ONCE Social Short-Form Video Marketing Research · Social Short-Form Video Marketing Research 06/10
Research article / literature comparison and creative analysis. Research cutoff: October 5, 2026.
A tool-light video appearing in a recommendation feed is not the same event as a colleague forwarding it into a work chat. In the first case, a platform gives the content an opportunity to be seen. In the second, a person chooses the next audience for the content, often with an implicit judgement: “this may be useful to you,” “this is like you,” or “we will understand this.”
Brands often call both events “diffusion” and then use share counts to judge whether the content has transmission power. The problem is that a share count compresses the content, sender, recipient, opportunity to see, and network position into one number. It describes how many actions occurred, but not why they occurred. By itself, it certainly does not establish incremental sales or long-term brand value.
Our central argument is that a short video becomes transferable not merely because it is stimulating, but because a sender judges it to be valuable to a particular recipient, useful for self-expression or relationship maintenance, and capable of surviving the path along which it travels. We therefore propose a four-layer model: recipient value, sender identity, relational fit, and transmission path. This is ONCE's organisation of the evidence, not a validated universal law.
Separate four events that are usually called diffusion
Algorithmic exposure, public reposting, private forwarding, and community participation can all expand reach, but they are not the same behaviour.
| Observable event | What it first indicates | What it does not establish |
|---|---|---|
| Algorithmic recommendation | The system gave content a chance to be seen | A user was willing to endorse it |
| Public share or repost | A sender made the choice visible to an audience | Recipients watched, understood, or bought |
| Direct-message or group forwarding | A sender selected a relationship and situation | Public engagement metrics captured it fully |
| Duet, remix, or participatory response | A user adopted the content as expressive material | The original brand information survived intact |
This distinction extends the conclusion of the first article in this series on recommendation and brand goals: exposure is conditional. A platform first influences who has an opportunity to see; viewers then decide whether to stay, respond, or forward. Public share counts are shaped by the content, initial exposure, audience composition, interface, and the sender's position in the network.
“A earned 10,000 shares while B earned 3,000” is therefore not a complete comparison. We also need to know how many eligible exposures each version received, how many distinct people forwarded it, whether transmission was public or private, and what recipients did next. Without a denominator, the share count rewards both content and opportunity. Without a path, a repost may be a public statement rather than an effective recommendation.
A sender does not press a button; a sender chooses the next audience
Alixandra Barasch and Jonah Berger's 2014 Journal of Marketing Research article, Broadcasting and Narrowcasting: How Audience Size Affects What People Share, reports six studies comparing communication to a large audience with communication to one person. When a specific recipient was salient, participants focused more on the other person and shared content useful to that recipient. In broader broadcasting, self-presentation played a larger role.
This classic research predates today's short-form recommendation environment. It is not a direct test of Douyin messages, WeChat Channels group sharing, or TikTok reposts. It does not provide a rule that private forwarding is always more effective. It provides a still-important mechanism candidate: the more concrete the recipient, the more sharing can become a relational judgement rather than a judgement about content alone.
The same product film can perform very different social tasks. Sent to a colleague, it may recommend a work method. Sent to a relative, it may reduce purchase risk. Reposted publicly, it may express professional interest. Sent to an enthusiast group, it may invite joint evaluation. If the creative team asks only where the film is most stimulating, it misses a better question: can viewers quickly identify who should receive it and why?
This is also why the claim that “useful content gets shared” is too simple. Information value is not a fixed property inside a video. A specification table may be meaningless to a stranger and highly valuable to someone actively comparing products. Usefulness has a recipient, a task, and a time. Making those legible is designing transferability; adding “send this to someone who needs it” at the end is not.
Emotion can propel sharing without giving the brand a direction
Jonah Berger and Katherine Milkman's 2012 article, What Makes Online Content Viral?, combines a three-month dataset of New York Times articles with experiments. Transmission was not explained by positive versus negative valence alone. High-arousal emotions such as awe, anger, and anxiety were associated with more transmission, while a low-arousal emotion such as sadness was associated with less. Practical utility, interest, and placement were also considered.
The material was news and email sharing, not today's shoppable feed video. The research supports arousal as one possible contributor to transmission. It does not support a rule that brands should manufacture anger because anger spreads. More sharing does not automatically mean stronger brand linkage, better product comprehension, or more purchases.
William J. Brady and Jay J. Van Bavel's 2025 Journal of Experimental Psychology: General article, Social Identity Shapes Antecedents and Functional Outcomes of Moral Emotion Expression, reports five preregistered experiments with 2,498 participants. According to the public abstract, moral-emotional language increased intentions to share political messages. Stronger partisan identification and ideological extremity predicted sharing of such messages. Evidence that brief identity-salience manipulations increased sharing was mixed. Sharers could be perceived as more strongly identified by their ingroup while also appearing less open-minded and less worthy of conversation to an outgroup.
The point is not to import a political experiment into a commercial script. It is to retain the cost the experiment reveals: an expression can become more transmissible within a group while becoming less discussable across groups. Sharing is not an unconditionally positive outcome. Controversy, humiliation, or factional hostility may enlarge a count while placing the brand inside a narrower and more antagonistic network. Without measuring brand safety, audience composition, and longer-term response, “provokes emotion” cannot be renamed marketing value.
An influential repost may first change the opportunity to spread
Takuro Niitsuma, Mitsuo Yoshida, Hideaki Tamori, and Yo Nakawake's 2025 open-access Scientific Reports article, Prestige Bias Drives the Viral Spread of Content Reposted by Influencers in Online Communities, analyses more than 55 million posts and 520 million reposts from Twitter, now X. The researchers reconstruct virtual timelines to estimate opportunities to encounter reposts and define a cascading repost probability: further reposts divided by estimated viewed reposts.
They find that content reposted by high-influence users is more likely to be reposted again. In the Japanese-language main data, the highest-influence group, approximately one percent of users, accounted for more than half of estimated views and reposts in secondary spread. The analysis also uses early time windows, mixed-effects models, and a one-day English dataset.
This is not a randomised brand experiment. Virtual timelines are reconstructed from following relationships and chronology. They cannot show precisely who actually saw a repost, and private accounts are unavailable. Parent links in cascades also require inference. The main data cover one month of Japanese posts, while the English robustness data cover only one day. The safer interpretation is that sender status, following networks, and opportunities to see jointly shape diffusion. Observed sharing cannot be attributed to content alone.
This is why follower count, repost count, and “ability to persuade target buyers” are not synonyms. An influential user can place content in a larger network and may also increase the conditional probability of further reposting. Whether the product was understood, the brand was correctly recalled, and recipients belonged to the commercial audience must still be measured separately. For the production and disclosure questions surrounding commercial speakers, see the previous article on four audits of creator trust.
Four layers of transferability
From this evidence, we propose four falsifiable questions for a script review, edit review, or test plan.
| Layer | Creative question |
|---|---|
| Recipient value | Will the recipient save a step or understand a difference, or merely receive an emotion? |
| Sender identity | Does forwarding make the sender look helpful, informed, amusing, intrusive, gullible, or overly promotional? |
| Relational fit | Can the viewer imagine a specific person and situation rather than an abstract “everyone”? |
| Transmission path | What information survives public reposting, direct messaging, group sharing, or remixing, and what brand cues disappear? |
These are not four scores to add. Strong recipient value may still fail if forwarding carries a high identity cost. Public expression may be identity-rich but leave the product interchangeable. A low-friction path does not make the content useful.
ONCE's published OLIGHT EDC tool-light short-form case provides material for a creative thought experiment. We read the case-page text and visible page information. We did not conduct complete-film frame-by-frame or sound analysis, and the page provides no publicly verifiable sharing or sales results. The case places the light on a desk and in actions such as inspecting a dark area, reading, and writing. Hands, paper, and localised illumination turn functions into visible acts.
From the transferability perspective, the same material can support at least three different hypotheses. A colleague-oriented version could emphasise temporary desktop inspection and note-taking, testing recipient value. An EDC-community version could emphasise object relationships and a new colour, testing identity expression. A broader version could use the movement from pocket to desktop, testing an entry point that requires less category knowledge. These are new creative hypotheses, not tests reported by the case and not claims about past performance.
Useful version production is not adding three different “share now” end cards. It means preserving different recipient cues, product actions, and brand resolutions before the shoot while holding product facts constant. ONCE's social and short-form video production solution shows how multiple openings, product actions, and platform versions can be planned in one production. The service page describes production scope; it is not evidence for this article's research claims.
Test whether a video deserves to be transferred instead of waiting for virality
The following is an experiment proposal. It has not been conducted.
Produce two versions for the same product. Version A uses a broadly high-arousal opening to create surprise. Version B, with the same duration, product facts, visual quality, captions, and action path, makes a recipient and task legible—for example, a colleague who repeatedly needs to find something in a dark space. The recipient cue need not be literal on-screen copy. It can be expressed through relationships, context, and action.
In an advertising or controlled-seed environment that can maintain reasonably comparable exposure, preregister the primary outcome as unique sharers per 1,000 eligible exposures, not total shares. Supporting outcomes can include correct product comprehension, unaided brand recall, qualified recipient visits or enquiries, and the costs of hides, reports, or negative response. If a platform provides only aggregate shares, record the unobservability of private destinations. Do not assume link copying, public reposting, and private messages are the same behaviour.
Network experiments add another complication: one person's treatment can affect a neighbour, who affects the next layer. Zahra Fatemi, Jean Pouget-Abadie, and Elena Zheleva's 2024 ICWSM paper, Cascade-Based Randomization for Inferring Causal Effects under Diffusion Interference, explains why cascade interference can bias ordinary individual randomisation and proposes assigning treatment from known cascade seeds through multi-hop neighbours. The paper evaluates its method on real and synthetic network datasets. It is not a brand-video field experiment. It reminds us that the experimental unit for diffusion may need to be a seed group, community, or geography rather than connected individual accounts.
If Version B produces more shares per eligible exposure but weaker brand recall, the recipient cue may have helped relational transmission while removing the brand from the information. If public shares rise but private destinations remain invisible, the team cannot claim that genuine word of mouth expanded. If a cascade follows a single high-influence repost, report the node effect separately from the content effect. A judgement that can fail is more useful than “the algorithm liked it.”
Conclusion: transmission power does not grow inside content by itself
A short video does not become transferable in a vacuum. A platform provides initial opportunity, a sender places it inside a relationship, a recipient decides whether to continue, and network structure amplifies or stops what follows. Emotion, utility, identity expression, influencer status, and social proof may all contribute. None is a viral formula by itself.
Our original contribution is to redefine sharing as a sender choosing the next audience for content, then to use recipient value, sender identity, relational fit, and transmission path to explain why a film may be watched but not forwarded—or widely reposted while leaving little product or brand meaning behind. The limits are explicit: we have no ONCE client network experiment, and public metrics cannot reveal all private transmission. The research spans news, political expression, and repost networks on X. Its mechanisms must be tested again before they are treated as commercial short-form effects.
If you are planning a social-content system that should be seen and transferred accurately, share the product, audience, publication path, and sharing hypothesis with ONCE. We can first write who the film is for, why someone would send it, and what must survive the transfer into the production brief, then decide the openings, actions, and versions.
Sources and reading scope
- Alixandra Barasch and Jonah Berger, Broadcasting and Narrowcasting: How Audience Size Affects What People Share, Journal of Marketing Research 51(3), 2014, DOI 10.1509/jmr.13.0238. Formal abstract, metadata, and study-scope description read.
- Jonah Berger and Katherine L. Milkman, What Makes Online Content Viral?, Journal of Marketing Research 49(2), 2012, DOI 10.1509/jmr.10.0353. Formal abstract, design overview, and conclusion scope read.
- William J. Brady and Jay J. Van Bavel, Social Identity Shapes Antecedents and Functional Outcomes of Moral Emotion Expression, Journal of Experimental Psychology: General 154(6), 2025, DOI 10.1037/xge0001753. Formal bibliographic record and public abstract read; sharing intentions are not rewritten as observed reposting or purchase.
- Takuro Niitsuma, Mitsuo Yoshida, Hideaki Tamori, and Yo Nakawake, Prestige Bias Drives the Viral Spread of Content Reposted by Influencers in Online Communities, Scientific Reports 15, 15282, 2025, DOI 10.1038/s41598-025-98955-4. Open full-text methods, results, and limitations read.
- Zahra Fatemi, Jean Pouget-Abadie, and Elena Zheleva, Cascade-Based Randomization for Inferring Causal Effects under Diffusion Interference, ICWSM 18(1), 2024, DOI 10.1609/icwsm.v18i1.31322. Formal abstract and method scope read; simulation evaluation is not described as a brand experiment.
- ONCE's Chinese and English short-form solution, OLIGHT EDC tool-light case pages, and completed series articles were read on October 5, 2026. The case is used for creative analysis without claiming undisclosed sharing, sales, or experimental effects.
