Scroll through TikTok for a few minutes and you will pass dozens of product pitches dressed up as entertainment. Some you swipe past without a thought. Others make you stop, tap the heart, leave a comment, or save the clip for later. What separates the videos that grab you from the ones you ignore?
A team of researchers set out to answer that question by taking apart the visual and audio ingredients of short video ads and measuring which ones line up with the actions people take. Their work appears in the Journal of Business Research, and it offers evidence that four specific features tend to move viewers toward engaging.
The gap the researchers wanted to fill
Plenty of earlier studies have looked at how short videos shape what people think about brands and whether they buy. Fewer have asked what comes first: which sensory elements actually prompt a viewer to like, comment, save, or share in the first place. That is the puzzle Zhipeng Zhang of Jiangsu University of Science and Technology, working with Keda Qiu of Wuhan University and Yan Ye of Northwest Agriculture and Forestry University, wanted to address.
The challenge is that short video ads pour several kinds of information at the viewer at once. There is what you see, what you hear, and the words being spoken. The researchers wanted to separate these streams and test each one against real engagement data, rather than relying on lab experiments or hand-coded surveys.
They built their analysis on two ideas from existing research. The first, called multimodal theory, holds that people absorb information more richly when it arrives through several channels at once, such as image, sound, and speech together. The second, the elaboration likelihood model, describes two ways people process messages. One is a quick, low-effort route that leans on surface cues and first impressions. The other is a slower, more analytical route that weighs the actual content. Because people often watch short videos to relax and scroll fast, the authors expected that snappy sensory cues would matter a great deal.
Four features, measured by machine
The researchers gathered 900 short video advertisements from TikTok, posted in the first half of 2023. They focused on two product categories: food, which they treated as an “experience” product because you only really judge it after consuming it, and digital products, which they treated as a “search” product because shoppers can look up specifications before buying. To make sure engagement numbers had stabilized, they only used videos posted at least 30 days earlier, with more than 100 likes and 10 comments.
From the first 10 seconds of each video, which prior research suggests carries the heaviest weight, they extracted four features using machine and deep learning tools.
Colloquial expression measured how casual and conversational the spoken language was. The team transcribed the audio and ran it through a language model that scored how informal the phrasing sounded, with chatty sentences scoring higher than stiff, formal ones.
Cadence captured the rhythm of the sound, meaning how much the volume rose and fell over time. They converted the audio into a curve and counted its peaks, with more peaks signaling more dynamic, varied sound.
Colorfulness measured how visually rich and vibrant the colors were, calculated from the pixels in sampled frames.
Visual prominence measured whether one object stood out sharply against its surroundings, using a key-point detection method that flags strong focal points in an image.
They then ran a regression analysis connecting these four features to four engagement actions: likes, comments, collects (saving a video), and shares. They also included control variables such as video length, title length, the influencer’s follower count, and the tone of the top comments.
What the analysis showed
All four features were positively associated with engagement, though not uniformly across every action.
Colloquial, informal speech showed the strongest and most consistent link, tied to more likes, comments, collects, and shares. The authors read this as evidence that casual language lowers the mental effort of watching and builds a sense of closeness with the viewer.
Cadence, the rise and fall of the sound, was also linked to all four actions. The researchers interpret rhythmic audio as a way to grab attention quickly and pull viewers into the video.
Colorfulness was tied to higher engagement across the board as well, consistent with the idea that vivid color stirs an emotional response. Visual prominence was linked to more likes, collects, and shares, but it showed no meaningful connection to comments. The authors suggest commenting takes more deliberate thought than a strong visual alone tends to spark.
Product type and timing shift the picture
The team also tested whether these effects changed depending on what was being sold and when the video went up.
For search products like digital gadgets, colloquial speech and visual prominence carried more weight. For experience products like food, the rhythm of the audio and the colorfulness of the visuals mattered more, at least for shares. The authors connect this to how people evaluate the two product types: shoppers tend to look up hard information about searchable goods, while they rely more on atmosphere and feeling when judging experiences.
Timing mattered too. The researchers split videos into daytime and nighttime releases. Colloquial expression, colorfulness, and visual prominence tended to be more effective during the day, when, they argue, viewers have more mental energy to process content. Audio cadence appeared to land harder at night, which the authors attribute partly to quieter surroundings making sound more noticeable.
A few of the control variables told their own story. Longer titles were associated with lower engagement, while a larger follower count was associated with more. Longer videos drew more likes and comments but fewer saves and shares.
What it means for video marketers
For people making short video ads, the study points to some practical levers. Simple, conversational narration, lively color, a clear visual focal point, and rhythmic audio all line up with stronger engagement. The researchers suggest matching the emphasis to the product: lean on sound and color for experience goods like food, and on plain speech and a sharp visual focus for searchable goods like electronics. They also suggest timing releases to fit the feature, putting color-heavy and chatty content out during the day and rhythmic audio at night.
The authors add that platform recommendation systems could fold these measurable features into their calculations to surface ads more likely to draw engagement.
A few caveats are worth keeping in mind. This is observational work, so it shows associations rather than proof that any one feature causes more engagement. The study covered only two product categories on a single platform during one stretch of time, and it measured engagement actions rather than actual purchases. The researchers also looked at each feature on its own, leaving open how these elements interact when combined, and they note that they did not measure how viewers personally perceive the audio and visuals. Even so, the work offers a structured way to think about why some short video ads make us stop scrolling.
