The Algorithm and the Hating: how outrage became fashion’s invisible engine
From Crocs to Balenciaga, ridicule has accelerated trends rather than ended them. Three voices from inside the industry explain why — and what it costs
Why engagement-based algorithms reward outrage and reaction over admiration, reshaping how trends form and how designers gain visibility in the fashion industry.
The major social media platforms do not reward the most admired content. They reward the most reacted-to content. That distinction, which the platforms understood long before the industry did, has restructured how trends form, how designers gain visibility, and how a mocked look can outsell a celebrated one. Outrage, in the attention economy, functions as distribution.
The industry has long known that controversy sells. What changed is the infrastructure behind it — and who benefits.
This investigation examines how engagement-based systems shape visibility, desirability, and trend cycles, through conversations with Nectaria Panagiotou, a London-based publicist and researcher in platform culture; Shannon Mirabelli-Lopez, former curator at the Costume Institute of the Metropolitan Museum of Art; and Victoria Carina, founder and designer of The Genuine Human.
How Facebook, Twitter, and Instagram replaced chronological feeds with engagement-based ranking systems built on behavioral signals rather than approval.
Facebook abandoned its chronological feed first — introducing engagement-based ranking as early as 2009, with significant updates through 2011 and 2016. Twitter and Instagram followed in 2016. Once the feed was ranked, every platform needed something to rank by. What they chose were behavioral signals: watch time, completion rate, comments, shares. Not one of those signals measures approval. They measure whether a user reacted.
Ms. Panagiotou, whose research examined how streaming platforms restructured the music industry before turning to digital culture, described the internal logic of that design choice: «The systems are designed to maximize reaction, and described as if they maximize satisfaction».
Both things can be simultaneously true. Internal Facebook research, leaked in 2020 and 2021, confirmed that the company knew its ranking systems favored divisive content. At one point, the ranking formula weighted anger emoji reactions more heavily than likes. That was a design decision, made and measured.
There is a direct parallel to music. Spotify’s algorithmic playlists are presented as a personal discovery service. The underlying system optimizes for retention. Scholar Robert Prey has described this as «curatorial power»: systems with commercial incentives, presented as mirrors.
How mockery of Crocs, Balenciaga, and MSCHF’s Big Red Boots turned public ridicule into a distribution engine for engagement and trend visibility.
The consequence of that design is that backlash acts as promotion. Ms. Panagiotou traces the Crocs arc as a case study.
For most of the two-thousand-tens, Crocs were the internet’s shorthand for poor taste. Every “worst shoe ever” post was, in algorithmic terms, a vote for more Crocs content. Christopher Kane put embellished Crocs on a London runway in 2016. Balenciaga followed with platform Crocs in 2017.
The coverage was overwhelmingly mocking and enormous. Reaction content — every stitch, duet, and outraged quote-post — handed the image to audiences the original never would have reached. Crocs posted record revenues within years of peak ridicule.
Ms. Panagiotou identifies the same pattern in compressed form with MSCHF’s Big Red Boots in 2023: «The mockery wasn’t an obstacle to the sale. The mockery was the campaign, just one nobody commissioned».
The algorithm read the spike of mockery as relevance and pushed the content into more feeds. Decades of behavioral research confirm what followed: repeated exposure produces familiarity. Familiarity shades into acceptance, even when initial encounters are hostile. Creators begin styling a mocked item ironically. Irony quickly collapses into sincerity.
Some publicists have drawn the obvious conclusion from watching a polarizing rollout outperform a celebrated one. Ms. Panagiotou is candid about what that looks like in practice: «Seed a detail that’s slightly too provocative, let the criticism do the distribution, and harvest the reach». The system rewards it. But she identifies a limit: «You can start that machine. You cannot steer it».
What separates aesthetic provocation from a trust violation in fashion controversies, from Balenciaga’s teddy bears to Burberry and Calvin Klein.
Ms. Carina, approaching the question as a working designer, draws a structural line. Controversy that challenges aesthetic preferences operates under one logic. Controversy that challenges shared values operates under another.
She identifies three cases where brands crossed from the first category into the second: Balenciaga’s teddy bear campaign, Burberry’s hoodie, and Calvin Klein’s teenage advertising from 1995. In each instance, the brand retreated publicly. «A simple disruption of established design standards can turn into a new trend», Ms. Carina said. «But a trust violation lies on the other side of that line».
The defining variable is not the degree of provocation. It is what the content reveals about the brand’s values. «A brand can survive backlash if a large number of people believe the provocation had artistic intent or was a simple mistake», she said.
Those three incidents span three decades, which raises the question of whether the threshold itself has moved. Ms. Carina argues that it has, but not uniformly. Consciousness around human exploitation has expanded. At the same time, Western audiences have grown more accustomed to sexual provocation. The line moves predictably, shaped by expanding cultural awareness in some areas and normalization in others.
Ms. Panagiotou adds a technical dimension. The line between criticism and harassment is not a content problem. It’s a scale problem. «The harm is emergent. Nobody in the pile-on thinks of themselves as a participant in harassment; they think they’re one person with an opinion. The algorithm is what assembles them into a crowd».
Moderation systems evaluate content one piece at a time. Aggregation — the platform’s own output, produced by its ranking system — goes unmoderated. The practical signals Ms. Panagiotou watches inside a backlash cycle: velocity; the shift in object, when discourse moves from the collection to the designer’s person or right to exist in the industry; and migration, when it crosses platforms and arrives in direct messages. When those three align, what began as commentary has become a safety situation.
Why a former Costume Institute curator argues that algorithms shape trends and visibility but not the historical or archival meaning of fashion
Ms. Mirabelli-Lopez, who spent twelve years at the Costume Institute before moving into design education at Pratt, Parsons, and West Valley College in Silicon Valley, approaches the same machinery from the institutional side.
She does not concede that algorithms have displaced curatorial judgment. «Algorithms don’t decide what fashion means. They might decide what trends in a given moment, but fashion is diachronic», she said. The historical record continues to document the larger social and cultural themes of an era: silhouette, materials, technology, aesthetics, innovation.
On the Met Gala, which now generates some of the most-engaged fashion-adjacent content of any given year, Ms. Mirabelli-Lopez does not read algorithmic amplification as institutional compromise. The Gala has functioned as the Costume Institute’s primary fundraiser since the arrangement between the industry and the museum was established. The red carpet consists of celebrities interpreting an exhibition theme, with results of varying depth. What concerns her is a specific conflation: viewers collapsing the red carpet with the exhibition’s academic content. The two are only superficially related.
On the democratization argument — that platforms bypass elite gatekeepers and let audiences decide — Ms. Mirabelli-Lopez acknowledges the strongest version of the case: «The best thing algorithms do is promote independent, new, or young designers — people who may not have come out of traditional institutions — in a way that reaches a lot of people and can launch careers».
That is real. But entry is not equivalent to attention. Recommendation systems concentrate visibility on a narrow slice of what’s available. Past engagement is the strongest predictive signal available to them. The funnel widened at the top. It narrowed in the middle.
How algorithmic virality favors striking visuals over craftsmanship, embroidery, and refined design that rewards sustained attention.
Ms. Carina identifies a concrete cost that the engagement economy has imposed on the industry. Platforms favor content that is immediately striking. Designs in unusual shapes, drastic colors, and proportions that defy convention perform better algorithmically than technically refined work that rewards sustained attention.
What the feed cannot capture is craftsmanship that requires physical proximity: the sensorial quality of silk, the precision of fine embroidery, the engineering of interior garment construction. Those properties are structurally incompatible with the browsing behavior that platforms produce.
Ms. Carina raises a further question about where this trajectory leads: «What if in a culture where everyone is loud and scandalous to stay relevant, the one way that will eventually stand out is quiet, effortless elegance?» — the one that does not need spectacle to travel, and does not need outrage to be seen.
Ms. Mirabelli-Lopez draws a longer institutional arc: «Trends and marketing aren’t fashion — that will never change». Museums collect pieces that carry social, artistic, or cultural significance. That criterion has not changed. Only time will establish how wide the gap is between algorithmic prominence and archival value.
The hidden human labor of content moderators and data annotators in the Philippines, Kenya, and India who train recommendation algorithms
The conversation about algorithms stops short of the people who build them. Ms. Panagiotou’s research addresses the human infrastructure behind AI systems that platforms describe in almost entirely automated terms.
Content moderators, data annotators, and quality evaluators form the workforce that trains, refines, and maintains recommendation systems. This population is concentrated in regions where labor costs are low: content moderation hubs in the Philippines, outsourcing firms in Kenya and India, and distributed micro-task workers on platforms such as Amazon Mechanical Turk.
Researchers Mary L. Gray and Siddharth Suri have described this workforce as ghost work in their eponymous study. Scholar Sarah T. Roberts has produced one of the foundational accounts of commercial content moderation. Investigative reporting has documented moderators in Nairobi reviewing graphic content on behalf of major technology companies for low wages.
«What is often described as an algorithmic decision is more accurately understood as the outcome of multiple layers of human judgment embedded within socio-technical systems», Ms. Panagiotou said.
How data annotation guidelines and classification decisions determine which aesthetics, designers, and bodies receive algorithmic visibility
Data annotation is presented as neutral, technical work. Each label is a judgment. Annotators apply classification guidelines developed by platform or product teams to large volumes of data. Those guidelines reflect the operational definitions and policy decisions established during system design. The criteria are not determined by the workers applying them.
Because these systems operate across diverse cultural contexts, the values embedded in their training carry consequences for which aesthetics, bodies, and designers receive amplified global visibility.
The result, as Ms. Panagiotou describes it, is a continuous feedback loop rather than a static reflection of audience preference: «Recommendation and user behavior develop through a continuous feedback process rather than a one-way reflection of existing preferences».
Ms. Carina identifies the same dynamic at the level of brand communication. Well-intended content can be taken out of context as audiences reinterpret and redistribute it. At a certain point, secondary content — analysis threads, long-form takedowns, reaction videos — detaches from the original work entirely. «It happens the moment people begin consuming the rhetoric around the controversy without bothering to see the underlying content for themselves», she said. «At that point the controversy becomes self-sustaining».
What recommendation algorithms and generative AI mean for how young audiences form taste, culture, and identity through platforms.
Ms. Panagiotou identifies two areas where the public conversation about algorithmic influence has not yet caught up with the scale of the problem.
The first concerns how young people form cultural preferences. Many now encounter music, clothing, and culture primarily through algorithmically curated environments. The effects of that exposure during periods of identity formation remain an active area of research.
The second concerns the relationship between recommendation systems and generative AI. Contemporary generative models train on large-scale datasets that include platform content. The visibility of that content is itself shaped by recommendation algorithms. As AI-generated content returns to the same platforms, the feedback loop between recommendation and generation has become a question for governance.
Ms. Mirabelli-Lopez, operating from inside a Silicon Valley institution with direct ties to the tech industry, offers a note of perspective: «It may look like they have outsized influence on the industry, but history is what actually decides. Only time will tell».
Ms. Panagiotou’s research on the music industry reached a similar conclusion. It did so around 2015, a decade after streaming had already restructured the industry’s foundations.
The Algorithm and the Hating: An investigation into how engagement-based algorithms reward outrage over appreciation — and what that means for visibility, taste, and desirability in the fashion industry. A parallel look at the hidden human labor behind AI systems: the content moderators, data annotators, and micro-task workers training the platforms. With contributions from Nectaria Panagiotou, Shannon Mirabelli-Lopez, and Victoria Carina.
