On November 3rd, LinkedIn began using public data from its users in the European Union and other regions to train generative AI models. The platform will use certain information, such as profile details, posts, articles, comments, and CVs submitted during job applications to develop its AI systems.
My previous decision was to opt out. But I’ve changed my mind and here’s why.
Growing momentum
You might have become aware over the past weeks of a growing frustration amongst women on LinkedIn. An increasing number of us have noticed that our content was not getting the reach it used to on the platform. Through a system of informal introductions, we started to realise that unconnected women around the world were noticing the same pattern. Our voices are being dimmed. Now we are starting to come together in a more coherent and cohesive voice.
I found that my own informal experiment which I carried out with Robert Baker (outlined HERE) is still gaining traction. It is currently at 62000 impressions, and the results aligns with the work of others. From our shared notes we are all coming to the same conclusion. Something is amiss.
It’s not that our ideas aren’t strong. It seems that the system doesn’t always see or value them in the same way.
The Research is Clear
Bias in AI and professional networking is well-documented:
- Network disparities: LinkedIn analysis in the U.S. found that women, Black, Latino, and lower-income users have smaller professional networks compared to men, White, Asian, and higher-income users. Even Zip codes can make a measurable difference
- Global under-representation: A global study LinkedIn ad audience estimates show women remain underrepresented across countries, industries, and seniorities in professional, technical, and managerial roles.
- Algorithmic amplification of bias: Machine learning systems not only reflect existing gender stereotypes—they can amplify them. Bias isn’t just in the real world; it’s baked into the code.
- Job recommendation bias: Models trained on biographical data (including LinkedIn profiles) can display gender bias, but research shows this can also be mitigated.
- Communication style bias: Posts written in collaborative, thoughtful, or understated tones, which many women naturally use, are often under-amplified. Bold, confident, or self-promotional posts tend to perform better algorithmically. Topics that interest women may be de-prioritised unless tied to social justice, where the content is sometimes contained. See research from Martyn Redstone
The result? Our voices aren’t silenced, but they are softened.
Women’s Voices For Training
That’s why I’ve chosen to let LinkedIn use my data for AI training, because if these systems are going to learn from us, I want them to learn from women’s voices too. The algorithms use our language, leadership styles, and our ways of connecting. Plus our topics of interest which will frequently be different from our male colleagues.
Digital Equity
It’s about whose voices get heard, how they are presented, whose values get reflected, and which experiences help shape the digital spaces we all rely on.
So yes, I changed my mind and have now opted in. The next generation of AI should recognise our way of speaking as powerful, too. Algorithmic equity and visibility equity are inextricably connected in the age of AI.
Join us at 3Plus International in amplifying diverse voices your organisation.
