Video Content Analysis
Hand a model the words in a video, not the view count.
A model cannot watch a video, so content analysis quietly becomes metadata analysis: captions, hashtags, and engagement numbers standing in for the thing itself. Transcripts remove the substitution. Find the videos by keyword or hashtag, take the transcript of each, and what reaches the model is what was actually said — on TikTok and on LinkedIn, through the same call shape.
Build it into a product.
Find the videos worth reading, take their transcripts, and hand a model text instead of view counts.
Supported responses use a consistent ActivityStreams 2.0-based model. Shared concepts map to consistent fields, while platform-specific capabilities remain documented. Standard calls use one credit; batch endpoints charge one credit per submitted ID. Failed and timed-out requests are automatically refunded.
Or run it in a tool you already use.
The same endpoints are available as MCP tools. Connect a supported client once and the work becomes a question you ask or a step in a scheduled workflow — with no integration to build.
Chat — read the video, not the caption
Ask ChatGPT or Claude to find videos on a topic and pull their transcripts, so the summary you get is of the content rather than the metadata.
Example prompts
- Find the top 5 TikToks about [topic], read their transcripts, and tell me what claims they make.
- Transcribe this TikTok and pull out every product mentioned.
- What do the highest-performing videos under #[hashtag] say in their first ten seconds?
- Compare how this LinkedIn video and this TikTok explain the same idea.
Automation — a transcript pipeline
Search, transcribe, and store on a schedule, so the text is already there when someone asks what a set of videos was about.
Example workflow
A nightly n8n workflow that pulls new videos under three hashtags, transcribes each one, and writes the text to a vector store your assistant already queries.
What this covers.
Public data only. Supported endpoints return what the platform serves publicly, and reaching it never depends on a connected account — so private analytics for accounts you own are outside what this can answer.
- Classify video content by what it says rather than by its caption
- Summarize a set of videos into the claims and hooks they have in common
- Give a product feature the text behind a video without hosting the video
- Compare how the same message is framed on TikTok and on LinkedIn
Try it with your own accounts.
Get a free API key and make the first call. 100 credits, every platform, both interfaces, no credit card.