A document from TikTok's engineering team in Beijing details the product's addictive algorithm, which optimizes for “retention” and “time spent”
It's the most successful video app in the world. Our columnist has obtained an internal company document that offers … Tweets: @zeyiyang , @akcronin , @jerrydunleavy , @mims , @gwbstr , @acerbialberto , @jimbethell , and @mims See also Mediagazer Tweets: @zeyiyang : this is basically how an American boomer decribes TikTok to other American boomers. 33% not understanding how algorithms work + 33% “It's made in China!!!” + 33% worrying about data security only when it's not done by U.S. companies or government. https://www.nytimes.com/... Audrey Kurth Cronin / @akcronin : Good nuanced story. TikTok optimizes for ‘retention’ and ‘time spent’, collecting highly personal info & preferences. Do you want to create a probing, accurate profile that others access? Who controls your data & what they do w it in the future are key. https://www.nytimes.com/... Jerry Christmas / @jerrydunleavy : “The doc lifts the curtain on the company's seamless connection to its Chinese parent company, ByteDance... The doc makes clear that TikTok has done nothing to sever its ties with its Chinese parent.” I mean ya— the new TikTok CEO this year was also ByteDance's CFO til last month. https://twitter.com/... @mims : “In a few hours, [Tiktok's] algorithm can detect his musical tastes, his physical attraction, if he's depressed, if he might be into drugs... It could potentially be used to micro-target him or make him more addicted to the platform.” https://www.nytimes.com/... Graham Webster / @gwbstr : Leaked document reveals TikTok is designed to maximize the value its owner gets from users. Familiar tools and goals, well implemented. What's wacky is these addicting practices read as increasingly illegal in China, while legal from DC's view and envied by some Valley types https://twitter.com/... Alberto Acerbi / @acerbialberto : Another leaked document - this time on TikTok algorithm. tldr: it is pretty boring. “There seems to be some perception (by the media? or the public?) that they've cracked some magic code for recommendation, but most of what I've seen seems pretty normal” https://www.nytimes.com/... Lord Bethell / @jimbethell : Chilling to read. Clearly this AI is hugely effective. Imagine what we could achieve if more of the great minds of the worlds' AI labs could be more focused on solving challenges of climate, equality, longevity, etc, rather than manipulating minds with this unsatisfying rubbish? https://twitter.com/... @mims : “This system means that watch time is key. The algorithm tries to get people addicted rather than giving them what they really want,” says founder of Algo Transparency, on revelations about TikTok's algorithm. https://twitter.com/... See also Mediagazer
Context & Ripple Effects
TikTok had already drawn concern over how its content choices could shape perceptions of real-world events, including the limited visibility of Hong Kong protest videos. The engineering document gives that debate a product-level frame: retention and time spent are explicit optimization targets, alongside data used to build user profiles.
The disclosure also sits within a broader argument for platform rules that do not hinge solely on a company’s home country, as outlined in the call for a more general Big Tech regulatory model. TikTok’s continuing connection to ByteDance makes algorithmic design, data handling, and geopolitical trust overlapping issues rather than separate ones.
First-order effects
- TikTok and ByteDance face more pointed scrutiny over whether engagement metrics and detailed preference profiles produce user harms, particularly for younger users.
- TikTok’s public trust effort must address the disclosed design goals as well as its China-parent relationship; a later Transparency and Accountability Center outreach campaign speaks to that same credibility gap.
Second-order effects
- Meta’s subsequent plan to recommend more posts from “unconnected” sources shows how TikTok’s engagement-led discovery model pressures rival social platforms to adopt similar feed mechanics.
- Regulators evaluating TikTok gain a concrete basis to focus on recommendation objectives and data-derived profiling, rather than treating content moderation or corporate nationality as the only issues.
Third-order effects
- If scrutiny continues to center on optimization objectives, platform accountability is likely to shift toward auditable recommender-system incentives—what feeds optimize for and whose data informs them.
- The later report that executives discounted internal evidence of teen harms suggests that disclosure disputes can evolve into broader governance questions over whether product-risk research changes platform design.
The trend: TikTok is a focal point in the shift from policing individual posts to scrutinizing the engagement incentives and data systems that determine what platforms distribute.