Chat · guide

Can ChatGPT analyze videos: the yes is not in the list

By Alberto Gulotta · Updated 23 September 2026 · 15 min read

Can ChatGPT analyze videos? Yes, and OpenAI says so in one sentence — published in its FAQ about images, not the one about files. A video goes in as an attachment, on Free plans too. The live camera is a separate feature, in the iOS and Android apps. Below: where each sentence is published, and what each one leaves out.

That placement is not a curiosity. A reader who wants to know whether a file type is accepted opens a page that lists file types — and there are two such pages, and neither of them names a video format. “What types of files are supported?” answers “most common file extensions for text files, spreadsheets, presentations, and documents… including XLSX, XLS, CSV, TSV, DOCX, PPTX, PDF, and TXT”. The list inside the image FAQ answers “PNG (.png), JPEG (.jpeg and .jpg), and non-animated GIF (.gif)”. Both are on help.openai.com.

So the honest shape of the answer is: the yes is in prose, and the lists are about something else. That is worth knowing before you conclude from a list that a thing is not allowed — which is what most of the pages answering this question have done.

What OpenAI promises, and what it does not

Saying a file is accepted is not the same as saying it is understood. OpenAI writes the second half itself, in the sentence immediately after the first.

Two practical things follow from it. The first is that a summary you get back is not a transcript: if the answer turns on a number said out loud, or on which of two people spoke, that is exactly the kind of detail the warning covers. The second is that a short clip is a different proposition from a long one, because “may not analyze the entire video” has more room to bite the longer the video runs.

There is also a small instruction in the same paragraph, about the step before the upload. OpenAI: “If you can’t select a video in Photos, check whether you can attach it through Files instead.” If Photos will not let you select the video, the upload has not started yet; try Files before anything else.

The numbers that stop a video at the door

Everything above assumes the file gets in at all. Three published limits decide that, and none of them is specific to video — which is the point, because OpenAI says video attachments “count toward your plan’s file-upload limits” like anything else.

The storage caps sit behind those: 25GB per user and 100GB per organisation, shared across chats, Projects and custom GPT knowledge. They matter less day to day than the rate, but they are the reason an “upload limit reached” message can appear on a day you have uploaded almost nothing.

Can ChatGPT watch a video while it plays?

This is a different feature with a different answer, and OpenAI keeps them apart on purpose: “Live camera video and screen sharing in Voice are separate features with their own availability and limits.” What the camera does is show ChatGPT what is in front of you, live, during a voice conversation — and the instruction is one line: “To share live video, select the camera button during a Voice conversation.”

Two conditions come with it, and both are narrower than the short answer suggests. It is in the iOS and Android apps, for eligible subscribers. And it belongs to one of the voice modes and not the other: “Live does not support video or screen sharing.” Live runs on GPT-Live-1 or GPT-Live-1 mini, depending on your plan. Advanced is the one that carries the camera, and you reach it under Settings → Voice.

That last line matters if the camera button seems to have gone. It has not been withdrawn: OpenAI says video and screen sharing “remain available” in Advanced.

One more thing about the camera is documented and worth knowing before you point it at anything: the clips are kept. “Audio clips from Live and Advanced Voice conversations, and video clips from Advanced Voice conversations, are stored with the transcript that appears in your chat history. Clips are retained for 30 days.” They are not used for training unless you turn that on yourself — OpenAI names the switch, “Include your video recordings”, under Data Controls.

Why pasting a YouTube link is the one route that is not documented

Of the ways people try, this is the one with nothing behind it. None of the five help.openai.com pages read for this guide on 23 September 2026 describes handing ChatGPT a video URL and getting the video back as understanding — and that includes the release notes, about 35,000 words of them, read end to end the same day. That is not the same as a rule forbidding it, and it should not be read as one. It means that nothing in the five pages read for this guide says what would happen, so the only evidence available is somebody trying it.

The link route, tested by somebody else. ZDNET’s David Gewirtz put the same job to three assistants, in an article dated 11 May 2026, and wrote the result plainly: “ChatGPT itself failed my tests. It couldn’t read the YouTube link. And while it theoretically can understand videos, they need to be under 500MB in size. Mine were not.” What worked for him was pairing ChatGPT with OpenAI’s Codex app rather than asking ChatGPT alone. Worth reading for exactly what it is — one person’s test, on one day, with his own files.

Between the documentation and that test, a workable order emerges for anyone who actually has a video and a question about it. Attach the file rather than the link. Keep it short and well under the size limit. Ask for the specific thing you need rather than “summarise this”, because a narrow question makes a gap in the analysis easier to spot. And if the answer will be quoted or acted on, check the part that matters against the video, because OpenAI warns that the analysis “can be incomplete or inaccurate”.

Not covered here. It will not tell you which formats are accepted. OpenAI names video as a category, and none of the five pages read for this guide lists an extension for it; inventing one from the image or document lists would be guessing in the direction the lists already got wrong.

It will not say what ChatGPT understood of any particular video. That depends on the file and on the day, and OpenAI’s own sentence is that the analysis may be incomplete.

And it will not cover generating a video, which is a different product with a different page. What holds instead is simple: the sentence that answers this is quoted from the page OpenAI publishes it on, with the date it was read, and the one route with no page behind it is named as such.