Every marketing conference has a talk about how to make things go viral. There are books, frameworks, checklists of emotional triggers. A great deal of money changes hands on the premise that this is a solvable problem.
The research is considerably less encouraging, and the reason it's less encouraging is quite interesting.
The cascade problem
Something goes viral through a cascade: person A shares it, a few of their contacts share it, a few of theirs, and so on. Whether that cascade sustains or dies depends on whether the average number of onward shares per viewer exceeds one.
That's the same mathematics as epidemic spread, and it has an awkward property: outcomes are extremely sensitive to conditions early in the process.
If the first few people to see something happen to be well connected, enthusiastic, and posting at a moment when their audience is paying attention, it spreads. If the same content reaches slightly different first viewers at a slightly different time, it doesn't.
This means the outcome depends heavily on factors that are essentially random from the creator's point of view. Two nearly identical pieces of content can have outcomes differing by orders of magnitude for reasons that have nothing to do with the content.
The experimental evidence
There's a well-known line of research on this in cultural markets, where the same set of items was presented to separate independent populations.
The findings were consistent: within each population, some items became enormously popular. But which items became popular differed sharply between populations, and popularity was only weakly predicted by independent quality ratings.
Quality mattered — genuinely poor items rarely succeeded anywhere. But among the reasonably good ones, which became a hit was substantially arbitrary, driven by early social influence within each group.
That's about as clean a demonstration as this field offers, and its implication is uncomfortable for anyone selling virality as a service.
What does predict something
It's not that nothing is predictable. Some factors reliably shift the odds, they just don't determine the outcome.
High-arousal emotion. Content producing awe, anger, anxiety or amusement is shared more than content producing sadness or contentment. The distinction isn't positive versus negative but activating versus deactivating. This finding has held up reasonably well.
Practical utility. People share things that make them look useful to others.
Social currency. Sharing something signals something about the sharer — taste, insider knowledge, values. Content that makes the sharer look good travels further.
Low friction. Obvious and neglected. Short, immediately comprehensible, works without sound, doesn't require context.
Each of these improves your chances. None gets you close to reliability.
The seeding effect
The factor that matters most is arguably who sees it first, and this is where a lot of apparently organic virality turns out to have had help.
Getting content in front of a small number of well-connected accounts early has an outsized effect on cascade probability. This is why influencer seeding works and why so much "spontaneous" viral content had a media budget behind its first few hours.
From outside it's usually impossible to distinguish an organic cascade from a seeded one. Which means a lot of the case studies used to teach virality are teaching the wrong lesson — the content wasn't magic, it was placed.
The survivorship problem
The deepest flaw in most virality advice: it studies successes.
You take twenty things that went viral, look for common features, and write them up as principles. What's missing is the enormous number of pieces of content with exactly those features that went nowhere.
Without the failures, any pattern found in the successes is uninterpretable. And the failures are invisible by definition, because nothing that doesn't spread gets studied.
This is a general problem with reasoning about outliers and it's particularly acute here, given the base rate. The overwhelming majority of content produced with viral intent reaches almost nobody.
The practical conclusion
If you're making things, the honest advice is: increase the number of attempts, make each one as good as you can, and accept that the outcome distribution is extremely skewed and largely out of your control.
That's not what anybody wants to hear, and it's why the alternative advice sells so well.
The one genuinely actionable insight I'd offer is that consistency beats optimisation. Somebody producing decent work regularly will eventually catch a cascade, because they've bought many lottery tickets. Somebody producing one heavily optimised piece has bought one.
Which reframes the question usefully. Not "how do I make this go viral" but "how do I build a practice that survives most things not going viral". That one has real answers.
A note on trying to manufacture it
Deliberate viral marketing has a specific failure mode worth naming: it is transparent. Audiences have spent two decades being marketed to through content, and the detection ability is now quite good. Something that reads as an attempt to go viral generally does not, because the sharing impulse depends on the sharer looking good, and passing along an obvious campaign does not achieve that.
Which produces an awkward conclusion for anybody selling this as a service. The techniques that improve your odds are largely the techniques of making something genuinely worth sharing, and those are not really techniques at all. They are just the work.