TikTok Shop Shoppable Video & Creative Finder

Find public TikTok videos connected to TikTok Shop products and turn them into structured creative intelligence.

Each dataset row represents one public video → product association. The Actor returns public performance metrics, creator and product context, caption-derived hooks, calls to action, formats, angles, and a transparent creative score.

What you can do

• find public videos for one or more TikTok Shop product URLs;
• discover products by niche and inspect their public video creatives;
• compare views, likes, comments, shares, and engagement rate;
• extract caption hashtags and deterministic hook/CTA patterns;
• identify caption-derived formats such as demonstration, tutorial, comparison, testimonial, unboxing, problem/solution, or deal promotion;
• rank creative leads by score, views, engagement, recency, or product sales;
• export results as JSON, CSV, Excel, API output, or MCP tool results.

Evidence boundaries

The Actor does not invent product attribution or claim to understand video frames it did not inspect.

• DIRECT_PRODUCT_MATCH / HIGH: structured public video data references the product ID.
• PRODUCT_PAGE_CONTEXT / MEDIUM: the video was exposed in public product-page context.
• SEARCH_QUERY_CONTEXT / LOW: a relevant public video was discovered through indexed/public search context. Treat it as a creative lead, not confirmed shoppable or affiliate attribution.

Hook, CTA, format, and angle fields are deterministic signals from public caption/indexed text. They are not frame, audio, transcript, or conversion analysis.

Input example

{
  "productUrls": [
    {
      "url": "https://shop.tiktok.com/us/pdp/hpbs-24000mah-power-bank-6000a-jump-starter-150psi-inflate/1731973421800919262"
    }
  ],
  "searchQueries": [],
  "region": "US",
  "maxProductsPerQuery": 1,
  "maxCreativesPerProduct": 10,
  "maxBrowserRequests": 20,
  "minVideoViews": 0,
  "searchPublicTikTokVideos": true,
  "sortBy": "CREATIVE_SCORE",
  "proxyConfiguration": {
    "useApifyProxy": true,
    "apifyProxyGroups": ["RESIDENTIAL"],
    "apifyProxyCountry": "US"
  },
  "maxConcurrency": 2,
  "navigationTimeoutSecs": 45,
  "includeRawData": false,
  "includeSearchFallbacks": false
}

Provide at least one productUrls entry or one searchQueries value.

Output example

{
  "creativeId": "7652247745117359382:1729545068746215852",
  "videoId": "7652247745117359382",
  "videoUrl": "https://www.tiktok.com/@creator/video/7652247745117359382",
  "caption": "Flat tire? Here is how this portable inflator works. Get yours today.",
  "views": 10000,
  "likes": 500,
  "comments": 25,
  "shares": 10,
  "engagementRate": 5.35,
  "creatorUsername": "creator",
  "productId": "1729545068746215852",
  "productName": "Portable Car Jump Starter With Air Compressor",
  "associationState": "SEARCH_QUERY_CONTEXT",
  "associationConfidence": "LOW",
  "hookTypes": ["QUESTION"],
  "ctaTypes": ["SHOP_NOW"],
  "formatSignals": ["PRODUCT_DEMONSTRATION", "PROBLEM_SOLUTION"],
  "creativeAngles": ["CONVENIENCE"],
  "creativeScore": 52.4,
  "creativeSignals": [
    "Strong engagement rate",
    "Caption hook: QUESTION",
    "Caption CTA: SHOP_NOW",
    "Search-context lead — product association is not confirmed"
  ]
}

Missing public metrics remain null. UNDETECTED means the public caption did not provide enough text evidence for that signal.

Creative score

creativeScore is a deterministic 0–100 ranking signal based on:

• public reach and engagement volume;
• engagement rate;
• recency when the publication time is available;
• product association strength;
• presence of caption hook, CTA, format, and hashtag evidence;
• public product opportunity score.

It is not a prediction of revenue, conversions, or future virality.

Dataset views

• Creative opportunities — compact ranking for research and outreach.
• Creative evidence — detailed metrics, association evidence, and caption-derived signals.

MCP and API

The Actor is compatible with Apify API clients and Apify's hosted MCP endpoint. Agent-facing guidance:

• use this Actor when the user asks for specific videos, creative examples, hooks, CTAs, formats, or ad inspiration;
• use the Affiliate Creator Finder when the goal is creator discovery and profile-level outreach;
• do not describe SEARCH_QUERY_CONTEXT as confirmed product or affiliate attribution;
• do not start parallel or replacement runs unless the user explicitly requests them.

The default dataset is available at:

https://api.apify.com/v2/datasets/{DATASET_ID}/items

The run summary is stored in the default key-value store under OUTPUT.

Delivery and diagnostics (v0.1.3)

Verified creative records are saved as soon as they are ready. The default dataset is in discovery order. sortBy applies within each ready batch and to the complete JSON export in RANKED_CREATIVES, linked from the Output tab at completion. The ranked export reuses delivered records and does not create extra paid dataset events.

Duplicate product/video associations are serialized and suppressed. On restart, delivered IDs are restored from the dataset. A write with a lost acknowledgement is reconciled against the dataset; if its outcome cannot be confirmed, delivery stops instead of automatically charging/writing the association again. This is not a transaction spanning the remote storage service and billing service.

For keyword jobs the first products with publishable videos fill the per-keyword output slots, up to maxProductsPerQuery. Within a product, the first verified creatives fill maxCreativesPerProduct; later discoveries cannot replace already delivered and charged rows. Search context is still LOW confidence.

The log contains a short Video discovery step per attempt. Full details are in OUTPUT.pipeline.affiliateDiagnostics: query, status, candidate and relevance counts, and outcomes including NO_CANDIDATES, RELEVANCE_REJECTED, BLOCKED, REQUEST_FAILED, and SEARCH_TIME_LIMIT. Zero search matches are not proof that no videos exist. Coverage states describe observed results and limits; they do not claim exhaustive TikTok coverage.

Resource controls

• maxBrowserRequests bounds scheduled browser requests; one retry per request and retry-only product homepage warm-up may add navigations;
• images, fonts, video playback and common binary media URLs are blocked;
• product homepage warm-up runs only on a retry;
• maxIndexedSearchSecs defaults to 30 seconds per product for indexed HTTP searches;
• indexed searches try a seller/model query and two short title-based variants;
• at most two extra links with insufficient snippet relevance are probed. Only relevant structured video payloads become output, never search snippets alone;
• the browser search fallback runs if indexed detail probes yield no output;
• each product has at most twice maxCreativesPerProduct scheduled detail pages;
• maxConcurrency is capped at 2;
• US residential access is recommended for geo-sensitive TikTok Shop pages;
• the Actor uses no external data APIs and calls no external Actors.

These controls reduce unnecessary work; they are not a guaranteed infrastructure dollar cap. No live success-rate or cost reduction is claimed from offline tests.

Limitations

• TikTok Shop availability and public payloads vary by region and session;
• public video metrics can change after extraction;
• deleted, private, restricted, or non-indexed videos cannot be returned;
• search-context associations require human verification;
• this release analyzes public caption text and metadata, not video frames, speech, audio, or comments;
• the Actor does not calculate sales attribution or creator commission.

Privacy and responsible use

The Actor processes public pages and public metadata. Use results lawfully, respect platform rules, and verify creator/product associations before outreach or publication.

Stop after the delivered-result limit

Before a queued VIDEO_DETAIL or VIDEO_SEARCH request is dispatched, the Actor checks confirmed dataset rows for that product. If maxCreativesPerProduct is already full, it skips navigation and records RESULT_LIMIT_REACHED. Other products continue. Rejected candidates and unsuccessful writes do not fill slots. Already running pages may complete when concurrency is greater than one.

OUTPUT.pipeline.skippedVideoRequestsAtLimit counts requests drained without site navigation. browserRequests/videoDetailRequests remain scheduling counts. The first qualifying rows still fill the limit; the separate ranked export sorts those same rows, not every video that might exist online. This update keeps extraction, relevance, scoring, view filters, sources, retries and timeouts unchanged.