Word of mouth is the most trusted and most valuable channel in marketing, while at the same time being the one you can least directly control. The useful skill is not "going viral" on demand, which almost no one can do, but building something worth talking about, understanding what actually gets shared, and measuring it honestly.
The starting fact is about trust. A recommendation from a person you know, or even a stranger's review, carries more weight than the same message paid for by the brand, which is why word of mouth (WOM) is treated as the most valuable channel a company can earn and the hardest to buy. The research splits into two eras. The pre-digital sociological tradition established that influence flows through personal ties and everyday conversation rather than straight from media to individual (Katz and Lazarsfeld, 1955). The post-2000 explosion of online WOM, reviews, shares, ratings, comments, then made talk visible and measurable at a scale the earlier researchers could only sample.
That visibility produced the field's most useful applied finding, about what actually gets shared. Studying which New York Times articles made the most-emailed list, Berger and Milkman found that the driver was not simply how positive or negative a piece was but how much arousal it triggered. Content that provoked high-arousal emotions, awe, excitement, amusement, anger, anxiety, was passed on more; content that produced low-arousal states like sadness or contentment was passed on less, even when people liked it. Practically useful and surprising content also travelled well (Berger and Milkman, 2012). Berger later distilled a working checklist for practitioners, summarised as STEPPS: things spread when they confer Social currency (sharing them makes the sharer look good), are tied to everyday Triggers (cues that bring the thing to mind), carry high-arousal Emotion, are Public (visible and imitable), have Practical value (worth passing to someone), and travel wrapped in a Story (Berger, 2013).
A second consensus point reshaped how WOM is measured. Godes and Mayzlin showed that the dispersion of conversation, how widely talk is spread across different communities, predicted outcomes (in their case television-show ratings) better than the sheer volume of talk. A thousand mentions inside one tight community are worth less than the same thousand scattered across many (Godes and Mayzlin, 2004). That distinction (spread versus loudness) is now built into serious social-listening.
Finally, the deliberate mechanism, referral programmes, has real if modest support. When a firm rewards existing customers for bringing in new ones, the referred customers tend to be more valuable and to stay longer than customers acquired other ways, though the size of the effect varies a great deal by category (Schmitt et al., 2011). Referral is the one part of WOM a company can actually engineer.
The central controversy is the word viral itself. The trade-press image of virality, a single clever piece of content spreading person to person through many generations, each infected person infecting several more, turns out to describe almost nothing. When Goel and colleagues mapped the actual diffusion structure of a billion online events, they found that the vast majority of even the most popular content spread by broadcast: one large source, a media outlet, a celebrity, a platform's front page, reaching many people more or less at once, with very little deep person-to-person chaining. Genuine multi-generation cascades were rare, and sustained self-propagating growth (each sharer reliably recruiting more than one new sharer) rarer still (Goel et al., 2016). The exponential "viral coefficient" that growth decks promised in the early 2010s essentially does not occur in the wild for organic content.
This connects to the influentials debunk covered in L5-04: the idea that WOM strategy means finding a few high-influence people and infecting them was dismantled by Watts and Dodds, who showed influence is structurally distributed and cascades usually start with ordinary people, not designated hubs (Watts and Dodds, 2007). Most of the WOM-consultancy industry was built on the story their work took apart.
The deepest correction is about causation. WOM is real and large, but it is mostly a downstream consequence of something else: a product good or distinctive enough to be worth mentioning, a triggering occasion that brings it up, a genuine reason to talk. It is largely an outcome, not an independent lever you pull. The mountain of "viral campaign" case studies in the trade press is badly survivorship-biased: we study the handful that took off and never see the thousands of near-identical attempts that vanished, so we over-learn from winners and mistake luck for method.
The headline "what gets shared" evidence is narrower than its fame suggests. The core arousal finding comes largely from one context (emailing text articles from one US newspaper), and STEPPS is a practitioner synthesis, a memorable and plausible checklist, not a validated causal formula that has been shown to make arbitrary content spread. eWOM measurement is noisy: sentiment and volume are read imperfectly by machines, and platforms differ. And the whole area is strongly culture- and platform-bound: sharing norms, which emotions are shareable, and the very architecture of spread differ across a WeChat or LINE or WhatsApp ecosystem versus an open feed like X, so findings from one platform and culture are priors to check, not laws to export.
Can virality ever be engineered before the fact, or only explained after it? How much of the referral-programme advantage is genuine causal lift versus selection, referrers simply passing the brand to people already like themselves and already likely to buy? And how do the mechanics of sharing change as messaging moves from open public feeds to closed encrypted groups, where the spread is invisible to measurement?
The usable core: stop trying to "go viral," build something distinctive and worth talking about, understand what gets shared and design for it, measure the spread of talk rather than its volume, use referral mechanics where the category rewards them, and never fake the talk itself.
Start by demoting virality from a goal to an outcome. A campaign brief that lists "make it go viral" as a deliverable is asking for the one thing the evidence says you cannot reliably command (Goel et al., 2016). The productive brief instead asks two answerable questions: is the product distinctive and good enough that people have a reason to mention it, and is there a trigger, a frequent, everyday cue that brings it to mind at the moment sharing is natural? Berger's most useful single idea is the trigger: a product linked to something people encounter often gets talked about often, which is why an occasion-linked brand (the Friday treat, the morning ritual, the match-day drink) out-shares a cleverer but un-triggered one. That is also where this connects to distinctiveness and mental availability (L5-02): the brands that get talked about are the ones already easy to bring to mind.
Use STEPPS as a design checklist, not a guarantee (Berger, 2013). Before launch, ask of the thing itself: does sharing it make the sharer look smart, kind, or in-the-know (social currency)? Is it tied to a recurring trigger? Does it carry a high-arousal emotion rather than a flat pleasant one (Berger and Milkman, 2012)? Is the behaviour public and imitable, can people see others doing it? Does it carry practical value worth passing on? Is there a story that carries the brand along with it, rather than a story people enjoy while forgetting who made it? Each yes improves the odds; none of them buys a cascade. Jonah Berger, whose Contagious (2013) popularised this, is a Wharton professor who has built a substantial applied practice around it, and the framework is worth using precisely as he frames it, as a way to improve shareability at the margin, not a recipe for guaranteed spread.
Measure the right thing. Track dispersion, how many distinct communities, segments, or networks the talk is reaching, not just the raw count of mentions (Godes and Mayzlin, 2004). A number that is large because one community is shouting is a weaker signal than a smaller number spread widely, and it often flags a bubble rather than broad appeal. And treat your social-listening volume as an outcome metric that tells you whether the product and its triggers are working, rather than as a lever you can crank directly.
Where WOM can be engineered is referral (Schmitt et al., 2011). Referred customers genuinely tend to be worth more and churn less, so a well-built referral programme can pay for itself, but three disciplines matter. Design the incentive to fit the motive: a double-sided reward, something for the referrer and something for the friend, tends to feel like generosity rather than a bounty on one's contacts, which is why the most-cited growth cases (the classic being Dropbox's give-storage-get-storage mechanic, widely reported to have driven a large share of its early signups) rewarded both sides with the product itself. Test empirically per category, because the effect size swings widely and what works for a subscription service may do nothing for a grocery brand. And run it against a holdout to separate genuine lift from selection, since referred customers may simply resemble the good customers who referred them (an open question the research flags).
The dual-use edge here is sharp and increasingly policed. Because earned talk is so trusted, there is a standing temptation to fake it: astroturfed "grassroots" enthusiasm, paid reviews posing as organic, undisclosed sponsored posts, bought followers and engagement. It can move numbers briefly. It also carries a real and growing cost: consumer-protection regulators in a number of markets now treat fake and undisclosed-paid reviews as unlawful, platforms purge inauthentic engagement, and the reputational damage when manufactured WOM is exposed lands precisely on the trust that made WOM valuable in the first place. The honest test is simple and worth applying to any WOM tactic: would this survive your customers knowing exactly how it was produced? Genuine referral, disclosed partnerships, and a product worth mentioning pass it. Manufactured talk does not, and the whole value of the channel rests on the difference.
The single most robust WOM lesson in politics is that trusted-peer contact beats broadcast. A message from a friend, neighbour, or fellow member carries a credibility that no paid advertisement matches, which is why relational organising, equipping supporters to talk to people they already know, consistently outperforms cold outreach (this is the mechanism behind the ground-game findings in L5-01). Build the campaign so that real supporters have something concrete and shareable to say to people who trust them, and give them the triggers and occasions to say it.
The honest cautions are two. First, most "movements" that appear to spread virally are, structurally, broadcasts amplified by media and a few large accounts (Goel et al., 2016), so a strategy that assumes organic person-to-person cascade will usually underdeliver, and organising for reach and repetition is the surer path. Second, the manufactured-grassroots temptation is even stronger here than in commerce and even more corrosive. An astroturfed hashtag or a staged "spontaneous" groundswell that no real community was already producing tends to read as inauthentic and can discredit the cause it was meant to serve, the same failure mode as the manufactured ritual in L5-05. Genuine WOM in politics is downstream of a message people actually believe and want to repeat.
Public communication runs on the same physics, and two applications matter most. For behaviour-change and public-health messaging, seed the message through trusted local messengers, community figures, clinicians, teachers, faith and civic leaders, whose word carries the peer-trust premium, and measure success by dispersion across communities rather than by total impressions, because a message that saturates one already-reached group while never entering another has failed at exactly the job that matters (Godes and Mayzlin, 2004). For misinformation and rumour, recognise that false content often spreads well precisely because it is engineered, deliberately or by selection, to carry high-arousal emotion, especially anger and fear (Berger and Milkman, 2012), and that it too usually spreads by broadcast amplification more than by deep cascade (Goel et al., 2016), which means countermeasures aimed only at "super-spreaders" will miss the broadcast sources doing most of the work. The platform-and-culture caveat is decisive for any government working across markets: how people share, and on which closed or open platforms, varies enormously, so a communication plan built for one country's media ecosystem should be treated as a starting hypothesis, not a template.
Three habits. Treat virality as an outcome, not an objective, and put the effort upstream into a distinctive product and a genuine trigger rather than into chasing a cascade you cannot command. Measure the spread of talk across communities, not just its volume, and read that talk as a thermometer of whether the product is working rather than as a dial you turn. And keep the talk real, engineer referral honestly and disclose partnerships, because the entire value of word of mouth is the trust that fakery destroys.
.png)