UGC voiceovers for TikTok Shop: what the data can tell us

1. Introduction

Short answer for the reader's decision: in our data, a voiceover is a reasonable default for TikTok Shop videos, but it is not the top performer. Voice-over videos sold at about the market average, while speaking on camera sold somewhat better per view. We recommend testing both rather than assuming one is best. This paper covers US TikTok Shop observations only; it does not speak to all UGC everywhere, to other languages, or to any particular voice.

The question matters because audio is an element creators can vary between video versions, yet brands and creators often choose it by habit. Voiceover production also adds cost and a casting decision, so it helps to know how voiceovers compare with on-camera speech, music and trending sounds, and which tones sit alongside higher or lower sales per view.

Prior work on short-video popularity shows that attention is long-tailed and that early popularity predicts later popularity [1], and that creator features tend to matter most in state-of-the-art popularity prediction [2]. Measuring the return to advertising and content from observational data is also hard [3]. These references do not address voiceovers specifically, so we treat our results as descriptive associations, not as a test of what audio choices cause.

2. Data

We tracked 51,843 videos selling on TikTok Shop in the US between 19 September and 18 October 2026, across 29 categories. Our AI analyst watched and analysed 9,285 of them in depth; this is a count of videos, not of views.

For each analysed video we recorded the audio style (speaking on camera, voice-over, music only, trending sound, product sounds) and the tone (for example enthusiastic, calm, urgent). Sales and views for each video give the revenue and attention figures used below. We did not record the language spoken, the voice's gender or accent, or production quality, so none of those can be compared here.

Daily rankings of top-selling TikTok Shop videos 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, hook source, length, pacing, what appears in the first second and how it is presented (people on screen, voice-over, setting, UGC or studio feel), 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 2026-09-19 to 2026-10-18: 51,843 videos tracked, 9,285 of them watched and analysed.

3. Methods

The main statistic is sales per view. For a group of videos (for example all voice-over videos), we divide the group's revenue per view by the market's revenue per view, so 1 means the market average and 1.13 means 13% more revenue per view than average.

Small groups give noisy ratios, so estimates are shrunk toward the average by empirical Bayes [4]. Intervals are 90% ranges computed on the log scale. Where we say a group's share of sales, we mean its share of revenue among the videos analysed in depth. We do not adjust for category, price, creator or paid promotion, and we give each audio style and tone separately, not in combination.

4. Results: audio style

Voice-over was the most common high-selling audio style: 3,341 videos carried 40% of sales among those analysed in depth. Their sales per view were 1.05× the market (90% range 1.02–1.07), close to average. Speaking on camera covered 2,869 videos and 35.5% of sales, with higher sales per view at 1.13× (1.10–1.16). The two ranges do not overlap, so on-camera speech sold somewhat better per view than voice-over in this window.

Audio without speech sold below average. Music-only videos (845 videos, 9.3% of sales) sold at 0.90× (0.85–0.95), and trending sounds (1,240 videos, 14.6% of sales) at 0.80× (0.76–0.83). Product sounds were rare, at 62 videos and 0.5% of sales, and sold at 0.40× (0.33–0.48); this group is small, so treat it as indicative.

All five groups had high confidence under our rule, but these are averages. A voice-over share that is the largest by sales does not mean it is the best per view, since volume and yield are different measures.

5. Results: tone

An enthusiastic tone dominated: 5,402 videos carried 67.5% of sales, at 1.08× the market (1.06–1.11). An informative tone sold at about average, 0.99× (0.95–1.04), on 1,116 videos and 12.4% of sales.

The highest sales per view appeared in two small groups. Urgent videos sold at 1.59× (1.33–1.90) but numbered only 79 and carried 0.9% of sales. Videos in a sceptic-convinced tone sold at 1.34× (1.18–1.53) on 97 videos and 1.5% of sales. The ranges exclude 1, but small groups mean these estimates rest on little evidence and may not hold at scale.

Lower-selling tones included funny (0.85×, 0.77–0.93; 315 videos), calm (0.77×, 0.73–0.81; 1,108 videos, 11.3% of sales) and emotional (0.62×, 0.56–0.68; 241 videos). We did not split tone by audio style, so we cannot say whether, for example, a calm voice-over differs from a calm on-camera delivery.

6. Discussion

For a brand deciding on voiceover production, our data suggest that voiceover is a workable default but that on-camera speech was associated with higher sales per view, and that non-speech audio (trending sounds, music only) sold below average. A tone that sounds enthusiastic was both common and above average; calm and emotional tones were associated with lower sales per view. These are associations. Alternative explanations include which creators, categories, prices and levels of paid promotion tend to use each style, and which products suit a voiceover (for example, products shown without a person). Observational data of this kind cannot separate those from the effect of audio itself [3].

Recommendation, not a finding: a practical test plan. (1) Pick one product and one script. (2) Produce two or three versions that differ only in audio: voiceover, on-camera speech, and optionally a different tone. (3) Keep the visuals, length, offer, price and posting conditions as similar as you can, and give each version comparable reach. (4) Compare revenue per view over the same period, and repeat across several products before deciding. (5) Note the language and voice you used, because our data do not cover them.

This paper does not show a universally best voice, and does not extend to languages or markets we did not measure. The small urgent and sceptic-convinced groups are better treated as ideas to test than as established advantages.

7. What this means for brands and founders

• Test voiceover against speaking on camera: in this window, on-camera speech sold at 1.13× the market per view (1.10–1.16) and voice-over at 1.05× (1.02–1.07), so run both versions on your own product.

• Think twice before relying on trending sounds alone: they sold at 0.80× the market per view (0.76–0.83), and music-only videos at 0.90× (0.85–0.95).

• Lean toward an enthusiastic delivery as a starting point: it sold at 1.08× the market (1.06–1.11) and covered 67.5% of sales, while calm (0.77×) and emotional (0.62×) tones sold below average.

• Treat urgent and sceptic-convinced tones as small experiments: they sold at 1.59× and 1.34× but on only 79 and 97 videos, so the evidence is thin.

8. Conclusion

In US TikTok Shop videos between 19 September and 18 October 2026, voice-over carried the largest share of sales among the videos analysed in depth (40%) and sold at about the market average per view (1.05×), while speaking on camera sold at 1.13×. Trending sounds and music-only audio sold below average, and an enthusiastic tone was both common and modestly above average.

The results are descriptive, cover one market and one 30-day window, and do not control for product, creator or paid promotion. They are best used to shape a controlled test of audio style and tone for your own products.