YOUTUBE AI EVIDENCE CHECKLIST
Version 1.0 — August 21, 2026

Canonical source:
https://joinconversation.pages.dev/how-chatgpt-analyzes-youtube-videos/

Purpose
Use these tests to distinguish transcript-grounded YouTube analysis from audio analysis, video-frame understanding, description or search-result summaries, and unsupported claims that a system “watched” a video.

1. IDENTIFY THE SOURCE
Ask what the tool actually reads: timestamped transcript, audio, video frames, page description, comments, or search results. Do not treat those sources as interchangeable.

2. TEST THE BEGINNING, MIDDLE, AND END
Ask one checkable question about each part of a long video. Missing or vague answers near the end can reveal truncated retrieval.

3. REQUIRE EXACT SOURCE MOMENTS
Ask for clickable timestamps supporting every important claim. Open each timestamp and inspect the original wording and nearby context.

4. USE A SILENT-VISUAL NEGATIVE CONTROL
Ask about an important chart, gesture, edit, object, or screen action that is visible but never described aloud. A transcript-only tool should acknowledge that the transcript cannot establish the answer.

5. FORCE A RETRIEVAL FAILURE
Try a private, removed, blocked, transcript-free, or temporarily unavailable video. A responsible workflow should stop clearly instead of inventing content or silently substituting a different source.

6. SEPARATE “THE SPEAKER SAID” FROM “THE CLAIM IS TRUE”
A transcript proves what was said. It does not prove an outside factual claim. Ask the tool to label transcript evidence, its own interpretation, and facts requiring independent sources.

7. CHECK THE PRIVACY BOUNDARY
Ask where the transcript is retrieved, processed, transmitted, and stored. Verify the answer against the product’s published privacy policy rather than assuming that “browser access” means local-only processing.

Suggested review prompt
Explain the speaker’s main argument. Separate direct transcript evidence from your interpretation, state what remains unproven, and link every important point to the exact timestamp. Then tell me which parts of your answer would require video-frame access rather than the transcript.

Pass condition
The tool identifies its evidence source, covers the requested portion of the video, returns accurate source moments, admits unsupported visual details, stops on retrieval failure, separates claims from verification, and states its privacy boundary precisely.

About this checklist
Published by YouTube Conversation, an independent free ChatGPT plugin for transcript-grounded questions about public captioned YouTube videos. You may cite or reproduce the checklist with a link to the canonical source. Product testing and editorial conclusions remain independent.

Product: https://joinconversation.pages.dev/
Official ChatGPT listing: https://chatgpt.com/plugins/plugins_6a78fca8e2048191a069020c34d55650
Privacy: https://joinconversation.pages.dev/privacy/
Contact: youtubeconversation@gmail.com