The Algorithm Doesn't Have Taste, and Neither Do We Anymore


There was a two-week window last year when everyone was a "tomato girl." Before that, a "coastal grandma." Before that, "mob wife." Each one arrived fully formed; a name, a mood board, a shopping list, and each one was dead within a month, replaced by the next aesthetic with its own tidy little label. We used to call this a trend cycle. But now, it would be more honest to call it a symptom.
Because here's the thing nobody in fashion media wants to say plainly: these aren't trends. A trend, historically, was the visible output of something slower and more human. A shift in what people wanted, filtered through designers, buyers, editors, and the slow social process of watching what your friends wore and deciding whether you wanted it too. It took a season to arrive and a season to leave, and somewhere in that span, a lot of people with actual judgment, bad judgment, biased judgment, but judgment, decided what was worth paying attention to.
What we have now isn't that. It's an algorithmic feedback loop with the humans edited out.

How the loop actually works
Start with the platform. TikTok's For You feed doesn't care what's beautiful, original, or well made. It cares what keeps a thumb from scrolling for another four seconds. A greige minimalist bedroom holds attention slightly better than a maximalist one this week, so the algorithm shows it to more people. More people means more videos performing "the aesthetic" back at the algorithm, which reads the volume as demand and shows it to more people still. Nobody chose "coastal grandma." A distribution system noticed that content tagged like it was outperforming content that wasn't, and did what distribution systems do: more of what works.
Now add the brands. A fast-fashion retailer's job used to be forecasting; reading runway shows, buyer meetings, cultural signals, and betting on what would sell six months out. That job barely exists anymore, because it doesn't need to. Social listening tools now tell a brand, in near real time, exactly which words, colors, and silhouettes are already spiking. The brand doesn't have to guess what people will want; it can wait until the platform has already told it, then manufacture directly into a demand curve that's already proven itself. This is why a "microtrend" can go from a single viral video to a product on a fast-fashion site within two weeks. Nobody at the company had a point of view about tomato-red sundresses. They just read the graph.
So the loop closes: platform rewards what performs, brand manufactures what performs, consumer buys what's available and performing, and the resulting purchase becomes more data confirming that the aesthetic is "trending", which was never anything but the loop describing itself.
What got lost in the loop
The uncomfortable part isn't that this is manipulative, exactly. It's that it's empty. A trend used to mean something, even when it was silly. It was a snapshot of what a particular culture, at a particular moment, collectively decided was worth wanting. The original trend of dressing like a “mob wife” told you something, even obliquely, about a cultural mood: an appetite for excess and drama after several years of quiet luxury and quiet everything. That reading might be wrong, but it was a reading, an act of interpretation by someone with taste, however contestable.
What the algorithm produces isn't an interpretation. It's a correlation. It has no opinion about whether tomato-girl summer means anything, because it isn't capable of having opinions. It only knows what holds attention. And when brands skip straight from "this is spiking" to "manufacture more of it," they cut out the one step that used to matter: someone deciding, on purpose, that a thing was good. Not high-performing. Good.
That's the real cost, and it's not primarily an environmental one, though the overproduction is real. It's that we've quietly let a system with no aesthetic values whatsoever become the arbiter of what an aesthetic is. Ask someone what "clean girl" actually means and they'll describe a product list; slicked-back bun, gold hoops, a specific brand of lip oil, because that's all it ever was: not a point of view, but an SKU list wearing the language of a point of view.

Taste as a refusal
None of this is an argument for nostalgia, for pretending editors were ever neutral or that the old trend cycle was some golden age of pure taste. It was often just as manufactured, just slower and gatekept by fewer people with different biases. But it required someone to decide, and decisions can be argued with. You can't argue with an engagement metric. It doesn't have a position to defend.
If there's a way out, it isn't refusing trends altogether, that's its own kind of performance. It's slower: noticing when you're dressing an algorithm's correlation rather than expressing an actual preference, and being willing to sit with something past its two-week shelf life simply because you still like it, whether or not the feed has moved on. That's not a moral stance. It's just the last available way to have taste at all, in a system that would very much prefer you didn't need to.


