In a video from a Turkish broadcaster, a Palestinian woman walks among the rubble, wailing. “Where are my three daughters, my husband, and my cousin?” She went out to get flour, she says, and came back to find her home collapsed, her family apparently trapped inside.
Hit pause on the clip and something strange happens: a prompt reading “Find Similar” pops up. When a viewer clicks the prompt, TikTok automatically pulls videos that look visually close to the footage of the woman — and more disturbingly, it suggests products on TikTok Shop that look like what she is wearing in the video. Among the products suggested are a “Dubai Middle East Turkish Elegant Lace-Up Dress” and “Women’s Solid Color Knot Front Long Sleeve Dress.” The scanning feature also identified the head covering the woman is wearing and the beige handbag she carries as she shouts out for her family — and suggested similar-looking products on the TikTok shopping platform.
The scanning feature is a new addition to the TikTok app that some users (including myself) noticed over the weekend. In a notification sent by TikTok, the company describes it as visual search tags that use AI to identify objects in content and display similar products or posts. “Due to limitations in AI technology, errors such as inaccurate or irrelevant results may sometimes occur,” the notification reads. TikTok also notes that users have the option to turn the feature off for their own posts and for videos on their feeds. The feature appears to be limited in access, as some users report not yet having the “Find Similar” pop-up.
TikTok did not immediately respond to a request for comment.
The best way to describe the new feature is a systematized way to find dupes on TikTok. It’s scaling up a behavior that users already do: they see a video with cool sunglasses, take a screenshot of the item, and reverse image search on shopping platforms or Google. The visual search tells us what we already know about social media platforms — that one of their primary functions is to show billions of users new stuff to buy. The feature drops any pretense that may have existed and centers the commerce element of social media.
Tech platforms like TikTok, Instagram, and YouTube have long tried to be both recommender and retailer: they want you to find a thing you might want to buy and purchase it through or even directly from them, therefore capitalizing on the transaction that happened because of content on their platform. With a built-in automatic visual search, every piece of content is an opportunity for profit, regardless of whether the original video or photo was meant to sell something. The reverse image searching feature shows up on innocuous videos, like influencers showing off handbags or fashion-focused creators modeling their outfits. But it also shows up on a video from Ms. Rachel, the child-focused content creator who has been an outspoken advocate for children in Gaza facing starvation, illness, Israeli airstrikes, and death.
“Something that people don’t understand is that my career and reputation will never be as important as standing up for kids,” she says in the clip. The visual search feature picks out the blue-and-white-striped dress she is wearing and suggests similar-looking items for sale on TikTok Shop.
The problem with trying to monetize every inch of online engagement is twofold. For one, there is content that we put online that can’t — or shouldn’t — be used to sell something, and the systems that power e-commerce cannot understand context. Videos shouldn’t need to depict what experts call a genocide to be off-limits to the aggressive marketing tactics employed by these systems; some people don’t want to be a digital billboard, period. The TikTok feature also represents the ruthlessness of tech platforms in squeezing every cent from user content, transforming the experience of being online into browsing an endless shopping mall. In this way, at least, the feature is honest about what the platforms want the internet to be.
[Notigroup Newsroom in collaboration with other media outlets, with information from the following sources]
