In Beauty, bold claims and filmed footage sell more per view than trend hooks and slideshows

1. Introduction

Beauty & Personal Care sellers and creators make many decisions before a video is posted: how to open, which format to use, whether to film or assemble a slideshow. This paper asks which of those choices are associated with more sales per view among top-selling TikTok Shop videos in the category, and what the category's seasonal pattern suggests for the coming months.

The question matters because attention on short-video platforms is long-tailed and early popularity tends to predict later popularity [1], so views alone say little about how efficiently a video converts. Measuring sales relative to views separates the creative choices that sell from those that simply reach people.

Prior work on short-video popularity has focused on predicting views or engagement. Multimodal benchmarks show that content signals carry predictive information [2], while the strongest recent systems find that creator features matter most [3]. We look instead at sales, and at creative attributes a brand or creator can choose.

2. Data

The sample consists of 1,456 top-selling Beauty & Personal Care videos on TikTok Shop observed between 2 September and 1 October 2026. Of these, 234 were analysed in depth. Market-level baselines draw on 3,340 analysed videos across categories, so an index of 1 represents the market average.

For each analysed video we recorded the opening hook type, the format, the production style (filmed live action or photo slideshow), and further attributes used by the predictive model, such as whether the product appears in the first second, whether AI is visibly used, and whether the video was boosted with ads. Not every analysed video received a classification on every attribute, so group sizes differ by dimension.

The window covers one month, and many hook and format groups contain fewer than 15 videos. Monthly seasonal indices describe the category's sales against its own average and are reported separately from the creative analysis.

Daily rankings of top-selling TikTok Shop videos, products and categories in the US were collected from several independent sources and merged into one record per video. Where sources overlap, their sales estimates are cross-checked against each other, and videos whose estimates disagree by more than half are flagged. An AI analyst watched each video and recorded its opening, format, angle, production style, use of AI and hook source, without seeing how the video sold. Sales per view is a video group's revenue per view divided by the market's, shrunk toward the average for small groups, with 90% ranges. This paper covers 2026-09-02 to 2026-10-01: 1,456 videos tracked in Beauty & Personal Care, 234 of them watched and analysed; the market baseline uses all 3,340 watched videos. Predictions come from a logistic model retrained daily and tested on recent weeks it had not seen.

3. Methods

Our main measure is a sales-per-view index: a group's sales per view divided by the market average, so 1 means average and 1.2 means 20% above. Because small groups produce noisy ratios, each estimate is shrunk toward the market mean using empirical-Bayes methods [4]; groups with few videos are pulled more strongly toward 1. We report 90% intervals and a confidence label (low, medium, high) that reflects group size and interval width.

To test whether creative attributes jointly predict top-selling status, we trained a predictive model on 2,667 rows and evaluated it with a time split, training on earlier videos and testing on later ones. Performance is summarised by the area under the ROC curve (AUC), which measures how well the model ranks top sellers above others, where 0.5 is chance [5]. The AUC range is a bootstrap interval [6]. Model drivers are reported as odds ratios.

The model was not restricted to analyses that never saw sales (cleanOnly is false). Some attribute labels may therefore have been produced with knowledge of performance, which could inflate the AUC and the driver odds ratios.

4. Results: Openings

Two hooks dominate sales. Bold-claim openings (33 videos) carry 22% of sales and sell at 1.21× the market per view (90% range 0.99–1.48); relatable-POV openings (35 videos) carry 21% and sell at 1.18× (0.97–1.43). Both are high-confidence estimates whose ranges sit mostly above 1. Problem call-outs (27 videos) carry 17% of sales at 1.06× (0.86–1.32), close to average.

At the low end, trend-or-sound hooks (15 videos) sell at 0.68× per view (0.50–0.92), the only hook whose range lies wholly below average. Visual pattern-breaks (0.73×, 0.51–1.05) and test-or-challenge openings (0.76×, 0.54–1.07) also point below average, but their ranges reach 1.

Price-or-deal openings show the highest point estimate at 1.31× (0.84–2.03), but rest on 7 videos and are low confidence. Social proof, questions, straight-into-demo and result-first openings each have 8 or fewer videos and ranges that straddle 1; we treat them as unresolved.

5. Results: Formats and production

Demo or tutorial (49 videos, 29% of sales) and talking head (37 videos, 26%) are the largest formats and sell close to average, at 1.04× (0.87–1.24) and 1.05× (0.86–1.29). Reviews or testimonials (1.12×, 0.85–1.48), voice-over B-roll (1.09×, 0.78–1.52) and before-and-after (1.04×, 0.77–1.40) are similar, with ranges spanning 1.

Slideshows stand out on the downside: 21 videos, 11% of sales, at 0.73× per view (0.56–0.95). The production split tells the same story. Filmed live action (134 videos) accounts for 88% of sales at 1.09× (0.97–1.21), while photo slideshows (22 videos) account for 12% at 0.74× (0.57–0.94).

Unboxing videos show the highest format estimate, 1.43× (1.03–1.98), with a range above 1, but this rests on 7 videos and is low confidence. The predictive model points the same way on production: filmed live action has the largest driver odds ratio (2.54), and no visible AI is associated with higher odds (1.71). Other drivers include before-and-after (1.66) and boosting with ads (1.59), while an education angle (0.59) and showing the product in the first second (0.62) are associated with lower odds. The model's AUC of 0.64 (range 0.57–0.70) indicates modest ranking ability, and because it was not trained only on sales-blind analyses, these odds ratios should be read cautiously.

6. Results: The season ahead

Against its own average (100 = a normal month), Beauty & Personal Care sales run below normal in October (89) and November (88), then rise to 110 in December. January (96) and February (92) are again below normal.

The strongest months in the index are March (119), June (118) and April (113), with September at 111. July (86) and August (87) are the weakest. The category therefore does not peak only in the holiday period; spring and early summer are at least as strong as December.

7. Discussion

For creators and brands, the evidence points to direct, filmed openings. Bold claims and relatable POVs both carry large shares of sales and sell above average per view, while trend-or-sound hooks sell well below it. Slideshow formats and photo-slideshow production sell at about three-quarters of the market rate per view. Within filmed formats, the common choices (demo, talking head, review, before-and-after) sit close to average, which suggests the opening and production style separate performance more than the format label does.

These results complement prior work. Popularity research finds that creator features matter most for reach [3], and our model's modest AUC is consistent with creative attributes explaining only part of sales outcomes. Long-tailed attention [1] means a few videos drive much of the sales, which is why per-view yield rather than raw sales is the more comparable measure across groups.

Alternative explanations remain. Hook and format choices may proxy for creator size, product price or ad spend; boosting with ads is itself associated with higher odds in the model. Slideshows may be used more often for lower-priced or less-established products. The one-month window falls in a single season, and labels used by the model may not be fully independent of sales.

8. Conclusion

Among top-selling Beauty & Personal Care videos on TikTok Shop, bold-claim and relatable-POV openings sell at about 1.2× the market per view, trend-or-sound hooks at 0.68×, and slideshows at about 0.73×. Filmed live action carries 88% of sales.

The seasonal index points to a softer October and November before a December rise, with spring and early summer the strongest periods. Given the short window, small groups for several hooks and formats, and a model of modest accuracy not limited to sales-blind analyses, these findings are best used as testable priors rather than rules.