A transcript is useful when it leads to a better brief, not when it becomes a pile of copied lines. This skill takes a small, intentional set of public TikTok URLs and returns the patterns worth discussing: the hook, the promise, the audience question, the evidence, and an original next-angle to test.
What this skill does
It starts with public source URLs. For each video, it calls the TikTok transcript endpoint, cleans the returned WEBVTT text, and keeps the URL beside every observation. Then it groups repeated themes into a practical brief a writer, editor, or marketer can review.
The API call is deliberately simple:
GET /v1/tiktok/video/transcript?url={tiktok_video_url}&language={language}
The endpoint returns id, url, and a timestamped transcript. Videos must be public. If a transcript is not available, the skill records that rather than inventing the spoken content. use_ai_as_fallback=true is an optional last step for a shortlisted video under two minutes; it adds 10 credits.
Copy the skill
name: tiktok-transcript-content-brief
description: Turn a small set of public TikTok videos into an auditable, original content brief. Preserve source URLs, use transcripts as evidence, and clearly separate observed patterns from new ideas.
---
# TikTok transcript to content brief
## Use this when
Use this skill when someone has public TikTok video URLs and wants to understand what is being said without manually watching every clip. It is for content research and briefing, not copying scripts or claiming that a format will perform.
## Inputs
```json
{
"video_urls": ["https://www.tiktok.com/@creator/video/123"],
"language": "en",
"goal": "Find useful hook and topic patterns for original content",
"max_videos": 12,
"audience": "Optional target audience or customer"
}
```
Defaults: process no more than 12 URLs in one pass. If the requester gives more, ask them to prioritize or sample the most relevant videos. Use public URLs only.
## API contract
All requests use `https://api.scrapecreators.com` with an `x-api-key` header held in an environment variable. Never put the key in the skill, a prompt, browser code, or the report.
```text
GET /v1/tiktok/video/transcript?url={tiktok_video_url}&language={language}
```
The response contains `id`, `url`, and `transcript` as WEBVTT. The transcript endpoint supports public TikTok videos. If needed for a single shortlisted video under two minutes, add `use_ai_as_fallback=true`; note that this adds 10 credits.
## Workflow
1. Canonicalize and dedupe the supplied URLs. Keep the original URL as the source of record.
2. Request one transcript per unique URL. Do not call fallback during the first pass.
3. Remove WEBVTT timestamps and obvious repeated fragments for analysis, while preserving the raw transcript in the research record.
4. For each usable transcript, identify:
- the opening hook or question,
- the promise or outcome,
- the main steps, proof, or examples,
- audience questions or objections,
- phrases that are useful as themes but must not be copied verbatim.
5. Group patterns across the sample. A pattern needs support from at least two videos; otherwise label it a single-source observation.
6. Draft three to five original content angles. Each angle must name the audience problem it addresses and say how it differs from the source material.
7. Flag claims that need fact-checking, regulated advice, unclear context, or a transcript that may be inaccurate.
## Missing or weak transcripts
If a transcript is unavailable, return `transcript_status: unavailable`. Do not infer the spoken content from a thumbnail, a creator's appearance, or a caption. If the requester approves it and the public video is under two minutes, retry once with `use_ai_as_fallback=true` and disclose that the fallback was used.
## Output
Return a compact brief with these sections:
1. **Sources reviewed**: URL, transcript status, and language.
2. **Observed patterns**: hook, promise, evidence, number of supporting videos, and source URLs.
3. **Original content angles**: a title, audience problem, suggested structure, and how it differs from the observed sources.
4. **Review flags**: missing transcripts, claims to verify, and limitations.
Use this shape for each observation:
```json
{
"source_url": "https://www.tiktok.com/@creator/video/123",
"transcript_status": "available",
"hook_pattern": "Starts with a specific everyday problem.",
"evidence": "Faithful short paraphrase of the spoken opening.",
"theme": "time-saving routine",
"confidence": "medium",
"review_flags": []
}
```
## Rules
- Public data only.
- Never fabricate transcript text, sources, or performance claims.
- Do not copy a creator's script, unique phrasing, or sequence of examples.
- Do not infer sensitive traits from content or appearance.
- Keep source URLs with all observations.
- Treat content ideas as hypotheses for human review, not a guarantee of views.
Keep the research honest
The useful output is a small brief with its receipts. Keep the public source URL, mark any transcript that is missing or generated through fallback, and do not turn a handful of clips into a universal conclusion. If an insight appears in just one video, say that.
For broader discovery before this step, use the TikTok keyword search endpoint. Then bring only the strongest public URLs into the transcript brief. That keeps the cost and the review load sensible.
