TikTok has paused the broad rollout of a new AI feature that automatically generates summaries of videos after users reported numerous nonsensical and misleading descriptions. The errors were not minor inaccuracies: in several instances the system produced clearly disconnected, surreal summaries.
Examples of the failures
Among the cases highlighted by Business Insider, one of the most notable involves Charli D’Amelio. Charli D’Amelio, who became the first TikTok creator to surpass 100 million followers in November 2020, had a simple talking-head video described by the AI as a “collection of blueberries with various toppings.” In another instance, a dog training video that explained canine behaviour was labelled by the system as “a complex origami art performance.” The AI described a Shakira promotional clip as “repeated motion of blue shapes,” and replaced the content of an Olivia Rodrigo promo with “nonsensical sequences of letters and numbers.” Several users said it appeared as if the system generated random text unrelated to the video being analyzed.
Response and limitations
Following user feedback, TikTok confirmed it will park the feature for general use. The company said it will repurpose the tool primarily for recognizing products shown in videos rather than automatically describing entire pieces of content. The development had been tested only with a limited number of users in the United States and a few other markets.
TikTok did not specify which AI model powered the summaries. In-app descriptions indicate the system may have relied on a mix of in-house technology and external models.
The broader issue: AI hallucinations
This incident is another example of how even advanced AI systems are prone to hallucinations — producing fabricated or baseless information. While AI developers regularly stress the technology’s potential to transform labour markets and digital services, errors like mistaking videos for blueberries or origami pose significant reputational risks for platforms.
Why the timing matters for ByteDance
The mishap is especially sensitive for ByteDance, TikTok’s parent company, which has seen rapid valuation growth. Recent private-market transactions value ByteDance at about $550 billion, compared with roughly $330 billion a year earlier — an increase of roughly two-thirds. Investors are watching closely in part because a future IPO could rank among the largest technology offerings ever.
ByteDance also survived an attempt to ban TikTok in the United States: a deal reached last December allowed the app to continue operating in the U.S. In that context, a visible AI failure is not only a technical embarrassment but also a potential business and reputational risk for a company many investors see as the next global tech giant.
Next steps
TikTok has scaled back the automatic summary capability and is focusing on refining the feature and investigating the errors. The company says testing will continue in a limited fashion while developers work to improve accuracy.
