Tiktok Research

TikTok research

Use Crawlora's normalized REST endpoints to answer TikTok-specific research questions. Keep the result tied to the returned public data: a feed or search sample is not a complete census, and a metric is not evidence of why a video or ad performed well.

Setup and requests

Set CRAWLORA_API_KEY to a key from crawlora.net, then run the helper bundled with this skill. It sends x-api-key to https://api.crawlora.net/api/v1; keep the key in the environment and never put it in a URL, prompt output, or saved file.

Set `CRAWLORA_API_KEY` in the environment before running the helper.

scripts/crawlora.sh /tiktok/profile/chatgpt | jq '.'
scripts/crawlora.sh /tiktok/search keyword="ai agents" count=20 | jq '.'
scripts/crawlora.sh /tiktok/post/7444278905264983342 | jq '.'

Read reference/endpoints.md for the complete endpoint list and current parameter names. Check both the HTTP result and the JSON envelope: a successful HTTP response can still contain a nonzero application code, a nested upstream error, or no usable results.

Choose the workflow

Profile, videos, and comments

1. Call /tiktok/profile/{handler} with the handle without @.
2. For the profile's videos, pass the returned user.secUid exactly to /tiktok/posts. Do not substitute the numeric user id or display name. sort_type is 0 latest, 1 popular, or 2 oldest.
3. Use /tiktok/post/{id} for one video's detail. Use /tiktok/comments with that video's aweme_id (the video id from its URL) for top-level comments.
4. Follow returned cursors only to the user's requested bound. Preserve the cursor type and stop when has_more/hasMore is false; do not assume every TikTok response uses the same casing.

Search and discovery

• /tiktok/search finds videos by keyword.
• /tiktok/search/hashtag finds hashtag/challenge candidates.
• /tiktok/search/user finds users by keyword.
• /tiktok/hashtag/{name} resolves a hashtag and returns its id; pass that id to /tiktok/hashtags to retrieve its videos.
• /tiktok/category discovers Explore categories; pass a returned category type to /tiktok/explore/{id}.

Use identifiers returned by the preceding call. Search matches and hashtag counts are evidence from TikTok's current index, not proof that every result is about the requested brand or topic. Verify identity from the returned profile, caption, source URL, or detail record before making a strong attribution.

Trending and Creative Center

• /tiktok/trending returns a current recommended/trending feed.
• /tiktok/creative-center/hashtags returns ranked hashtags for a country and period.
• /tiktok/creative-center/videos returns ranked videos, optionally filtered by content label or organic_only and sorted by views, engagement, or six_second_views.
• /tiktok/popular-trend/country-industry-meta provides country and industry metadata when a Creative Center context needs to be explained.

Always report country, period, sort/filter choices, retrieval time, and sample size. “Trending” is surface- and time-dependent; do not present it as a global or historical ranking.

Top Ads intelligence

For a matched ad study, first call /tiktok/top-ads/filters to discover current dynamic country, industry, objective, language, and pattern-label ids. Keep period, country, objective, and sort consistent across comparisons. Then:

1. Search /tiktok/top-ads/list with the agreed filters and a bounded limit. State period (7, 30, or 180) and order_by (for_you, impression, ctr, play_2s_rate, play_6s_rate, cvr, or like) in the output.
2. Deduplicate the returned materials and pass each returned id as material_id to /tiktok/top-ads/detail. This is a Top Ads material id, not the ad's TikTok video id; id and materialId are not accepted by the detail endpoint.
3. Use /tiktok/top-ads/analysis for interactive-time charts and percentiles. Match metric and period_type across ads. Supported metrics are retain_ctr, retain_cvr, click_cnt, convert_cnt, and play_retain_cnt.
4. Use /tiktok/top-ads/recommend or /tiktok/top-ads/spotlight only as supplemental discovery. Do not mix those handpicked/related materials into a matched filtered sample without labeling them.

For a dedicated competitor-ad comparison or creative brief, the narrower tiktok-ad-research skill provides the same Top Ads endpoints with a more focused deliverable.

Interpretation and limits

• Public data only: no login, private content, hidden contacts, or audience demographics. A verification badge does not establish audience authenticity.
• Anonymous Creative Center access is limited. Hashtags return at most three records; videos return only page one (four videos), regardless of the apparent upstream totals. Treat those totals as upstream-reported context, not accessible coverage.
• Creative Center video availability varies by country. US, JP, ID, VN, and TH have reliably returned populated samples; a valid empty response in another market can mean that TikTok has no anonymous data for that market. Do not retry it indefinitely or call it an outage without other evidence.
• Creative Center engagement_rate and six_seconds_vtr are ratios, not percentages. Period-scoped and lifetime fields must not be compared as if they covered the same window.
• A Top Ads percentile is not a raw CTR/CVR, spend, conversion count, or ROI. Preserve the metric label, period, and units exactly as returned.
• Signed image/video URLs may expire. Retain stable TikTok source links and ids so the research can be repeated. Do not claim to have watched or heard a creative unless the environment actually exposed the media.
• Missing fields are unknown, not zero. A failed live refresh may support a clearly labeled stale/dataset observation, but must not silently become a current fact.

Output and reliability

Return the requested answer with source URLs or stable ids, retrieval time, filters, and sample size. For comparisons, keep the same definition and window across rows. Separate observed facts from interpretation and label hypotheses as hypotheses.

Use bounded requests. On 429, back off; retry one transient 5xx when useful. Stop on 401/403 and report the access limitation. Preserve a clearly labeled partial result when some details or comments fail, and inspect the application-level code before treating an empty or nested response as valid data.