TikTok Creator Research

Vet public TikTok creators and content

Build a current, evidence-backed view of public TikTok creators. Keep identity, content fit, posting activity, and engagement observations separate from claims about audience demographics, authenticity, or partnership history.

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. Read reference/endpoints.md for exact parameters.

sh

Set `CRAWLORA_API_KEY` in the environment before running the helper.
scripts/crawlora.sh /tiktok/profile/username | jq '.'
scripts/crawlora.sh /tiktok/search/user keyword="running coach" | jq '.'

Check both the Crawlora envelope and any nested upstream application code before using a response. Keep the key only in CRAWLORA_API_KEY.

Workflow

1. Translate the brief into topic, geography, follower range if available, creator count, recency window, content-fit criteria, and any disqualifiers. Ask only for missing details that would change the shortlist.
2. Discover handles with /tiktok/search/user when needed. Search results are candidate evidence, not a verified identity or complete creator directory.
3. Refresh each candidate with /tiktok/profile/{handler} using the handle without @. Capture the returned stable user id, uniqueId, nickname, verification flag, bio, secUid, follower count, heart count, and video count. A missing/deleted/banned handle is not a successful refresh.
4. For recent or popular content, pass the profile's returned user.secUid exactly to /tiktok/posts. Do not substitute the numeric user id, nickname, or handle. Use sort_type=0 for latest, 1 for popular, or 2 for oldest; take a bounded, consistent sample and follow its cursor only as needed.
5. Use /tiktok/post/{id} to enrich selected videos. Use /tiktok/comments with the video's aweme_id to sample top-level comments when engagement quality matters. Comments are a sample, not a complete sentiment census.
6. Rank candidates only against the user's stated criteria. Deduplicate by stable creator or video id and preserve the refresh time for every row.

Interpretation

• A profile's follower count, likes, views, and verification status are public profile signals. None proves audience authenticity, demographic composition, location of followers, or conversion potential.
• Bio keywords and a small post sample show candidate content fit, not sustained niche ownership. Record sample size and dates before describing cadence or consistency.
• If calculating engagement, state the formula, numerator fields, denominator, sample size, and dates. Do not replace unavailable metrics with zero or rank creators on incomparable windows.
• Public contacts may be unavailable. Do not guess email addresses or hidden contact information; do not send outreach unless the user separately asks.
• Link identities across platforms only when a public profile link or other direct evidence establishes the match. Similar handles are not proof.
• If a live refresh fails, a candidate can remain in a clearly labeled unrefreshed/candidate state; it must not silently become a current observation.

Output and reliability

Return the requested shortlist with handle, stable creator id, profile URL, topic-fit evidence, follower/video counts, verification status, refresh time, sampled video URLs, metric definitions, and unknowns. Include a brief reason for each inclusion and disclose the discovery and sampling limits.

Use bounded profile, post, detail, and comment calls. On 429, back off; retry one transient 5xx when useful. Stop on 401/403 and report the access limitation. Preserve a partial shortlist when only some refreshes succeed, and inspect application-level code before treating data as valid.