The X Effect on Sky News presents the findings of an excellent piece of research into bias on the X social media platform. Over nine months, Sky News’ Data and Forensics team investigated whether X’s algorithms amplify right-wing and extreme content. Their findings were clear: Elon Musk’s X is boosting right-wing and extreme content, as well as politicians Musk favours.
The research was designed in consultation with multiple academics and researchers and built on the methodologies used by other similar researchers including The Wall Street Journal and Global Witness – it was thorough, objective and very well executed.
If bias exists in the content presented to users on X does it happen accidentally, or is it built into X’s algorithms? That begs the question, what are these algorithms we keep hearing about? If you are a mathematician or computer scientist you will already know, but most of us aren’t, so let’s look at what an X algorithm might look like.
The X Effect shows us the right-wing bias in X is not accidental
Musk has made it harder (more expensive) for researchers and academics to access data on X. For Sky News to research whether there is any bias in posts on X, they created nine new user accounts to emulate different British people. Three were politically left-leaning, three right-leaning, and three neutral.
The researchers wanted to see what content would be shown on each of these accounts, so they gathered data on each over a period of two weeks, collecting a dataset with nearly 90,000 entries. The dataset was then analysed using a large language AI model specifically trained and validated for this purpose. Details of how the research was conducted can be found on the Sky News website.
The research findings are clear: right-wing content was shown more prominently, regardless of the users’ political leaning. For example, neutral users saw twice as much right-wing content as left-wing content and almost none of the political content shown to neutral users was non-partisan, with them being shown twice as much right-wing content as left-wing content.
The X algorithms are pushing posts to new users that don’t align with their interests and this can’t be happening by accident. We know that many left-wing voices have left X and moved to alternatives such as Bluesky, but the Sky research ruled this out as a possible explanation of the right-wing bias.
The only possible explanation is that there is a right-wing bias built into the X algorithms and, according to Ned Mendez, data analyst at digital consultancy FourOneOne, and one of the Sky research team, “an algorithmic bias must be decided by senior people at the channel”.
What does social media know about us?
To understand how social media algorithms work we first need to be clear eyed about what social media platforms know about us as individuals – it’s a lot!
They collect this personal information because their business model relies on targeted advertising. Individuals and organisations create content and then pay the platform to put that content in front of specific types of user: you. To do this the platform needs to know about you.
They collect vast and detailed amounts of information, building comprehensive profiles that go far beyond the data you explicitly post.
Social media platforms are likely to know some or all of the following about you:
- Self-reported information: This is the data you provide directly when setting up and using an account, including identity and contact information, as well as personal details.
- Content and behavioural data: This includes almost everything you do on the platform, such as user activity (posts, comments, likes, reactions, and so on), your direct messages, and purchasing habits.
- Inferred data: The platforms analyse your activity to infer deeper, often sensitive, characteristics about you that you never directly stated. Your inferred data includes your interests and preferences, your psychological states, and your location.
- Off-platform and device data: This is data collected when you’re not actively using the app, such as web activity, device information (your IP address, operating system, etc) and data gathered from your permissions settings.
- Artificial intelligence (AI): Machine learning models analyse all the collected data (posts, photos, interactions) to make highly accurate inferences about your personality, interests, political leanings, life events, and purchasing intent, making your profile far more detailed than the data you entered manually.
- Facial and object recognition: AI is used to analyse photos and videos you upload to identify people (including those not tagged), objects, activities and locations.
What is a social media algorithm?
X gets around 500 million posts every day so how does it decide which ones you see and in what order? That’s the algorithm: a set of step-by-step instructions for solving a problem.
How might a social media algorithm work? It collects all the recent posts from people the user follows plus a few recent posts from accounts the user might like based on similar interests or what’s popular. The algorithm then calculates a “Relevance Score” for each post, assessing user interaction, the popularity of the post in general, and which formats (video, photo, text) the user usually engages with.
That’s fine so far but there’s another, big factor to take account of: advertising. Individuals and businesses, including political parties, pay to have their posts put in front of you and the more they pay, the more people they can reach. This is taken care of by another step in the algorithm.
The algorithm checks its database for all ads that are currently running campaigns targeting the user’s profile. For each targeted ad, the algorithm calculates an Ad Score, based on the bid (how much the advertiser is willing to pay to show the ad), the Estimated Action Rate (a prediction of how likely the user is to interact with the ad), and the Ad Quality/Relevance (how good the ad is, with a clear image and positive user feedback). The higher the Ad Score, the more valuable it is to both the advertiser and the platform, although not necessarily to you.
Finally, the algorithm needs to merge ad posts and the organic posts. It does this by interleaving and displaying the feed, using a set of rules to insert high-scoring ads. The final feed is then a ranked list of posts and ads, and this is what you see.
Following such an algorithm should be politically neutral – posts should reflect your own politics. But do they?
Now we have proof of bias
Thanks to Sky News we now have hard evidence that the right-wing bias we have seen subjectively in X’s algorithms is real, deliberate, and comes from the top. We don’t need research to show us the bias in Trump’s Truth Social and we have our own views on Facebook bias.
X has over 600 million monthly active users. The right-wing bias inherent in the platform is ultimately under the control of one man, the platform’s owner. The threat posed by so much power and reach resting in the hands of one man should be a concern to all of us.
