How to let Claude or ChatGPT watch a screen recording (2026)
Checked and updated
A chat assistant can't press play on a video link, but it can read text and look at images. So give it the recording as a transcript plus stills of the moments that matter. Every Tellscreen video has that version: the Copy for AI button on the video page copies a link an assistant can open, with the summary, the full timed transcript, and a still image of any second you ask for.
Do it
- Record or upload a video on Tellscreen.
- On the video page, click Copy for AI.
- Paste it into Claude, ChatGPT or any agent that can open web links, with your question: "What did I ask Sam to do?" or "Find the moment the error appears and tell me what's on screen."
The assistant needs to be able to open links, so its web browsing or fetch feature has to be on.
What the assistant gets
The copied link points at /v/<id>/agent.json, a small file that maps the video:
- the title, length and when it was recorded
- the AI summary, key points and chapters, each with its time
- the transcript, inline for shorter videos, plus a link to the full timed transcript at
/v/<id>/transcript.json - a list of recommended moments, each with a link to a still image
- a link template for a still of any second:
/v/<id>/frame.jpg?t=42
The stills show what viewers see, camera bubble included, at up to 1280 by 720.
If the assistant is given the plain share link instead, it still works: the page includes a text copy of the summary and transcript and points to the same file.
Private and password-protected videos
A private or password-protected video won't open for an assistant from its normal link. When you, the owner, click Copy for AI, the link carries a key that opens that one video to agents for 90 days. Changing the video's password retires every key made before.
Why not just upload the video file?
Some assistants accept video files, but uploads are often capped in size and length, and a screen recording's details sit in small text that is easier to read from a full-size still than from a compressed frame. A transcript plus chosen stills is a fraction of the size and keeps the timestamps, so the assistant can say when something happened.
The idea of links agents can read comes from Builder.io's open-source Agent-Native Clips; see how the two compare.