Tellscreen

Can ChatGPT watch videos? What it does with a screen recording (November 2026)

Checked and updated

ChatGPT accepts a video file as an attachment, but it doesn't watch it the way you do: OpenAI says it "may not analyze the entire video or accurately interpret its audio". In our test, OpenAI's GPT-5.6 Sol given frames from a narrated screen recording found 0 of the 5 changes asked for; given the same frames plus a transcript, it found all 5.

What OpenAI says

OpenAI's help centre says ChatGPT can take video attachments, including on the Free plan, that availability varies by app and account, and that ChatGPT "may use tools to analyze uploaded videos, but its analysis can be incomplete or inaccurate." Video attachments count toward your plan's upload limits.

In practice that means pictures sampled from the video. Anything you only said out loud (and in a screen recording that's usually the point) may never reach the model.

What we measured

We recorded two narrated screen recordings of this website, each asking for five specific changes, and gave the same request to several models. The GPT rows:

What GPT-5.6 Sol gotPlain recordingPointing recording ("change this to…")
Frames every 2 seconds, no sound0 of 50 of 5
Frames plus a timed transcript5 of 55 of 5
Tellscreen's change list (any model can read it)5 of 55 of 5

Given only frames, GPT declined to guess, which is the honest answer and still leaves you with nothing to act on. The full study, with Claude and Gemini.

How to give ChatGPT a video it can use

  1. Record with Tellscreen and talk through what you want, or upload a video you already have to your Tellscreen library.
  2. When it's ready, press Copy prompt for your AI on the video page.
  3. Paste it into ChatGPT. You get every request with your exact words, where on screen it was, and a link to a screenshot of that moment, plus the full transcript.

From a terminal (or from Codex, which runs on the same models):

curl -fsS -T recording.mov "https://tellscreen.com/api/v1/convert?wait=110"

When ChatGPT is enough on its own

If the video has no speech and the question is about what's visible (what does this chart show, which button is highlighted), sampled frames can be enough. If you're asking it to act on what someone said, give it the words.

Sources

Read next