How Europe disagrees online, and what turns disagreement toxic
Online polarisation is usually treated as a problem of too much disagreement. Across five countries and six contested topics, this study found the real issue is where disagreement is aimed.
The problem
Polarisation across Europe is usually met with calls for civility, fact-checking, or content takedowns. The Evens Foundation's Polarity project started from a different premise: most people are less extreme than the loudest voices online suggest, yet those voices are increasingly mistaken for the norm, because hostility travels faster and further than substance. Evens needed to understand what actually turns a disagreement toxic, so that the creators and communicators it supports could engage contested issues without feeding the dynamics they are trying to counter.
What we did
Ripple Research examined the conduct of disagreement rather than its volume. Immigration was held constant across all five countries as a shared reference point, and each country added one distinctive local topic chosen from outside the usual culture-war flashpoints: wolf and bear conservation in France, tradwife content in Germany, the Amsterdam meat advertising ban in the Netherlands, wildfires in Portugal, and nuclear power in Sweden. The unit of analysis was the interaction pair, one video matched to one top-level comment, so that audience behaviour could be read against the creator's original framing. Every pair was coded twice, once by a human analyst and once by a language model applying identical definitions, with the analyst resolving any disagreement.
What we found
The target of disagreement, not its strength, is the strongest predictor of hostility.
Two posts arguing the same anti-immigration position, one in Germany and one in France, produced very different comment sections. The German post named and accused its opponents and drew contempt in about 15% of replies. The French post argued the same case at length without naming an adversary and drew contempt in about 7%. Sharp disagreement stayed productive as long as it aimed at the claim, and turned toxic the moment it aimed at the person.
How a creator frames a post sets the starting conditions for everything that follows.
Across the 4,750 coded interactions, informative posts drew contempt in roughly 13% of cases, argumentative posts in 21%, and accusatory posts in 28%. Accusatory framing attracted more than twice the contempt of informative framing, and the ordering held across nine of the ten country-topic combinations.
Accusatory framing draws about twice the contempt of informative framing, and the ordering holds in every country and on all six topics.
What a post is about matters less than how it is framed, and how it is framed is the part a creator controls.
Low contempt often signals agreement rather than healthy debate.
Informative posts drew the least contempt, but around 63% of their comments sat in what the study calls the Comfy Bubble, where like-minded people agreed without testing a position. A single hostility score cannot tell a calm exchange of views apart from a room where no one disagrees.
Much of the contempt carried no abusive language at all.
A recurring move was to deny an opponent standing to speak, without a single offensive word.
“Your opinion about Germany is completely irrelevant.”
Said to a commenter living abroad. No abusive term, yet it removes the person from the conversation entirely.
Keyword and profanity filters miss this quieter exclusion, even though it shuts a person out as effectively as an insult.
Constructive disagreement is observable and repeatable.
A review of the productive exchanges, which the study calls the Fair Fight, found five recurring behaviours present in every country and topic: frame the argument around the claim, provide a reason or example, make the trade-offs visible, be forceful about the idea without attacking the speaker, and leave room for a reply. This turns the abstract appeal to be civil into a set of habits that can be modelled and rehearsed.
Sharp disagreement does not have to become hostility. The shift happens when the object of disagreement moves from the claim to the person.
What it led to
The research produced the Beef to Debate Decoder, a framework that classifies both a creator's framing and the resulting comments, and that now underpins the Polarity network's training and editorial work with creators across Europe. It translates into a short pre-publication framing check for creators, a shared vocabulary for civil society teams briefing the creators they work with, and a conduct-based lens that speaks directly to platform obligations under the Digital Services Act, where systemic-risk assessments still lean almost entirely on keyword toxicity. The method is portable to any campaign team wanting to see whether its content invites disagreement or invites a verdict.
Working on a similar question?
Most engagements start with a conversation about what you're trying to understand and what a useful answer would look like.