Tactic · Propaganda
Emotional bait
Content engineered to produce outrage, fear or vindication before it produces a thought. Not proof of falsehood, but a reliable signal the content was optimized to spread rather than to inform.
What am I feeling right now, and is that feeling doing the arguing?
What it looks like
The tell is sequence, not content. A jolt arrives first and a judgment forms behind it, already shaped. The item may be entirely true; what marks it is that the emotional payload does more work than the claim.
This is worth saying plainly because the inference runs one way only. Emotional content is not evidence of falsehood. Real events are often enraging. What the emotion tells you is that this was built to travel, which is a reason to check before sharing rather than a reason to disbelieve.
Why it works
The clearest causal evidence comes from an unusual archive: 22,743 randomized headline tests run by a single publisher, covering roughly 5.7 million clicks across 370 million impressions. For a headline of average length, each additional negative word raised the click-through rate by about 2.3 percent, and positive words lowered it.2 These were controlled experiments, not observations, which is why they carry more weight than the correlational literature.
On the diffusion side, a study of roughly 126,000 story cascades spread by about three million people between 2006 and 2017 found false stories travelling significantly farther and faster than true ones — and identified the emotional signature: false stories drew replies expressing fear, disgust and surprise, while true ones drew anticipation, sadness, joy and trust.3 Automated accounts spread true and false material at the same rate. The difference was people.
The single strongest predictor found so far is not emotion in general but hostility toward a named group. Across 2.7 million posts, each term referring to a political out-group raised the odds of a post being shared by about 67 percent — several times stronger than negative affect or moral-emotional language on their own.5 The mechanism is content-neutral; it is the naming of a group to be angry at that does the work.
click-through rate per additional negative word in a headline of average length. Positive words reduced it.Robertson et al., Nature Human Behaviour, 16 Mar 2023, 22,743 randomized trials, 370 million impressions
increase in the odds of a post being shared, per term referring to a political out-group.Rathje, Van Bavel & van der Linden, PNAS, 29 Jun 2021, n=2,730,215 posts
increase in diffusion per additional moral-emotional word, within ideological networks more than across them.Brady, Wills, Jost, Tucker & Van Bavel, PNAS, 26 Jun 2017, n=563,312 messages
of people across 48 markets say they sometimes or often avoid the news, up from 29 percent in 2017 — the joint highest recorded.Reuters Institute Digital News Report 2025, 17 Jun 2025, 97,000+ respondents
Documented cases
A panic with nothing underneath it, 2019
In early 2019 a claim spread that a menacing figure was appearing in children’s videos and instructing them to harm themselves. It produced school warnings and police statements in more than a dozen countries.
Every relevant safeguarding organization said there was no evidence for it. The Samaritans said they were "not aware of any verified evidence in this country or beyond" linking it to suicide. The NSPCC said it had received more calls from journalists than from concerned parents. The UK Safer Internet Centre called it a myth and noted that children were now frightened of a threat that had not previously existed.6 The coverage generated the harm it reported.
A deliberately bad study, 2015
The journalist John Bohannon ran a study he designed to produce a false positive: 16 participants across three arms, measuring 18 variables, which gives roughly a 60 percent chance that something crosses the significance threshold by chance alone. It was published within 24 hours by a journal that did not review it, and he issued a press release.7
It was picked up by newspapers, magazines and television in several countries. The emotional payload — that a treat you already want is good for you — was the product. The study was the packaging. Bohannon’s sting itself drew criticism for deceiving readers, which is worth noting since this page is about being honest with them.
What to do about it
- Name the feeling before you judge the claim. Outrage, fear or vindication arriving first is a signal about how the item was built, not about whether it is true.
- Restate the claim flatly and see what is left. If the sentence has no content once the emotional language is removed, there was no claim to begin with.
- Look for the out-group term. A post that names a group to be angry at is the highest-sharing form of content yet measured. Treat "they" as a flag.
- Put sixty seconds between reading and sharing. Sharing is reinforced by the response it gets; the delay is what interrupts that loop.
- Treat novelty as a reason for caution. Surprise predicts spread. "I have never heard this before" is not the same as "this is important."
Sources
Every figure on this page comes from one of these. Each entry names the publisher, the date and the sample size, so you can check it rather than take our word for it.
- Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A. & Van Bavel, J. J. — "Emotion shapes the diffusion of moralized content in social networks"PNAS 114(28):7313–7318 · 26 June 2017 · n=563,312 messageshttps://doi.org/10.1073/pnas.1618923114
- Robertson, C. E. et al. — "Negativity drives online news consumption"Nature Human Behaviour 7(5):812–822 · 16 March 2023 · 22,743 randomized trials, 370 million impressionshttps://doi.org/10.1038/s41562-023-01538-4
- Vosoughi, S., Roy, D. & Aral, S. — "The spread of true and false news online"Science 359(6380):1146–1151 · 9 March 2018 · ~126,000 cascades, ~3 million peoplehttps://doi.org/10.1126/science.aap9559
- Brady, W. J., McLoughlin, K., Doan, T. N. & Crockett, M. J. — "How social learning amplifies moral outrage expression in online social networks"Science Advances 7(33) · 13 August 2021 · 7,331 users, 12.7 million tweets, plus experiments with n=240https://doi.org/10.1126/sciadv.abe5641
- Rathje, S., Van Bavel, J. J. & van der Linden, S. — "Out-group animosity drives engagement on social media"PNAS 118(26) · 29 June 2021 · n=2,730,215 postshttps://doi.org/10.1073/pnas.2024292118
- Waterson, J. — "Viral ‘Momo challenge’ is a malicious hoax, say charities"The Guardian, via The Irish Times · 28 February 2019https://www.irishtimes.com/news/world/viral-momo-challenge-is-a-malicious-hoax-say-charities-1.3809573
- Bohannon, J. — "I Fooled Millions Into Thinking Chocolate Helps Weight Loss. Here’s How."Gizmodo · 27 May 2015 · n=16 participants, 18 measured variableshttps://gizmodo.com/i-fooled-millions-into-thinking-chocolate-helps-weight-1707251800
- Reuters Institute — Digital News Report 2025Reuters Institute for the Study of Journalism, University of Oxford · 17 June 2025 · 97,000+ respondents across 48 marketshttps://reutersinstitute.politics.ox.ac.uk/digital-news-report/2025
What this page is not sure about
Every page here carries this section. A site that tells you to check its sources should be the first to say where its own evidence is thin.
- Several of the studies behind this page were built on US political corpora. We cite the mechanisms, which are content-neutral, and deliberately do not repeat their findings about which political groups did what — that is outside what this site will adjudicate.
- The headline experiments are from a single publisher and end in April 2015. They are the best causal evidence available on negativity and attention, and they are one publisher, ten years ago.
- The Brady 2017 diffusion figure is observational, not experimental. It shows association, not that the moral-emotional words caused the sharing.