What makes small-business videos win: our 2026 study

Small-business videos that beat their own channel's normal views are about twice as likely to have surprise as their main emotion (×2.0), 1.6 times as likely to hold the reveal until the end, and 1.57 times as likely to use original sound rather than a trending track. Videos over 90 seconds show up more often among winners (×1.54), while tips told straight to camera show up less (×0.59). To get there, we collected 293,926 Instagram and TikTok posts over the 12 months to 12 October 2026, built a catalogue of 5,576 small-business and creator channels, and compared 2,608 winning videos with 1,600 ordinary posts from the same channels.

Key findings

How to read ×1.6. A winner is a post that got far more views than its own channel usually gets. ×1.6 means the feature appears 1.6 times as often among winners as among ordinary posts from the same channels. For the reveal at the end, that's 4.3% of winners against 2.7% of ordinary posts. Below ×1 means the feature is rarer among winners. The multiplier includes a small-numbers correction, so it can differ a little from dividing the two shares. Everything below is a reliable finding or a result from our product report, unless it says otherwise.

• Surprise as the main emotion: ×2.0. 3.8% of winners vs 1.9% of ordinary posts.
• A strong hook: ×1.70. Hooks rated 5 out of 5 (214 winners, 142 ordinary posts). A weak hook: ×0.62.
• Reveal at the end: ×1.6. 4.3% vs 2.7%. The product or the best frame is held back until the last seconds.
• Original sound: ×1.57. A trending track: ×0.89, slightly below ordinary posts.
• Videos over 90 seconds: ×1.54. 21.9% vs 14.2%.
• Green-screen commentary: ×1.54. A person talking over a screenshot or an image behind them. 5.7% vs 3.7%.
• Relatable humor: ×1.5. 6.0% vs 4.0%.
• UGC selfie with the product in hand: ×1.49. A phone-style clip where someone holds the product and talks about it.
• Probable: an AI-character story, ×4.39 (0.9% vs 0.2%, a small group of posts), a top-N list video, ×1.32, and raw phone footage, ×1.10.

And the formats that show up less often among winners:

• A tip told straight to camera: ×0.59. 6.8% of winners vs 11.6% of ordinary posts.
• Pretty product footage without a voice: ×0.63.
• Packing-orders ASMR: ×0.43.

Findings by audience

Online stores. The weakest evidence of the four groups: only 1 of 46 store findings is reliable. That one is counters of social proof in the frame ("10,000 orders"): 3.7% of winners vs 1.0% of ordinary posts, ×3.37. In the separate hook check, store videos with a strong hook came out at ×1.95 (a split of the overall ×1.7 result by audience, less certain than the overall number).

Marketplace sellers. Step-by-step tutorials: 25.3% vs 15.8%, ×1.6. A hook that names the viewer's role ("Handmade sellers: ..."): 6.3% vs 2.0%, ×3.0. Videos over 90 seconds: 17.2% vs 9.5%, ×1.79. Pretty B-roll with text on top is rarer among winners: 3.4% vs 7.7%, ×0.44.

Local services and salons. The group with the most reliable findings. An odd first frame: 9.7% vs 4.0%, ×2.37. Real orders on camera: 9.4% vs 5.0%, ×1.84. Over 90 seconds: 28.9% vs 15.0%, ×1.91, while clips under 15 seconds are rarer (12.9% vs 20.5%, ×0.63). Tips to camera fall to ×0.38 (4.7% vs 12.5%), and a before / after story to ×0.6 (6.1% vs 10.2%).

Creators and agencies. Surprise is even stronger here: 8.7% vs 3.5%, ×2.42. A contradiction that breaks the pattern: 14.8% vs 9.8%, ×1.51. Talking about having no money for creative: 13.5% vs 7.5%, ×1.79. Posts about AI tools: 23.2% vs 13.5%, ×1.71, partly because many channels in this group are about AI tools. Posts pitched as "useful for a client" are rarer: 11.1% vs 19.8%, ×0.56.

What didn't matter

Some popular rules turned out to be hygiene, not levers. Winners and ordinary posts do them equally often:

• Text in the hook. 89% of winners and 89% of ordinary posts.
• A face in the first second. 70% vs 71%.
• Readable with the sound off. 84% vs 88%.
• Fast cutting. More than 5 cuts per 10 seconds: 14% of winners vs 16% of ordinary posts.
• Hashtags. 0 to 5 hashtags made no difference.

For these we used values measured by software (on-screen text, faces, cuts), not the AI labels. Keep text in the hook, a face early and a video that works muted: most winners and most ordinary posts already have them. Cutting speed and hashtags made no difference either way. None of these will make a post win on its own.

Engagement benchmarks

Engagement rate (ER) here is likes, comments and shares divided by views. Instagram doesn't give shares, so its ER is likes and comments divided by views. Medians:

• Instagram videos: ordinary posts 2.04% (1,037 posts), winners 3.64% (1,732 posts).
• TikTok videos: ordinary posts 6.45% (562 posts), winners 10.89% (876 posts).
• Facebook Reels: 0.98% across 1,648 Reels from 216 pages, with no split into winners and ordinary posts.

Length works differently by platform. On Instagram, ER grows with length: ordinary posts go from 1.73% under 15 seconds to 3.07% over 90 seconds, winners from 3.22% to 4.44%. On TikTok, winners score best under 15 seconds (13.61%). On Facebook, ER grows from 0.68% under 15 seconds to 1.41% over 90 seconds. Smaller channels tend to have higher ER, so compare yourself with channels of your size.

Methodology

• Data. 293,926 posts collected: 120,742 from Instagram and 173,155 from TikTok (the total also includes 29 of our own test videos). From them we built a catalogue of 5,576 channels (2,127 on Instagram, 3,449 on TikTok): 1,996 local services, 1,895 creators and agencies, 869 marketplace sellers and 816 online stores. Period: 12 months to 12 October 2026.
• What "winner" means. For each post we divided its views by the channel's normal: the median views of the channel's other recent posts in the same format, adjusted toward the median of channels of a similar size. Winners were picked mostly on that ratio, with engagement, comments and views as smaller parts of the score. Instagram and TikTok were never mixed.
• Excluded. Posts marked as ads (13,816) or paid partnerships (4,726), lead-magnet posts ("comment a word and I'll DM you"), pinned posts and Reels under 7 days old.
• Sample. 2,808 winners (2,448 main picks, 160 absolute hits, 200 carousels) and 1,600 control posts, 400 per audience group. Carousels have no control group, so the comparisons in this article use the 2,608 videos. Control posts are ordinary posts from the same channels, with views between 0.7 and 1.3 of the channel's normal.
• Lift. The share of winners with a feature divided by the share of control posts with it, with a correction for small numbers. Confidence intervals come from resampling whole channels, and the results are corrected for testing many features at once.
• Reliability. A finding is reliable when it has at least 15 winners from at least 8 channels, its interval doesn't include ×1, and it repeats in at least 2 types of channels. Weaker ones are probable or hypotheses. Of 267 findings, 43 are reliable, 9 probable and 215 hypotheses. This article uses the first two.
• Labels. Views, likes, comments, length, cuts, faces, on-screen text and loudness were measured. Hook, format, story arc, theme and emotion were labelled by AI that watched each video with sound, and the hook score of 1 to 5 was given without seeing the post's numbers. People audited a sample of 50 labelled videos by hand.

Limitations

• Organic posts, not paid ads. Ads were removed. What wins in a feed is a strong hint for ads, not proof.
• Correlation, not causation. "×2.0" means more common among winners, not that the feature caused the views. With many features tested, a few reliable results may still be chance. Trust the ones that repeat across audience groups.
• AI labels are imperfect. In the manual audit, theme was right 64% of the time, hook type 68% and format 72%. Plain facts like on-screen text and faces were right 98% of the time, and for those we also had measured values.
• Carousels have no control group. So we don't report carousel results here.
• YouTube is not covered. Facebook appears only as an engagement benchmark.
• No saves or shares on Instagram. Instagram doesn't report them, so Instagram ER counts likes and comments only.

How to use this for your ads

1. Spend your effort on the first second. Videos with a strong hook were 1.7 times as common among winners, and those hooks more often hold a contradiction, an unexpected object or something visually odd. Text and a face there are the minimum, not the edge.
2. Save the reveal. Show the product, the finished result or the price at the end, not in frame one.
3. Use your own voice. Winners more often had original sound (×1.57), and less often a trending track (×0.89) or silent pretty footage (×0.63).
4. Show the product in a hand. Phone-style selfie clips with the product in hand were more common among winners (×1.49) than polished showcases.
5. Swap the tip for a story. If you give advice to camera, wrap it in a surprise, a joke or a real order.
6. Test a long cut. In organic feeds, videos over 90 seconds are more common among winners, especially for local services.

Vendy makes the UGC format from one product photo. A UGC video has an AI creator talk to camera about your product, with captions and a script you can edit. The AI video ad generator also makes product showcases and avatar slideshows. Vendy's clips run 3 to 30 seconds, so they suit ads and hooks more than 90-second posts. For more on openings, read video ad hooks, and for formats that work, UGC ad examples.

Cite this study

Questions about the study

What is a winning post in this study?

A post that got far more views than its own channel usually gets. We compare it with ordinary posts from the same channels, so big accounts don't win just for being big.

What does ×1.6 mean?

The feature appears 1.6 times as often among winners as among ordinary posts. For example, 4.3% of winners hold the reveal until the end, against 2.7% of ordinary posts.

Do longer videos really do better?

They are more common among winners: videos over 90 seconds appear 1.54 times as often among winners as among ordinary posts. That doesn't prove length itself brings views. On TikTok, though, winners have their highest engagement under 15 seconds.

Does a face in the first second help?

Not measurably. 70% of winners and 71% of ordinary posts show a face in the first second. The same goes for text in the hook: 89% in both groups.

Can I use these numbers in my article?

Yes. Quote any number with a link to this page, and use the suggested citation above.