What sells in Food & Beverages on TikTok Shop: direct claims, demos and mid-priced products

1. Introduction

Brands selling food and drink on TikTok Shop, and the creators who promote them, face two practical questions: which products and price points to back, and which kinds of video to make. Food and drink is an everyday, low-consideration purchase, so the opening seconds of a video and the way the product is shown may matter more than long explanations. We ask what the category's top-selling videos and products have in common.

Prior work shows that online attention is long-tailed and that early popularity predicts later popularity [1]. Recent short-video prediction research finds that creator features carry most of the signal, with content features adding less [2, 3]. Live selling is argued to build trust in social commerce [4], and affiliate commissions are set to balance seller margin against referral effort [5]. Price endings can shift retail sales [6], and online retail sells a long tail of niche products beyond bestsellers [7]. Returns to advertising are hard to measure from observational data [8], which is why we report associations only.

2. Data

Our analysis covers 257 Food & Beverages videos tracked on TikTok Shop between 2 September and 1 October 2026. Of these, 133 were watched and analysed in depth by our AI analyst. The figure is a count of videos, not of views. Video analyses are compared with a market-wide baseline of 3,342 watched videos across categories.

Marketplace figures (prices, commission, product age and competition) rest on 100 top-selling products in the category. Momentum and channel figures are as of 29 September 2026, and price and commission figures as of 1 October 2026. For each analysed video we recorded attributes such as the hook type, format and production style. Revenue is estimated, not reported by shops.

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 and what appears in the first second, 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. Weekly category figures for the US (estimated revenue, its split between video, LIVE and the shop tab, affiliate revenue, shops and active products, and a year of weekly revenue) and daily rankings of each category’s top-selling products (unit price, commission rate, launch date and revenue by channel) were collected from several independent sources. Medians of price and commission are weighted by revenue; momentum compares the latest weeks with the same number of weeks before. This paper covers 2026-09-02 to 2026-10-01: 257 videos tracked in Food & Beverages, 133 of them watched and analysed; the market baseline uses all 3,342 watched videos; 100 top-selling products in Food & Beverages. Predictions come from a logistic model retrained daily and tested on recent weeks it had not seen.

3. Methods

Sales per view is a group's revenue per view divided by the market's, so 1 is the market average. Small groups are shrunk toward the average by empirical Bayes [9], and we report 90% intervals on the log scale. The predictive model is a logistic model of top-selling videos, retrained daily and tested on recent weeks it had not seen. It is reported as AUC [10] with a bootstrap range [11].

Momentum compares estimated revenue in the latest 4 and 13 weeks with the same number of weeks before, using weekly revenue over the past year. Channels split latest-week revenue into shoppable video, LIVE and the shop tab, plus the share credited to affiliate creators. Prices and commission rates are revenue-weighted medians and ranges of the tracked top products, with revenue per product shown against the average product. Product age is measured in days from launch.

Competition is measured by shops and active products, average revenue per shop, and concentration among the top products (the top ten's share and the Herfindahl–Hirschman index [12]). Seasonality compares each month with the category's average month over the past year, excluding partial months. Separating season from trend follows [13].

4. Results: momentum, channels and competition

Food & Beverages accounted for 2.1% of market revenue in the latest four weeks, ranking 17th of 28 categories. Revenue fell 1.7% against the previous four weeks and 1.7% against the previous 13 weeks, while the market fell 11.1% over 13 weeks. The category is shrinking, but more slowly than the market.

In the latest week 66.6% of category revenue came through shoppable video, 19.3% through LIVE and 14.1% through the shop tab. The market splits 64.9%, 17.1% and 18.0%, so the category leans slightly more on video and LIVE and less on the shop tab. Affiliate creators were credited with 76.9% of revenue.

The category has 3,420 shops and 22,676 active products, compared with 4,461 shops in the median category. Average revenue per shop is $2,401, and the ten largest tracked products hold 33.9% of top-product revenue; the median category matches both figures. The Herfindahl–Hirschman index among top products is 188, which indicates low concentration.

5. Results: prices, commission and product age

The revenue-weighted median price of top products is $29.99, with the middle half between $23 and $43. The $20–35 band holds 38 products and 40.7% of top-product revenue, and the $35–60 band holds 27 products and 29.1%. Revenue per product is about average in both (1.07× and 1.08×). Products at $10–20 earn 0.60× the average product. The $60–100 band earns 2.02× the average, but it holds only 6 products, so we treat it as indicative.

The revenue-weighted median commission rate is 14%. The 20–30% band has 9 products and 18.0% of revenue, with revenue per product at 1.70× the average. The 5–10% band is at 0.79×. The share of product revenue sold through video falls as commission rises, from 86.2% at 5–10% to 73.6% at 20–30%. The under-5% and 30%-and-over bands have only 2 and 3 products, so we do not read their shares.

Products launched in the last 90 days produce 20.5% of top-product revenue. Products 30–90 days old earn 1.72× the average (9 products), whereas those under 30 days earn 0.83× (6 products). Products 6–12 months old account for the largest share of revenue, 31.3%. These age groups are small, and the 30–90 day result may reflect a few strong launches.

6. Results: hooks, formats and what predicts top sellers

Among watched videos, bold-claim openings sell at 1.96× the market average per view (90% range 1.46–2.64; 17 videos) and account for 19.0% of sales. Relatable POV openings reach 1.62× (1.28–2.06; 21 videos, 26.8% of sales), and price or deal openings 1.42× (1.07–1.87; 15 videos). Question (1.03×) and problem call-out (0.96×) hooks sit near the average, with ranges that include 1. Story openers are lowest at 0.47× (0.31–0.71), but only 8 videos support that estimate and confidence is low.

By format, demo or tutorial videos sell at 1.79× (1.32–2.43; 16 videos). Routine or GRWM videos reach 1.62× (1.20–2.19) on 9 videos and low confidence. Talking-head videos sell at 1.25× (0.94–1.68), the best-supported estimate with 30 videos. Voice-over B-roll gives 21.4% of sales but sells at 0.74× (0.49–1.11), and reviews or testimonials at 0.82× (0.63–1.07). All 79 videos in the production comparison were filmed live action (1.02×), so there is no production contrast within this category.

The model of top-selling videos has an AUC of 0.64 (90% range 0.57–0.70), trained on 2,667 rows. It was not trained only on analyses that never saw sales (cleanOnly is false), so its signal may partly reflect sales-aware labelling. The strongest associations are with filmed live action (odds ratio 2.54), a visual pattern-break combined with a review format (1.92), and no visible AI (1.71). Education content (0.59) and showing the product in the first second (0.62) are associated with lower odds. Predictive power is modest.

7. Discussion

For brands, the evidence points to products priced around $20–60, where most top-product revenue sits, and to claim-led, demonstration-led content. Sales per view appears higher when a video states a clear promise or shows the product being made or used, than when it tells a story or relies on voice-over footage. For creators, commission rates in the 20–30% band coincide with higher revenue per product, but this may reflect which products offer higher rates, not any effect of the rate. Higher-commission products also sell less through video.

Seasonality in this category is modest. Over the past year, December (1.20×), March (1.18×) and June (1.19×) ran above a normal month, and January was lowest (0.86×). Halloween is under way now. Black Friday / Cyber Monday starts in 40 days, holiday gifting in 45 days and New Year resolutions in 92 days. With only one year of history we cannot separate repeated seasonal peaks from one-off events [13].

Several alternative explanations apply. Larger or established creators may use bold claims more often and would sell more regardless of the hook [2]. Ad boosting is associated with higher odds in the model, and advertising returns are hard to isolate from observational data [8]. Newer products doing well may mean fresh launches get promotion, not that novelty sells. Category revenue is estimated and not audited.

8. Limitations

Most videos in the sample are top-ranked, so findings mostly separate strong sellers from good ones rather than from all videos; typical and weak videos are being added. Product figures cover the top-selling products we track in each category, not every listing, so they describe what sells best rather than everything on sale. Results are associations, not causes. Paid promotion, creator audience size and product price are not fully controlled for. Revenue, views and sales are estimates, not figures reported by TikTok. The window is one month, and most hook and format groups contain fewer than 25 analysed videos, several with low confidence. Revenue is estimated. The seasonal index covers one year, so it cannot distinguish repeat seasonality from one-off events. Creator size, ad spend and product quality are not controlled for. The $60–100, under-5% and 30%-and-over commission, and under-30-day bands each contain fewer than ten products. The model has modest accuracy and was not trained only on analyses that never saw sales.

9. Conclusion

In Food & Beverages on TikTok Shop, sales are concentrated in mid-priced products and in shoppable video with affiliate creators. Bold claims, relatable POV and price or deal openings, and demo or tutorial formats, are associated with sales per view above the market average. Story openers and voice-over B-roll are associated with below-average sales per view.

The category is declining slightly while the market falls faster, and the coming retail moments give a natural test window. These results are associations from a one-month window, and most video groups hold fewer than 25 videos. They should guide what to test, not what to assume.