What the research says about why content gets shared
Two findings from the word-of-mouth literature sit awkwardly against how most content gets briefed. Sharing is driven by arousal rather than by how positive something is. And the thing that makes content interesting is not the thing that keeps people talking about it.
Both have direct consequences once AI makes it cheap to produce a lot of content quickly.
The full citations are in the Word of Mouth & Social Transmission research index. The papers below are peer-reviewed marketing science, several using natural experiments rather than surveys.
Arousal moves content, not pleasantness
Berger and Milkman (2012) analyzed every article the New York Times published across a three-month window and measured what predicted sharing. Positive content was more viral than negative, and that is the part usually quoted. The stronger effect was physiological arousal.
High-arousal emotions increased transmission whether the valence was positive or negative. Awe traveled. So did anger and anxiety. Sadness, a low-arousal deactivating emotion, reduced sharing.
This is an uncomfortable result for brand content, most of which is engineered to be agreeable and calm. Agreeable and calm is precisely the quadrant with no transmission mechanism. A piece can be well made, accurate, on brand and completely inert.
The practical read is not to manufacture outrage. It is to notice that if a concept produces no activating response in the person who reads it, sharing is not a realistic goal for that piece, and it should be measured on something else.
Interesting buys a spike; cues buy duration
Berger and Schwartz (2011) ran the sharpest test of the "make it more interesting" brief. Working from everyday conversation data across more than 300 products, a large field experiment and a controlled lab study, they compared the drivers of immediate word of mouth against ongoing word of mouth over months.
More interesting products got more immediate word of mouth. They did not get more ongoing word of mouth.
What sustained conversation was different: products cued more often by the environment, or more publicly visible. Being encountered regularly in ordinary situations kept a product in conversation long after novelty had worn off.
| Goal | What the evidence supports | What it does not support |
|---|---|---|
| Immediate spike | Novelty, high-arousal emotion | Calm, agreeable content |
| Sustained talk | Frequent environmental cueing, visibility | Interestingness alone |
| Travel to new audiences | Dispersion across communities | Raw volume in one community |
That table is a planning tool. A content program aimed at sustained conversation and built entirely on novelty is working against its own goal.
Volume and spread are different measurements
Godes and Mayzlin (2004) studied online conversation as a proxy for word of mouth and found that dispersion across communities carried explanatory power in their model separately from conversation volume. The same quantity of talk means something different depending on whether it sits in one community or spreads across many.
They also named a problem worth carrying into any measurement discussion: word of mouth is both an outcome of sales and a predictor of them. Reporting that treats conversation purely as a leading indicator is describing half a loop.
Chevalier and Mayzlin (2006) got closer to causation on the review side by comparing the same books across two retailers, which separates the effect of reviews from the quality of the book. An improvement in a book's reviews at one site was associated with an increase in that site's relative sales.
One detail from that paper deserves more attention than it gets. Reviews at both sites skewed overwhelmingly positive. The baseline is not neutral, so an average rating reads as a weak signal rather than a middling one.
What this changes for AI-produced content
The instinct with a fast content tool is to produce more concepts and hope one catches. The research suggests that strategy is buying spikes, and spikes are the effect that decays.
The durable alternative is less exciting and more demanding. Attach content to situations the audience actually encounters on a repeating basis, then stay recognizable across all of it, so the cue keeps leading back to you. That is a consistency problem rather than a creativity problem.
It is also where volume can work against a brand. Fifty pieces that each look slightly different train no cue at all. Copper Sun holds brand context across projects so the recognizable signals survive the volume, which is the condition environmental cueing depends on.
Ask two questions of a content plan. Does anything here produce an activating response? And is there a recurring situation this content attaches itself to? A plan that answers no to both will produce work, and not much conversation.
Frequently Asked Questions
Is positive content more shareable than negative content?
Berger and Milkman (2012) found positive content more viral overall, but valence is the weaker half of the result. Arousal is the stronger driver. High-arousal emotions increase sharing whether positive, such as awe, or negative, such as anger and anxiety. Low-arousal sadness decreases it. A positive but calm piece is poorly positioned to travel.
Why do interesting campaigns stop being talked about?
Berger and Schwartz (2011) tested this directly and found that interesting products got more immediate word of mouth but not more ongoing word of mouth over subsequent months. Interestingness is a novelty property and novelty decays. Sustained conversation came from environmental cueing and public visibility, which is a matter of being reminded rather than being remarkable.
Do online reviews cause sales or just correlate with them?
Chevalier and Mayzlin (2006) approached the causal question by comparing the same titles across two competing retailers, which separates review effects from book quality. Improvements in a book's reviews at one site were associated with higher relative sales at that site. The design is why the paper is cited for causation where most review studies can only show correlation.
How should a content team apply this?
Split the plan by goal. Pieces meant to travel need an activating idea, and pieces meant to sustain conversation need to attach to a recurring situation. Then hold the brand signals steady across both, because cueing only works if the cue keeps pointing at the same brand. The consistency argument is developed further in what the research says about how brands actually grow.