Documentation
Everything you need to understand and use Veractum.
Getting Started
Veractum analyses YouTube videos in three steps:
- Paste a YouTube URL into the input on the homepage
- Sign in with Google, GitHub, or Twitter if prompted
- Wait 1–3 minutes while the AI processes the video
The video must have captions or auto-generated subtitles enabled on YouTube. Videos without any subtitles cannot be analysed.
How Analysis Works
When you submit a video URL, Veractum:
- Fetches the transcript via the Supadata API, which retrieves YouTube captions including auto-generated ones
- Generates a summary using Gemini 2.5 Flash — key points, topic segments with timestamps, and an overview
- Extracts claims — factual statements, opinions, and predictions identified in the transcript
- Fact-checks each claim — Gemini searches its knowledge and (when configured) the web to assign a verdict and confidence score
Processing runs as a background job via Inngest with no time limit, so long videos are fully supported.
Understanding Results
Supported
The claim is backed by credible evidence. Confidence is typically 70–100%.
Contested
Evidence contradicts or raises significant doubt about the claim. Confidence is 30–70%.
Opinion
The statement is subjective and cannot be objectively verified.
Insufficient Data
Not enough evidence was found to reach a verdict.
The confidence score (0–100%) reflects the AI's certainty in its verdict. Always use results as a starting point, not a final answer.
FAQ
Why does my video fail with 'no captions available'?
The video must have either manual captions or auto-generated subtitles enabled by the creator. Live streams and some private videos are not supported.
How long does analysis take?
Typically 1–3 minutes, depending on video length and current server load. The Inngest background job has no hard timeout.
Can I analyse the same video twice?
Yes. Each analysis request creates a new independent analysis and counts toward your monthly quota.
Why is the timeline only showing the first part of a long video?
Veractum samples the transcript evenly across the full video duration to generate distributed timeline segments and claims.