AI UGC vs filmed UGC: how to evaluate the evidence

1. Introduction

The short answer for anyone deciding between AI-generated and filmed creator content: in these US TikTok Shop observations, fully AI-generated videos sold at higher sales per view than the market average, but they are a very small group, and filmed live action still carries almost all of the sales. The evidence supports testing AI formats alongside filmed content in your own category. It does not support replacing creators, and it says nothing about cost.

The question matters because brands and creators must choose how to spend limited production budgets, and claims about AI content are easy to make and hard to check. Observational data on advertising and content is hard to read causally, because who makes a video, what is sold and how it is promoted all differ between groups [1]. Prior work on short-video popularity also finds that attention is long-tailed and that creator features matter most [2, 3], which is a reason to be careful about attributing differences to the format alone.

This paper therefore reports what we observed, with sample sizes and uncertainty for each group, and separates those findings from our recommendations. The observations cover US TikTok Shop only and should not be read as describing all UGC everywhere.

2. Data

The window runs from 19 September to 18 October 2026. We tracked 51,843 videos across 29 categories, and our AI analyst watched and analysed 9,371 of them in depth. "Watched" is a count of videos analysed, not of views.

For each analysed video the analysis recorded how much AI was visible (for example fully AI-generated, AI voice-over, AI shots mixed with real footage, no visible AI, or unclear) and the production style (for example filmed live action, AI-generated footage, AI avatar presenter, photo slideshow, mixed media). The group counts in the tables below are smaller than the number analysed in depth, so some analysed videos are not in these groups. Shares of sales refer to the sales of watched videos only.

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, 9,371 of them watched and analysed.

3. Methods

Sales per view is a group's revenue per view divided by the market's revenue per view, so 1 is the market average. Estimates for small groups are shrunk toward the average by empirical Bayes [4], which stops a handful of videos from producing an extreme figure. We report 90% intervals, computed on the log scale.

Share of sales is the group's share of the sales of watched videos. Confidence labels (high or medium) are those attached to each group in our analysis. The AI-use and production labels come from what the AI analyst saw in each video. We have not validated them against creators' own disclosures, so some misclassification is possible.

Each interval compares a group with the market average, not with another group. We have not run a formal test of the difference between, say, fully AI-generated and filmed live action.

4. Results: how much AI a video shows

Fully AI-generated videos (103 videos, 1% of sales) sold at 1.6× the market's sales per view, with a 90% range of 1.4–1.9× and high confidence. Videos with no visible AI (6,056 videos, 81.8% of sales) were at 1.0× (0.98–1.02×). AI voice-over (1,278 videos, 16.3% of sales) was at 0.99× (0.94–1.03×), indistinguishable from the average.

Two groups carry little information. AI shots mixed with real footage has 12 videos and 0.1% of sales, with medium confidence and a range of 0.69–1.40× that spans the average. The unclear group (57 videos, 0.7% of sales) was lower at 0.73× (0.60–0.90×).

So the only AI-use group clearly above the average is also tiny in its share of sales. The higher sales per view is a rate, not a volume: this group accounted for 1% of the sales we observed.

5. Results: production style

Filmed live action (6,711 videos) accounted for 89.4% of sales at 1.0× (0.999–1.037×). AI-generated footage (91 videos, 0.9% of sales) sold at 1.5× (1.3–1.8×) with high confidence. AI avatar presenters were highest at 2.1× (1.6–2.8×), but this rests on 16 videos and 0.2% of sales, with medium confidence, so we treat it as a hypothesis to test rather than an estimate to plan on.

Below the average were photo slideshows (585 videos, 8.4% of sales) at 0.8× (0.78–0.89×) and mixed media (97 videos, 1% of sales) at 0.8× (0.66–0.95×). Not every non-filmed format sold better: the pattern is specific to AI-generated footage and avatars.

The intervals show how precisely each group is estimated against the market. They do not account for differences in creators, products, categories or paid promotion between groups.

6. Discussion

For brands and creators, the observations suggest that AI-generated footage and avatar presenters are associated with higher sales per view than the market average in this window, while AI voice-over on its own is not. The AI groups are small, however, and we cannot say they generalise to other categories, products or periods. The size of the filmed live-action group means its estimate is the most precise.

Several alternative explanations remain. Creators who adopt fully AI formats may differ from the rest in audience, posting volume or paid reach, and creator features matter strongly in short-video prediction [3]. The products sold with AI formats may differ. The AI analyst's labels may contain errors. A small group of videos can also be dominated by a few successes in a long-tailed distribution [2]. Observational comparisons like this one are a weak basis for judging returns [1]. We also hold no production-cost evidence, so we cannot say whether AI content saves money.

Recommendation, not a finding: a practical test plan. (1) Pick one product and one category, and write down the decision you want to make. (2) Make a filmed version and an AI version of the same brief, with the same offer, hook and call to action. (3) Record production cost and time for each, including edits and revisions. (4) Post from comparable accounts or creators, with the same paid budget or none, over the same days. (5) Compare sales per view and revenue per unit of cost, and decide in advance how many videos per arm you need before reading any result. (6) Check platform labelling and disclosure rules directly with the platform, as this paper does not cover them. (7) Repeat before changing your main approach.

7. What this means for brands and founders

• Treat AI formats as a test, not a switch: fully AI-generated videos sold at 1.6× the market's sales per view (90% range 1.4–1.9×), but on 103 videos and 1% of sales.

• Don't expect AI voice-over alone to change results: across 1,278 videos it sat at 0.99× the market (0.94–1.03×), with 16.3% of sales.

• Keep filmed content in the mix: live action carried 89.4% of watched-video sales at about the market average (1.0×), so it is the base to compare any AI test against.

• Hold avatar presenters as a hypothesis: they showed 2.1× (1.6–2.8×) but rest on only 16 videos with medium confidence, so test before planning around them.

• Be wary of defaulting to photo slideshows: they were associated with 0.8× the market's sales per view (0.78–0.89×) across 585 videos.

8. Conclusion

In US TikTok Shop observations from 19 September to 18 October 2026, fully AI-generated videos (1.6×, range 1.4–1.9×) and AI-generated footage (1.5×, range 1.3–1.8×) were associated with higher sales per view than the market average. These groups made up 1% and 0.9% of sales. Filmed live action remained the base at 1.0× and 89.4% of sales, and AI voice-over was at the average.

The evidence supports running controlled tests of AI formats next to filmed content, with costs recorded. It does not support a claim that AI replaces creators, that AI saves money, or that small-group results will hold elsewhere.