A study of 1.6 million social media posts found that adding moral language increases engagement, but only up to a point. The most moralistic posts perform worse than posts with no moral language at all.
The debate about moral language and social media has been running for years. Researchers have documented that posts expressing moral outrage travel farther and faster online than purely factual content. Activists have internalized this finding, loading political content with moral framing to maximize reach. Platforms have been criticized for algorithmically rewarding the most emotionally charged moral language. The conventional wisdom has converged on a simple story: more moral language means more engagement.
A new study published in Nature Human Behaviour has tested that story against the largest dataset of social media posts ever used for this purpose. The story is more complicated than the conventional wisdom suggests, and the complication has practical consequences for anyone who has ever tried to communicate something important online.
Researchers led by Cristian Candia at Northwestern University’s Kellogg School of Management, with collaborators Mohammad Atari at Harvard University, Nour Kteily, and Brian Uzzi, analyzed 1,621,147 social media posts across three platforms. Their Twitter dataset contained 530,104 posts. Their Reddit dataset contained 1,048,653 posts. Their 8chan dataset contained 42,390 posts. The posts were drawn from 13 distinct socio-political topic areas including immigration, gun control, climate policy, COVID-19 policy, and racial justice, selected to capture a broad range of political content across the ideological spectrum.
To measure moral language, the team used Distributed Dictionary Representations, a technique that maps individual words through neural word embedding models to measure their proximity to moral concepts. Rather than simply counting morally charged words, the method captures the semantic field around each word, detecting moral loading even when it is conveyed through phrasing rather than explicit moral vocabulary. The researchers then scored each post on two distinct dimensions.
Moral loading measured how morally relevant the post was overall. Did it engage with themes of right and wrong, fairness and harm, purity and betrayal? A high moral loading score indicated a post dealing with genuinely moral subject matter.
Moral density measured something different: how concentrated the moral content was within the post’s text. A post about immigration policy might engage with moral themes at a high level while distributing that moral content sparsely across many words, or it might pack dense moral language into nearly every phrase, word after word carrying moral valence.
The distinction between these two dimensions is what unlocks the study’s most important finding.
The moral engagement curve
When the researchers modeled the relationship between moral language and engagement, they found that the two dimensions behaved in opposite ways.
Higher moral loading was consistently associated with higher engagement across all three platforms. Posts about genuinely morally relevant topics attracted more responses, more sharing, more interaction than posts with less moral relevance. This confirms the prior literature: moral stakes draw people in.
But higher moral density, conditional on moral loading, was consistently associated with lower engagement. The more a post concentrated its moral language into dense, repetitive moral framing, the less engagement it attracted.
Taken together, these two findings produce a curve. As a post moves from no moral content toward a moderate density of moral language, engagement increases. Beyond a specific threshold, engagement begins declining. At maximum moral saturation, engagement collapses entirely.
The threshold the researchers identified sat at approximately 30% moral density. Posts hitting this level of moral content concentration attracted the highest engagement. Below this threshold, engagement was 2.28 times lower than at the peak. Above it, engagement was 2.78 times lower than the peak. The most morally saturated posts in the dataset, the ones where nearly every word carried a moral charge, performed worse than posts with no moral language at all.
The curve held across all three platforms, across all 13 socio-political topics, and across platforms ranging from mainstream social media to extremist imageboards. The platform differences produced different absolute levels of engagement, but the underlying curve shape was consistent.
“Moral language often travels widely online, but does more moral content always correspond to higher engagement?” the researchers asked at the outset. Their answer was unambiguous: no.
Why the curve exists: what dense moral language does to a reader
The researchers propose several mechanisms for why moral density beyond the sweet spot reduces engagement rather than amplifying it.
The first concerns attention. When moral language is distributed at moderate density across a post, moral cues direct the reader’s attention selectively to the most important arguments and framings. When moral language saturates every phrase, attention disperses across too many evaluative cues simultaneously. No single argument receives the focus that would provoke a response. The post becomes a wall of outrage rather than a pointed argument.
The second concerns information structure. Posts that moralize without restraint tend to sacrifice logical structure for moral intensity. The argument becomes harder to follow, the specific claim harder to extract. Prior research on social media engagement has found that informational clarity is one of the strongest predictors of whether people engage, share, or respond. Dense moral language can displace that clarity, producing what the researchers describe as moral “gibberish” in which the emotional register is high but the propositional content is low.
The third concerns audience saturation. Readers of political social media are already exposed to extremely high volumes of moral content daily. A post that is entirely saturated with moral language reads as indistinguishable from hundreds of other morally saturated posts, blending into the background noise rather than standing out. The marginal impact of additional moral language eventually reaches zero and then turns negative.
What the Reddit data shows specifically
The Reddit dataset of over one million posts deserves particular attention because it allows analysis across subreddits with very different community norms and topic focuses.
Reddit’s internal culture has its own relationship with moral language that differs from Twitter in specific ways. Reddit’s upvote and downvote system, combined with subreddit-specific rules and norms, creates a context in which overly moralistic posts frequently generate critical responses rather than engagement, as community members push back against what Reddit culture sometimes calls “moralizing” or “lecturing.”
The same inverted U-curve appeared in the Reddit data. Moral loading increased engagement, consistent with Reddit communities being willing to engage with genuinely moral questions. But moral density beyond the threshold reduced engagement, consistent with Reddit’s cultural pushback against posts that read as sermons. The effect was statistically significant across all 13 topic areas.
This means that the Reddit users who report disliking what they perceive as preachy content are expressing a response that aggregates into measurable engagement data. Moralizing posts get fewer upvotes, fewer comments, and less cross-posting. The cultural intuition and the data point in the same direction.
Implications for political communication
The finding has direct practical implications for anyone who creates or analyzes political content online, which in the social media era means virtually everyone.
Activists and advocates who have been told that moral framing is the key to online influence need to add a caveat: moral framing up to a point. Beyond that point, moral language becomes a tax on engagement rather than a boost to it. The most morally saturated climate content, the most morally saturated gun control content, the most morally saturated immigration content all performed worse than moderately moral content on the same topics.
This does not mean abandoning moral framing. The data show clearly that posts with no moral relevance perform worse than posts with moderate moral engagement. The optimal strategy is not to remove moral language but to calibrate it, to select the moral framing that most efficiently communicates the stakes without overwhelming every word with moral charge.
For researchers studying political polarization, the finding suggests a possible mechanism for why the most extreme content, which tends toward maximum moral saturation, may not spread as far as conventional wisdom suggests. The posts that perform best are not the most outraged ones. They are the ones that engage with moral stakes while maintaining enough argumentative clarity to generate genuine engagement.
What the study cannot establish
The analysis is correlational within each platform. The researchers measured the association between moral language density and engagement in the existing corpus of posts, which means the direction of causality cannot be definitively established. Posts might receive lower engagement because they are morally saturated, or morally saturated posts might be produced by a different population of users or in a different context that independently reduces engagement.
The 30% moral density threshold is derived from statistical modeling across 1.6 million posts and represents the population average peak. Individual posts, topics, communities, and time periods may show different optimal thresholds. The curve provides a population-level description of the engagement-moral density relationship, not a precise formula for maximizing any specific post’s performance.
The three platforms studied, Twitter, Reddit, and 8chan, represent a specific slice of social media that skews politically engaged and male. Whether the same curve appears on Instagram, TikTok, Facebook, or other platforms with different demographics, content norms, and algorithmic structures requires separate investigation.
What the study establishes with 1.6 million observations across three platforms and 13 socio-political topics is the existence and consistency of the moral engagement curve. Moral content is not simply more engaging or less engaging. It follows a structured relationship with engagement that rewards the right amount and punishes excess, and that relationship is stable enough to appear across platforms ranging from mainstream social networks to extremist imageboards.
The study, “Saturation of moral language predicts lower content engagement on social media”, was authored by Cristian Candia, Mohammad Atari, Nour Kteily, and Brian Uzzi at Northwestern University and Harvard University, and published September 2, 2026 in Nature Human Behaviour.
Source: Kellogg School of Management, Northwestern University. DOI: 10.1038/s41562-026-02560-y
