Nobody tells you what to post — how platforms discipline the digital public

What is it about?

Do digital platforms simply connect us, or do they also shape how we interact and divide? Our study develops the idea of platform disciplinary mechanisms. Discipline here does not mean censorship or instruction. It means something quieter. Interfaces make certain responses effortless — a tap, a share, a reaction icon — while more considered ones become costly. Recommendation systems determine what comes to feel like "the world." Over time, people come to regulate themselves. We argue that this discipline has changed form. In an earlier phase, platforms watched and directed users from above. More recently, enforcement has become horizontal: approval and disapproval come from peers, and emotional conformity is policed within communities rather than imposed on them. To see whether that shift left a measurable trace, we tracked two things across 2,475 consecutive days between 2015 and 2021. The first was how widely emotional tone diverged across global news coverage each day. This captures what the information environment looked like, not how individuals felt inside it. The second was what five of the largest technology firms were building: 49,775 research abstracts from Apple, Amazon, Meta, Google and Microsoft, sorted into ten technology domains by machine learning. The clearest change is in how long things last. Before 2020, a surge in emotional divergence lost half its intensity within about ten hours. Afterwards, the same decay took nearly two days. That shift was most strongly associated with front-end technologies — recommendation and interaction design — rather than back-end computational capacity. Divergence had become less episodic — long enough that one disturbance could still be fading when the next arrived.

Featured Image

Photo by Steve A Johnson on Unsplash

Photo by Steve A Johnson on Unsplash

Why is it important?

Platform regulation is largely organized around one question: what harmful content is present, and how quickly is it removed. Our findings point to something that question cannot capture. The shift we observe after 2020 is one of duration: emotional divergence lasted longer after the events that prompted it. Duration belongs to the environment rather than to any single post, which is why counting removals cannot detect it. Reading this through the idea of discipline sharpens the point. If emotional norms are increasingly enforced by users on one another, the relevant lever is not the individual message but the architecture that makes such enforcement cheap and rewarding — reaction buttons, ranking, feedback loops. That is where our associations are concentrated. One result complicates the familiar narrative. The technology domains did not all move in the same direction: one showed a negative association with emotional divergence, suggesting that some design and data-handling choices may be linked to dampening rather than amplifying effects. Platform architecture does not appear to be uniformly polarizing. For regulators implementing frameworks such as the EU Digital Services Act, this points toward a different object of attention — the design conditions under which emotional polarization becomes self-sustaining.

Perspectives