Social media offers many benefits but also carries risks, including exposure to distressing content. The UK’s Online Safety Act requires certain platforms to empower users to control the content they see. Content controls can reduce users’ exposure to sensitive content. However, there is little public data on how platform design shapes the use of these controls. In our online randomised controlled trial on a simulated social media platform, participants were given an initial choice between seeing ‘All content types’ or ‘Reduced sensitive content’. After browsing, they were given the opportunity to change their choice. In the Control arm, none of the options were pre-selected. 24% chose ‘Reduced sensitive content’. Pre-selecting ‘All content types’ reduced this proportion to 15%. Conversely, adding a description of ‘sensitive content’ on the choice page increased that figure to 29%. The initial choice proved to be ‘sticky’. When invited to review after browsing, those defaulted away from ‘Reduced sensitive content’ did not switch any more than those whose choice was not influenced by a default. Overall, user choice was susceptible to choice architecture, and users’ tendency to update their initial choice was weak. This highlights the importance of platform design to deliver genuine user empowerment.