TikTok Shop Review Insights

The Actor turns a TikTok Shop product's reviews into one analysis record per product: the complaints and compliments customers raise most often, each ranked by how many reviews mention it, what share of the review sample that is, the average rating among reviews that raise it, and a handful of verbatim quotes as evidence. It does not return the reviews themselves; for that, see the TikTok Shop Reviews Scraper.

{
  "product_ids": [
    "https://shop.tiktok.com/us/pdp/1730927783781307026",
    "1729447567685489298"
  ],
  "region": "US",
  "reviewSampleSize": 100
}

Accepted input

product_ids is required and takes page links or bare product IDs, mixed freely in the same list. Each product is analysed separately; one that cannot be read is logged as a warning and skipped rather than ending the run, so a bad ID in a ten-product list costs that product only. region is required and selects the storefront read — US, GB, SG, MY, PH, TH, or VN — which changes the products and reviews returned, not only the display language. reviewSampleSize defaults to 100 and bounds how many recent reviews are pulled per product before analysis runs; 0 requests the upstream ceiling of 500. A larger sample surfaces themes that show up too rarely for a small sample to catch, at the cost of one larger request per product.

How the analysis works

The analysis is deterministic: keyword and phrase matching against sixteen fixed themes — sizing and fit, quality and durability, shipping and delivery, packaging, value and price, customer service, authenticity, appearance versus the listing, functionality, scent and taste, comfort, missing or wrong items, effectiveness and results, adverse reactions, damage on arrival, and instructions and ease of use — combined with the star rating each review carries. A review rated 1 or 2 stars that matches a theme counts toward that theme's pain_points entry; a review rated 4 or 5 stars that matches the same theme counts toward praise_points. Three-star reviews contribute to rating_average and rating_distribution but are not counted toward either list, since they carry the weakest sentiment signal. No language model is called and no request beyond the review fetch itself is made — the whole analysis runs on the review batch already in memory, which keeps the event affordable and the output reproducible: the same reviews always produce the same record.

Themes match English and Spanish. Roughly a quarter of the US-storefront reviews sampled while building this Actor were written in Spanish, so calidad, llegó roto, no sirve and their neighbours count toward the same themes as their English equivalents rather than being silently dropped.

Measured coverage. Against 300 real reviews pulled from ten US products across headphones, sneakers, serum, supplements, leggings and kitchen goods, the theme table matched at least one theme in 68% of reviews overall and 73% of 1–2 star reviews. The remaining third are reviews that are purely evaluative ("love it", "not for me") or describe something no fixed theme covers. Those reviews still count toward reviews_analysed, rating_average and rating_distribution — they simply do not add a themed mention, so a product whose complaints are unusually phrased will show a shorter pain_points list than its rating distribution implies.

The theme list is fixed and does not learn new topics from a specific product's reviews. Keyword matching can also cross themes: a sizing complaint that also contains the word "return" is counted once under sizing and fit and once under customer service, since a single review can raise more than one theme.

Result fields

{
  "product_id": "1730927783781307026",
  "reviews_analysed": 10,
  "rating_average": 2.9,
  "rating_distribution": { "1": 3, "2": 2, "3": 1, "4": 1, "5": 3 },
  "pain_points": [
    {
      "theme": "Sizing & fit",
      "theme_key": "sizing_and_fit",
      "mentions": 4,
      "share_of_reviews": 0.4,
      "avg_rating_when_mentioned": 1.5,
      "example_quotes": [
        "Runs way too small, had to return it. Very disappointed.",
        "The size is off and it does not fit like the photos showed."
      ]
    }
  ],
  "praise_points": [
    {
      "theme": "Shipping & delivery",
      "theme_key": "shipping_and_delivery",
      "mentions": 4,
      "share_of_reviews": 0.4,
      "avg_rating_when_mentioned": 4.75,
      "example_quotes": [
        "Arrived way faster than expected, great shipping!",
        "Fast delivery and the packaging was neat."
      ]
    }
  ],
  "verdict_summary": "Based on 10 reviews averaging 2.9★ — the top complaint is sizing & fit (4 mentions, 40% of reviews, 1.5★ average when raised); the most praised aspect is shipping & delivery (4 mentions, 40% of reviews).",
  "trend": { "theme": "Sizing & fit", "direction": "stable", "earlier_share": 0.4, "recent_share": 0.4 }
}

This is a real record produced by the analysis, from a ten-review fixture rather than a live product, kept short enough to read in full; pain_points and praise_points hold up to five themes each on a real run, ranked by mentions. share_of_reviews is mentions divided by reviews_analysed, not by the size of the pain or praise group alone, so it reads as "the share of the whole sample," comparable across both lists. avg_rating_when_mentioned is the average rating among only the reviews in that theme's own band — a pain_points entry averages 1 to 2, a praise_points entry averages 4 to 5 — so it says which end of that band a theme leans toward, not whether it is positive or negative. example_quotes holds up to three excerpts, truncated to 160 characters, from the reviews that raised the theme. verdict_summary restates the top entry of each list in one sentence; it is a template filled with the same numbers already in pain_points and praise_points, not a separate summary.

Trend

trend reports whether the top pain point is mentioned more in the recent half of the timestamped sample than the earlier half, splitting the sample by review_time at its midpoint. direction is rising, falling, or stable when there are at least ten timestamped reviews in the sample and at least four of them mention the top pain point across both halves combined; below either threshold it reports insufficient_data rather than a direction computed from too little evidence. trend is null when no pain point was found at all — a product with no theme in pain_points has nothing for trend to track.

Charging

Each product that returns an analysis record is billed the review-insights event once. A product with no reviews, or one whose lookup fails, yields no row for that product and is not billed; the run continues to the next product in product_ids rather than failing outright.

Related Actors

For the underlying review text and star ratings themselves, with photo, verified-purchase, and rating filters, use the TikTok Shop Reviews Scraper. For a product's full page — description, variants, seller profile, and a page of reviews alongside everything else — use the TikTok Shop Product Scraper. For the seller behind the product, use the TikTok Shop Seller Info Scraper.