Social media fragmentation: do online platforms drown democracy in noise?

Big platforms like X promised to become a digital meeting place akin to town squares; not everyone is convinced it has worked out this way. Guiziou Franck / AFP

With online platforms under pressure and artificial intelligence set to potentially flood the internet with content, is it all over for grand ideas about a harmonious digital public sphere?

When someone with fewer than 5,000 followers quits the social media platform X, they shouldn’t expect to make the news. But not everybody is the Swiss Interior Minister. So, when Elisabeth Baume-Schneider posted her last tweet in October 2024, saying X had “profoundly changed” and she’d had enough of the debate culture, it didn’t go unnoticed.

In jumping ship, the left-wing politician was jumping on a global trend. In 2022, after buying the platform, Elon Musk promised a “common digital town square” which would be “warm and welcoming for all”; by January 2025, an advisory group to the Swiss government was warningExternal link that the commercial and opinion-forming power of X – and other big platforms like Facebook or TikTok – was a threat to democracy itself.

Elisabeth Baume-Schneider’s retreat to politer Instagram

Indeed, over the past two years, hate speech has increasedExternal link on X and the tone has roughened, partly due to eased moderation rules and a culture shift, re-igniting debates about what a decent online public sphere would look like. And while Musk’s overhauls – and the similar approach decided more recently by Meta – have come as good news to free speech fans, others have been moving elsewhere – including many advertisers.

But for people worried about polarisation and anti-democratic discourse, does abandoning a platform like X achieve anything? Or is an exodus to politer places – in Baume-Schneider’s case, it was Meta-owned Instagram – just a retreat into “digital gated communities”, as the Neue Zürcher Zeitung wrote?

“Echo chambers do exist, but it’s not clear how big a problem they are,” says Emma Hoes, a digital technology and online behaviour researcher at the University of Zurich.

Emma Hoes: In the echo chamber of echo chambers

Hoes reckons talking about echo chambers all the time might even make them into more of an issue than they are, pushing us into an “echo chamber of echo chambers”. “Ultimately, at least some research shows that what we see on social media is more diverse than what we see in our offline life,” she says. Even in places like the independent service Bluesky – where many of Hoes’ colleagues have moved – “there’s accidental exposure to things we didn’t choose to see”.

In any case: if you did design a system to push opposing views in front of users all the time (a model which sounds like how some free speech absolutists would describe X) it might not lead to more open-mindedness. “One of the most consistent research findings is that political views remain super stable over time,” Hoes says. “People form their outlooks early in life and are not easily swayed by tinkering with social media feeds.” Even in the age of endless opinion and information, “people don’t really change their minds”.

Emma Hoes is a postdoctoral research fellow at the University of Zurich, where she studies how digital technologies influence the quality of information ecosystems. Courtesy of

It’s also easy to over-estimate the prevalence of political debates on the internet. In reality, Hoes says, most people consume “crap” online, whether it’s entertainment or culture-war fodder. Add TikTok, porn, gambling, gaming and the rest to the mix, and the picture is not necessarily one of polarising policy discussions but one that looks more like “news deprivation”.

It’s also easy to be overly idealistic about the prospect of a digital public sphere – an online space of deliberation which allows for many views to be channelled into something representing a democratic will. Rather than aggregate information in this way, one of the main points of the internet has always been to connect similar, often niche, interests. People find others who share their language, whether it’s harmonious or hate-filled; communities meet to discuss crosswords or the ellipsoid nature of the Earth.

This can lead to a splintering of political opinion into disparate groups which find it difficult to coalesce around a common cause. In a 2024 paperExternal link, Hoes’ University of Zurich colleagues Renate Fischer and Otfried Jarren wrote that the sheer size, speed, and diversity of ideas online makes it difficult to pull out a common public opinion that can be used to take action in a democracy. The public sphere then “loses its power to stabilise society and to integrate and it is becoming increasingly difficult to transform deliberative processes into political ones”, they wrote.

In short: we’re all online, giving our opinion, surfing the growing seas of content; at the same time, with trust in democracy sinking in many western countries, it’s hard to see what all the online noise adds up to – if it adds up to anything.

Hannes Bajohr: ChatGPT as a discourse machine

“Well, that’s the whole point of the internet – it’s so vast that no one can process it all,” says Hannes Bajohr from the University of California, Berkeley. Bajohr, who researches how AI and large-language models (LLMs) influence text and writing, also says things could be about to become much vaster.

LLMs like ChatGPT, which enable basically anyone to become a creator of text, video, or music, could hyper-accelerate the already countless streams of online stuff. This could make it even harder than it already is to find good information. At the extreme end, it could even lead to “artificial public spheres”, Bajohr says – entire spaces where you can’t tell if something is written by a human or a computer. Bajohr warns that such uncertainty would put key democratic ideas like trust, truthfulness, and responsibility under pressure.

Hannes Bajohr is an assistant professor in the Department of German at the University of California, Berkeley. He previously held post-doc positions at the University of Basel and the Collegium Helveticum in Zurich. Jen Siska / Courtesy of

LLMs could also change the type of discourse humans use in the first place. Since they are programmed using specific data and according to specific choices, tools like ChatGPT don’t produce the kind of neutral language their creators claim, Bajohr says. Like social platforms, they have biasesExternal link and they write a certain way. And since they are statistical, they run the risk of “value lock”, Bajohr says: that is, getting trapped in a way of speaking which can’t quickly adapt to political or linguistic shifts in the real world.

For Bajohr, these might not be such big issues if there were millions of such LLMs to play with. But in a field limited – at least for now – to a few profit-driven firms, it could lead to “a certain streamlining of language and a certain type of discourse, prescribed by these companies”. In the long run this could even lead to a streamlining of thought that would remove some of the communicative competencies which humans need for democratic deliberation, he reckons.

Should you nudge people to be nicer?

Meanwhile, efforts to regulate AI and big platforms don’t generally focus on how people say things or where they say them, unless it comes to enforcing laws against hate speech. But on a free internet, you can’t force people to gather in the same spaces, even if these spaces are well-designed. Neither can you prosecute people for being just mildly offensive.

Whether you can nudge people to be nicer, or to write a certain way, is another question. On the corporate level, it’s already happening. Apple, for example, recently plugged an AI tool that can make your emails less abrasive; auto-complete tools and spell-check features could even be seen as precursors to such well-meaning tools. These are all fine as long as they are opt-in, Bajohr says – the dystopian idea is that such features get slotted into communication tools without us knowing about it.

Overall, both Bajohr and Hoes agree that meddling with what people say and share is tricky. For Hoes however, it’s important not to get too wrapped up in the problems. “While there’s a lot of ‘bad’ content online, it’s still a minority of what people consume,” she says. She thinks all the ingredients – accurate and diverse information, broad participation – for a well-working online public sphere are already in place. It’s up to people to find them, or to help others find them. “People have all the tools they need. They just don’t always use them,” says Hoes.

Edited by Benjamin von Wyl/ac