TikTok allegedly gave thousands of users, including children and teenagers, non-functional versions of a safety feature while studying how the tool affected engagement and advertising revenue, according to newly unsealed allegations in New York’s lawsuit against the platform.
Reuters reported that users who selected TikTok’s Algo Refresh feature were allegedly placed into ghost or placebo experiments in which some received a version that did not change their feeds, despite believing the feature had been activated.
The allegations are part of a case brought by New York Attorney General Letitia James, one of more than two dozen state lawsuits accusing TikTok of misleading users about safety and designing features that encourage compulsive use.
Algo Refresh was designed to give users more control
TikTok introduced its refresh feature in 2023 as a way for people to change recommendations when their For You feed no longer felt relevant.
The company’s own documentation says users can refresh your For You feed to help us reshape your For You feed recommendations, alongside controls such as “Not interested,” keyword filters and topic management.
When TikTok originally announced the feature, it said activating refresh would let users see recommendations as though they had just signed up, with future content shaped by new interactions.
New York’s lawsuit alleges that some users who deliberately requested that reset instead received a non-working version during experiments conducted repeatedly in 2023.
An employee reportedly questioned the experiment
According to Reuters, an internal TikTok product manager working on digital wellbeing raised concerns about using placebo groups for Algo Refresh.
The employee reportedly wrote that the feature concerned “algorithm transparency and control” and that a placebo group conflicted with that purpose. TikTok nevertheless allegedly compared people receiving functional and non-functional versions to measure time spent on the platform and effects on advertising revenue.
TikTok disputes New York’s characterization of the experiments. A company spokesperson said TikTok routinely tests products and features in real-world environments and maintains that protecting younger users is a priority.
Safety experiments create a different product problem
A/B testing is standard practice in technology. Product teams routinely show different interfaces to groups of users to determine which version produces better outcomes.
Safety controls create a harder ethical boundary.
If someone deliberately activates a tool because they want to escape unwanted or potentially harmful recommendations, silently assigning that person to a placebo group changes what is being tested. The experiment is no longer simply about button placement or engagement.
It is testing what happens when a user believes a protection is functioning when it is not.
For product teams, the allegations raise a question that extends far beyond TikTok: safety features cannot necessarily be experimented on under the same assumptions as ordinary engagement mechanics.
The more platforms describe controls as giving users agency, the more important it becomes that the control actually works when the user chooses to exercise it.
