proxies you discussed were already being accounted for previously.
Great question, and honestly, one I should have addressed more directly.
There are a few possibilities, and I should probably acknowledge that I’m not certain which one it is:
1. LinkedIn is being technically truthful but practically misleading. They may not use the gender field directly, but the gender marker might trigger other systems—ad targeting, recruiter search visibility, content categorization—that indirectly affect how often your posts get shown. The algorithm doesn’t “see” gender, but the broader platform ecosystem does.
2. The marker itself becomes a proxy. Once you declare male, you get sorted into patterns associated with male users—your content gets tested against audiences that engage more with “male-coded” profiles, you show up in different network clusters, etc. The marker isn’t used as a variable, but it changes which variables get applied to you.
3. A/B dynamics in the feed. The algorithm doesn’t just evaluate content—it evaluates who’s likely to engage with content from whom. If historical data shows users engage more with content from male-presenting profiles (consciously or not), then switching the marker changes which test audiences your posts get shown to first. Better early engagement = more amplification.
LinkedIn is probably telling the truth when they say gender isn’t a direct signal. But the gender marker likely affects which other signals get applied to you.
Think of it this way: the algorithm doesn’t ask “is this person male?” But it does ask “who is likely to engage with this content?”
And if historical patterns show that users (consciously or not) engage differently with profiles that present as male vs. female, then the marker changes which audiences your content gets tested on first.
Better early engagement = more amplification.
The algorithm never “saw” gender. But it didn’t need to.
That’s the insidious part.
The system can discriminate by gender without gender ever appearing in the code.
Lots of great stuff in here, Justin. I am confused by one thing, however. If the algorithm doesn't include gender in its analysis, why did changing a marker from male to female have such a powerful effect? Presumably, all of the proxies you discussed were already being accounted for previously.
Great question, and honestly, one I should have addressed more directly.
There are a few possibilities, and I should probably acknowledge that I’m not certain which one it is:
1. LinkedIn is being technically truthful but practically misleading. They may not use the gender field directly, but the gender marker might trigger other systems—ad targeting, recruiter search visibility, content categorization—that indirectly affect how often your posts get shown. The algorithm doesn’t “see” gender, but the broader platform ecosystem does.
2. The marker itself becomes a proxy. Once you declare male, you get sorted into patterns associated with male users—your content gets tested against audiences that engage more with “male-coded” profiles, you show up in different network clusters, etc. The marker isn’t used as a variable, but it changes which variables get applied to you.
3. A/B dynamics in the feed. The algorithm doesn’t just evaluate content—it evaluates who’s likely to engage with content from whom. If historical data shows users engage more with content from male-presenting profiles (consciously or not), then switching the marker changes which test audiences your posts get shown to first. Better early engagement = more amplification.
LinkedIn is probably telling the truth when they say gender isn’t a direct signal. But the gender marker likely affects which other signals get applied to you.
Think of it this way: the algorithm doesn’t ask “is this person male?” But it does ask “who is likely to engage with this content?”
And if historical patterns show that users (consciously or not) engage differently with profiles that present as male vs. female, then the marker changes which audiences your content gets tested on first.
Better early engagement = more amplification.
The algorithm never “saw” gender. But it didn’t need to.
That’s the insidious part.
The system can discriminate by gender without gender ever appearing in the code.
Good on you for realizing you benefited from the Algorithm, even if it was uncomfortable. We need more people doing this.
This was a great read Justin.