Essay

How to go viral: what the research and 26 viral spikes show

Published · Updated 18 min read

Virality is the spread of a piece of content, a product or an idea from person to person, so that each share brings in new people who share it again. The word borrows from epidemiology: in theory, an idea "goes viral" the way an infection does, passing through chains of contacts rather than arriving from one big broadcast.[3] If you want to know how to go viral, the research gives a clear and slightly sobering answer. You can raise the odds that people share your work. You cannot make spreading likely, and once a spike arrives, it leaves fast.

This essay brings together three bodies of evidence. The first is the best known field study of what makes people share, built on almost 7,000 New York Times articles.[1] The second is large-scale data on how sharing really travels across networks, from studies covering seven online platforms and about a billion Twitter events.[2][3] The third is our own study: we measured the rise and fall of public attention for 26 documented viral products, memes and shows, using a decade of Wikipedia pageview data. We close with a framework, the Spread and Hold matrix, and a step-by-step method any marketing team can use.

Idea in brief

The problem. Most shared content goes nowhere. Across seven online platforms, between 73% and 96% of sharing chains never went past the first person, and fewer than 1% grew past seven people.[2]

Why it happens. Sharing is driven by feeling and usefulness, and most business content offers neither. Content that sparked anger or awe was more likely to be widely shared; sad content was less likely.[1] Even when content does spread, it mostly spreads through a few big broadcasts, not long chains of friends.[3]

The gap (our data). Across 26 viral spikes, public attention halved in a median of 10 days. One year later, the median topic held 2.2% of its peak attention. Products that people kept using held on far longer than memes.

The answer. Design for spread and for hold. Build an idea that sparks a strong feeling or solves a real problem, put it in front of large audiences on purpose, and give people a reason to come back once the spike is over.

Key findings

The essay in 2 minutes
  1. Finding 134%

    High-arousal emotion spreads.

    A one standard deviation increase in the anger an article evoked raised its odds of making the New York Times most e-mailed list by 34%. For awe the figure was 30%. Sadness lowered the chance by 16%.[1]

  2. Finding 2<4%

    Almost all sharing stops at the first step.

    In seven online domains, 73% to 96% of cascades showed no diffusion at all, and fewer than 4% reached further than one step from the person who started them.[2]

  3. Finding 30.025%

    Big cascades are very rare.

    In a sample of about a billion Twitter diffusion events, only 0.025% grew to at least 100 people, roughly 1 in every 4,000.[3]

  4. Finding 410 days

    Attention halves in about ten days.

    Across our 26 viral spikes, the median topic lost half its peak attention in 10 days and fell below 10% of its peak in 64.5 days.

  5. Finding 52.2%

    A year later, almost nothing is left.

    The median topic kept 2.2% of its peak attention after a year. Only 3 of 26 kept 10% or more.

  6. Finding 628 vs 6

    Products hold, memes fade.

    Products and apps took a median of 28 days to lose half their attention. Memes and challenges took 6.

Chapter 1 · 1 min

What virality is

Researchers use "viral" in two ways, and the difference matters for marketers. In everyday speech, viral means very popular, very fast. In network research, viral describes how something became popular: through many generations of person-to-person sharing, rather than through one large broadcast such as a TV spot, a celebrity post or a front-page feature.[3] Goel and colleagues call this second idea "structural virality". Their main finding is that popular content usually grows mostly through broadcasts, with some person-to-person spread mixed in.[3]

So when a business asks how to go viral, it is usually asking two questions at once. How do we make something people want to pass on? And how do we get it in front of enough people for the passing on to matter?

Terms used in this essay

TermMeaning
CascadeThe chain of shares that starts with one person (the seed) and grows as others adopt or reshare.
SeedThe first person or account to post a piece of content in a cascade.
Structural viralityA measure of how much a cascade grew through many generations of sharing rather than one big broadcast.[3]
ArousalHow activated or fired up an emotion makes a person feel. Anger, awe and anxiety are high-arousal. Sadness is low-arousal.[1]
Peak multipleOur measure: attention on the peak week, divided by normal daily attention a month earlier.
Attention half-lifeOur measure: the number of days after the peak until daily attention falls below half the peak level.

Broadcast and viral spread compared

Broadcast spreadViral spread
How it travelsOne source reaches many people at oncePeople pass it to people, over many steps
Typical exampleA TV ad, a big creator post, a front-page storyA tool or joke passed around a team, then a company, then an industry
What drives itReach of the sourceEach sharer's reason to pass it on
How commonAccounts for most popularity online[3]Rare in pure form[2][3]
What it means for a businessPick the right big audiencesMake each share worth the sharer's reputation

Chapter 2 · 2 min

The evidence, part 1: what makes content shareable

The best known study of sharing is "What Makes Online Content Viral?" by Jonah Berger and Katherine Milkman of the Wharton School, published in the Journal of Marketing Research in 2012.[1] They captured every article on the New York Times home page between 30 August and 30 November 2008, a total of 6,956 articles, using a crawler that visited the page every 15 minutes. Their measure of virality was whether an article made the paper's most e-mailed list. Of the articles in the data, 20% did.[1]

The team coded each article for positivity, emotionality and specific emotions, and controlled for how long and where it was featured on the home page.

Figure 1. Anger, awe and usefulness predict sharing. Sadness does not.

Anger (+1 SD)
+34%
Awe (+1 SD)
+30%
Practical value (+1 SD)
+30%
Interest (+1 SD)
+25%
Anxiety (+1 SD)
+21%
Time at top of home page (+1 SD)
+20%
Emotionality (+1 SD)
+18%
Surprise (+1 SD)
+14%
Positivity (+1 SD)
+13%
Sadness (+1 SD)
-16%

Change in probability of making the list (%), -20 to 40

Percentage change in the fitted probability of an article making the New York Times most e-mailed list, for a one standard deviation increase in each characteristic. Source: Berger and Milkman, Journal of Marketing Research, 2012, Figure 2.

The pattern is about arousal, not only about good or bad feelings. Anger, anxiety and awe all made sharing more likely, while sadness made it less likely. The authors put the size of the effect in practical terms: one standard deviation more anger raised the odds of making the list by 34%, which they say is equivalent to spending an extra 2.9 hours as the lead story on the website, nearly four times the average time articles spent in that spot.[1]

Look at the third bar. Practical value matched awe at 30%. For a business, that may be the most useful result in the paper. A piece that helps someone do their job better is as shareable as one that amazes them.

Follow-up lab experiments, with 45 to 49 participants each, tested cause and effect: stories that evoked more amusement or more anger were more likely to be shared, and a sadder story was less likely.[1]

Limits of this evidence

Berger later turned this research and related work into the book Contagious, which sets out six principles that drive sharing.[6] The Hive's notes on Cialdini's Influence cover social proof, the related idea that people share what they see others sharing.

Chapter 3 · 2 min

The evidence, part 2: most sharing never spreads

The Berger and Milkman study tells you what makes a piece more likely to be shared. It does not tell you what happens next. For that, the strongest evidence comes from studies that traced every step of sharing across whole networks.

Sharad Goel, Duncan Watts and Daniel Goldstein studied seven very different online domains, from Yahoo! messaging products and an online game to news stories and videos shared on Twitter.[2] They traced each cascade back to its seed and counted how far it went.

Figure 2. In every domain, most cascades never left the first person

Twitter Videos
96%
Twitter News Stories
95%
The Secretary Game
94%
Yahoo! Voice
92%
Yahoo! Kindness
89%
Friend Sense
84%
Yahoo! Zync
73%

Cascades with no diffusion (%), 0 to 100

Share of cascades that consisted of the seed alone, with no further adoption. Source: Goel, Watts and Goldstein, ACM Conference on Electronic Commerce 2012, Table I.

In every domain, most cascades showed no diffusion at all: from 73% for Yahoo! Zync to 96% for videos shared on Twitter. The mean cascade size was between 1.1 and 1.4 people. In every domain, fewer than 1% of cascades included more than seven people, and fewer than 4% went more than one step from the seed.[2]

A later study by Goel, Ashton Anderson, Jake Hofman and Watts looked at about a billion diffusion events on Twitter, covering news stories, videos, images and petitions.[3] About 99% of adoptions came either from the seeds themselves or from their direct followers. Only 0.025% of cascades reached 100 people, roughly 1 in every 4,000. The authors found that structural virality was typically low and stayed low regardless of size, which suggests that popularity is driven mainly by the size of the largest broadcast in a cascade.[3]

This changes the question a business should ask. A piece that reaches many people usually does so because a large account, a publication or a platform feature carried it. Person-to-person sharing mostly works one step out from those big moments.

Tip: Plan the broadcast, not only the post. List the five accounts, newsletters, trade publications or associations whose audience your buyers trust, and design the piece so it is worth their sharing.

Limits of this evidence

If you want a team to check whether your content gives a stranger any reason to share it, the Brave New audit looks at what a new visitor sees and whether it travels.

Chapter 4 · 3 min

Original research: how fast viral attention fades

The studies above describe what drives sharing and how it spreads. We wanted to answer a different question that matters for budgets: once something goes viral, how long does the attention last?

On 3 November 2026 we pulled the daily English Wikipedia pageviews for 26 well documented viral topics from the Wikimedia pageviews API, which records human page views back to July 2015. Wikipedia views are a public, comparable proxy for how many people were looking something up on a given day. We grouped the topics into three types: products and apps (12), memes and challenges (9), and media and entertainment (5). For each topic we found the peak week inside its breakout year, then measured how quickly attention fell. The full method is at the end of this essay.

Finding 1: attention halves in about ten days

Figure 3. Days until attention fell to half its peak, by topic

Threads (social network)
3
Consumption of Tide Pods
4
Tesla Cybertruck
4
Storm Area 51
5
Yanny or Laurel
5
Barbenheimer
6
Chewbacca Mask Lady
6
Hawk tuah
6
Pesto (penguin)
6
Pokémon Go
6
Bernie Sanders mittens meme
7
Skibidi Toilet
7
Tiger King
8
Moo Deng
12
Mannequin Challenge
13
Grimace Shake
14
Squid Game
14
Clubhouse (app)
16
Wordle
25
Fidget spinner
31
Among Us
32
Labubu
38
BeReal
66
Dubai chocolate
66
Old Town Road
69
ChatGPT
79

Median: 10 days: 10

Days until attention halved

Lime: product or app. Navy: meme or challenge. Gray: media or entertainment.

Days from the peak (7-day average of daily English Wikipedia pageviews, including redirects) until attention first fell below half the peak. Source: Brave New Media analysis of Wikimedia pageviews data, 3 November 2026.

The median topic lost half its peak attention in 10 days. Twelve of the 26 (46%) lost half within a week. The fastest was Threads, the social app, which dropped to half its launch-week attention in 3 days. The median topic fell below 10% of its peak in 64.5 days.

Pokémon Go is the clearest case of how steep a spike can be. In the week of its peak in July 2016, its Wikipedia attention ran 129.6 times its normal level of the month before. Six days later, it had halved.

Finding 2: products that people use hold attention longer

Figure 4. How fast attention fades, by type of viral topic

Half of peakBelow 10% of peak
10
64.5
All events
28
146.5
Product or app
6
49
Meme or challenge
8
37
Media or entertainment

Median days

Medians across topics. Source: Brave New Media analysis of Wikimedia pageviews data, 3 November 2026.

Products and apps took a median of 28 days to lose half their attention, more than four times as long as memes and challenges (6 days). Products also took a median of 146.5 days to fall below 10% of their peak, against 49 for memes.

A meme gives people one thing to look at. A product gives them something to do: a game to play again, a tool to use at work, a shelf item to hunt for. That matters for any business, because a product is what most companies have to sell.

Finding 3: a year later, almost nothing is left

Figure 5. Share of peak attention left after 30 days and after one year

Day 30After one year
24.4%
2.2%
All events
50.1%
3.8%
Product or app
18.8%
1.7%
Meme or challenge
13.6%
1.2%
Media or entertainment

Share of peak attention (%), 0 to 60

Average daily attention from day 335 to day 365 after the peak, and from day 28 to day 34, each as a share of the peak 7-day average. Source: Brave New Media analysis of Wikimedia pageviews data, 3 November 2026.

After a month, the median topic held about a quarter of its peak attention. After a year, it held 2.2%. Only three topics kept 10% or more a year later: ChatGPT (17.8%), Dubai chocolate (14.7%) and BeReal (10.4%). For the median topic, the seven days around the peak drew 15.9% of all the attention it received in the year starting a month before the peak.

Finding 4: half of these topics barely existed a month earlier

For 13 of the 26 topics, Wikipedia did not yet have a live article, which we defined as one drawing at least 100 views a day, a month before the peak. The topic went from unknown to famous in weeks. For the 13 that did have a baseline, the median peak week ran 2.5 times normal attention, with a wide range: from 1.2 times for Dubai chocolate to 267.1 times for Squid Game.

What our data does and does not show

Wikipedia views measure people looking something up, not sales, shares or sign-ups. The sample is 26 topics we chose because they were widely described as viral, so it says nothing about the far larger number of pieces that never spread. And English Wikipedia leans toward English-speaking readers. Read the numbers as a picture of how public attention behaves after a spike, not as a forecast for any one campaign.

Chapter 5 · 1 min

A framework: the Spread and Hold matrix

The research points to two separate jobs. Spread is about getting attention: an idea that sparks feeling or solves a problem, carried by the right broadcasts. Hold is about keeping it: a reason to come back after the spike. Most viral advice covers only the first.

Low hold (one-off interest)High hold (repeat use or need)
High spreadFlash. Big spike, fast fade. Typical of memes and challenges, which in our data halved in a median of 6 days.Engine. Spike that leaves a base behind. Products and apps in our data halved in a median of 28 days, and the strongest kept over 10% of peak attention a year later.
Low spreadNoise. No spike and no base. Where most content sits: 73% to 96% of cascades never pass the first person.[2]Staple. Steady, small, loyal audience. Useful to existing customers but rarely found by new ones.

Where each move sits, with the evidence

MoveBuildsEvidence
Spark a high-arousal feeling (awe, surprise, the useful kind of worry)SpreadAnger +34%, awe +30%, sadness -16% in the NYT data[1]
Make it useful enough to forwardSpread and holdPractical value +30%[1]
Plan the big broadcastsSpreadPopularity driven mainly by the largest broadcast[3]
Give people something to do, not only to seeHoldProducts halved in 28 days, memes in 6 (our data)
Make it recognisably yoursHoldA spike nobody links to you builds someone else's memory

Most business content sits in the Noise box. The goal is not to become a Flash. The goal is to move toward Engine: content that spreads because it is useful or striking, attached to something people return to.

Chapter 6 · 3 min

How to go viral: a step-by-step method

Use this for one piece of content or one campaign, not a whole content plan.

Step 1: Decide what a spike is worth

Write down what one new customer is worth and who that customer is. An industrial supplier does not need a million views. It needs the few thousand engineers, buyers and plant managers who could ever place an order. The Hive entry on mental and physical availability explains why reaching future buyers before they are shopping pays off later.

Step 2: Generate many ideas, then cut hard

Write at least 20 ideas, and cut any you cannot explain in one line or that only existing customers would care about. This is where we build AI flows for clients: one brief turned into dozens of angles and headlines in minutes, so the team spends its time judging, not drafting.

Step 3: Name the feeling or the use

For each idea, write down the emotion it should cause or the job it does for the reader. Aim for awe, surprise, laughter or a useful worry, such as a cost the reader did not know they were paying. If the honest answer is "mild interest", the research says it will not travel.[1] The Hive entry on somatic markers covers how feelings feed into decisions, and the debate around that idea.

Step 4: Run the stranger test

Ask: if you saw this by chance, from a company you had never heard of, would you send it to a colleague? "It's well made" is not enough. "This saves our team a day a month" is.

Step 5: Choose the broadcasts

List the accounts, newsletters, trade titles, associations and partners whose audiences match your buyers. Shape the piece so each of them has a reason to carry it: a data point their readers have not seen, a tool their members can use, a story about their industry.

Step 6: Build the hold before you launch

Decide what a new visitor can do after the spike: a calculator to come back to, a monthly test, a free tool, a newsletter, a product page that answers the next question. Our data suggests this is what separates an Engine from a Flash.

Step 7: Make it recognisably yours

A piece that spreads without your name on it builds someone else's attention. Use the colours, voice or recurring format people already link to you. The Hive explains how distinctive brand assets do that job, and why familiar signals make new work easier to process, as covered in mere exposure.

Step 8: Plan for the half-life

Expect attention to halve in days, not months. Have the follow-up content, the sales outreach and the versions for other markets ready before launch, so you use the peak week instead of reacting to it. An AI flow can draft those versions in a day, with a person checking each one. If you want one that takes your best ideas to more markets, start with the Brave New audit.

A made-up example

This case is made up, but it's typical. Say a 40-person fastener manufacturer usually posts "New product line now available!" and reaches only its existing customers.

It runs the method. The new idea is a monthly blind pull test on camera: its bolt against two unbranded imports, with nobody knowing which is which until the end. The feeling is suspense and a useful worry about imports. The broadcasts are two trade publications and an engineering association newsletter. The hold is a public results table updated every month, so engineers have a reason to come back.

Which moves to make first

MoveEffortLikely impactWhen
Name the feeling or use for each pieceLowHighNext piece
Stranger test on every ideaLowHighNext piece
List the five broadcasts that reach your buyersLowHighThis month
Make every piece recognisably yoursLow to mediumMediumOngoing
Build one reason to return (tool, test, table)MediumHighThis quarter
Prepare follow-up and market versions before launchMediumMediumEach launch

Chapter 7 · 1 min

What this means for leaders

For manufacturing and supply-chain businesses, the research takes pressure off and adds discipline. It takes pressure off because nobody can reliably manufacture a viral hit, and the data shows that most of the attention from one fades within weeks. It adds discipline because the parts that can be controlled are clear: content that is useful or striking, planned placement in front of large trusted audiences, and a product or tool that gives people a reason to return.

The growth case is in the hold. A spike that reaches a new country, a new industry or a new buyer group is only worth what you capture before it fades: the sign-ups, the sales conversations, the list of companies that looked. That is why we build AI flows that prepare every market version and follow-up before the peak, so a small team can act on a two-week window in ten markets instead of one. For a weekly look at how attention and money move in industrial markets, read The Brave New Business Journal.

Chapter 8 · 1 min

Common mistakes

Chapter 9 · 2 min

Methodology

Sample. 26 topics widely described as viral since mid-2015, chosen for having an English Wikipedia article: 12 products and apps (Pokémon Go, Fidget spinner, Wordle, Grimace Shake, Dubai chocolate, Labubu, Among Us, Tesla Cybertruck, ChatGPT, Threads, Clubhouse, BeReal), 9 memes and challenges (Mannequin Challenge, Yanny or Laurel, Storm Area 51, Moo Deng, Hawk tuah, Bernie Sanders mittens meme, Consumption of Tide Pods, Pesto the penguin, Chewbacca Mask Lady) and 5 media and entertainment topics (Skibidi Toilet, Squid Game, Tiger King, Old Town Road, Barbenheimer). The sample begins in mid-2015 because the pageviews API starts on 1 July 2015.

Data. On 3 November 2026, a script requested daily pageviews from the Wikimedia REST API (English Wikipedia, all access methods, agent type "user", which excludes known bots) for each article and for every redirect pointing to it, then added them together, because traffic often sits under an older title. It identified itself with the user agent "BraveNewResearch/1.0" and a link to brave-new.com.

Measures. We smoothed each series with a centred 7-day average so a single odd day could not set the peak. The peak is the highest 7-day average inside each topic's breakout window (usually its breakout calendar year; the windows are listed in the published data). Attention half-life is the number of days until the 7-day average first fell below 50% of the peak; we measured the same for 10%. Day-30 attention is the average of days 28 to 34 after the peak; one-year attention is the average of days 335 to 365. The baseline is the median daily views from 37 to 8 days before the peak, counted only when all 30 days had data and the median was at least 100 views a day. Group figures are medians across topics.

Limits. Pageviews measure interest, not shares or sales. The sample is chosen, not random, and leans to English-speaking audiences. Some topics had earlier waves outside the window we measured. Twenty-six topics is a small sample; treat the group medians as direction.

The data behind every figure in this essay is published alongside it as CSV files, with the scripts that produced it.

Chapter 10 · 1 min

Frequently asked questions

How do you go viral on purpose?

You cannot guarantee it, but you can raise the odds. Make content that sparks a strong feeling such as awe, surprise or anger, or that is useful enough to forward. Then plan where it will be broadcast, because research on a billion Twitter events found that popularity is driven mainly by large broadcasts, not long chains of sharing.

What makes content shareable?

In a study of 6,956 New York Times articles, content that evoked high-arousal emotions was more likely to be widely e-mailed. More anger raised the odds by 34% and more awe by 30%. Practical value showed an effect as large as awe. Sad content was less likely to be shared. Useful, interesting and surprising content also helped.

How long does viral attention last?

Not long. In our study of 26 viral topics using Wikipedia pageviews, attention halved in a median of 10 days and fell below a tenth of its peak in a median of 64.5 days. A year later, the median topic held 2.2% of its peak attention. Products held attention longer than memes.

How rare is it for content to go viral?

Very rare. Across seven online platforms, 73% to 96% of sharing cascades never went past the first person. In a sample of about a billion Twitter events, only 0.025% of cascades reached 100 people, roughly 1 in 4,000. Most popular content reached its audience through a few large broadcasts.

Can a marketing agency help a business go viral?

An agency cannot promise a viral hit, and you should be wary of one that does. It can help you raise the odds and use the attention you get. Brave New Media is a marketing agency that builds AI flows to test ideas, prepare versions for new markets and follow up during the peak, with experience in supply chain and manufacturing.

Your next step

This week, take the last three pieces of content you published and score each one on two questions. What feeling or use does it offer a stranger? What would bring that stranger back a month later? Any piece that scores zero on both is Noise. Rewrite one of them with a named feeling, a named broadcast and a reason to return.

If you want a marketing agency to turn this into a system that keeps producing shareable ideas and carries them to new customers and markets, start with the Brave New audit or see Brave New Media.

Brave New Media is a marketing agency that builds AI flows to help companies scale their ideas, grow their customer base and expand into new markets, with experience across supply chain and manufacturing.

See also

References

1. Berger, J. and Milkman, K. L. "What Makes Online Content Viral?" Journal of Marketing Research 49(2), 2012, pp. 192-205. faculty.wharton.upenn.edu

2. Goel, S., Watts, D. J. and Goldstein, D. G. "The Structure of Online Diffusion Networks." Proceedings of the 13th ACM Conference on Electronic Commerce, 2012, pp. 623-638. 5harad.com

3. Goel, S., Anderson, A., Hofman, J. and Watts, D. J. "The Structural Virality of Online Diffusion." Management Science 62(1), 2016, pp. 180-196. 5harad.com

4. Wikimedia Foundation. Pageviews REST API, per-article daily views, English Wikipedia. wikimedia.org

5. Brave New Media. Attention decay of 26 viral topics, Wikipedia pageviews, 3 November 2026. Data and scripts published with this essay.

6. Berger, J. Contagious: Why Things Catch On. jonahberger.com