The short version
ChatGPT does not play video. It has no way to watch or listen to anything on YouTube. When you paste a link, one of three things happens. It tells you it can't access the video. It browses to the page and works from whatever text it finds there. Or it writes a confident summary from almost nothing.
The third case is the one worth understanding, because it does not look like a failure.
It reads pages, it does not watch video
A YouTube URL is a web address, not a media file ChatGPT can open. With browsing on, it can fetch that page like any other page. What sits there as text is the title, the channel name, the description, the tags, sometimes a few comments. The video itself is audio and pixels, and neither reaches the model.
Behaviour also changes depending on which model and which mode you are in, which is why the same link can work on Tuesday and fail on Thursday. That inconsistency is not you doing it wrong.
The failure worth worrying about is the one that looks like success
A refusal is honest. It tells you nothing happened.
The harder case is when ChatGPT produces a summary anyway. It has the title. It has a description the creator wrote. It has whatever it knows about the topic, and possibly search results about the video or the person in it. From those pieces it can assemble something that reads exactly like a summary of a forty-minute talk it never processed.
The output is not random. That is the problem. It is plausible, on-topic and structurally correct. It just isn't from the video. If you are studying from a lecture, checking whether a claim was made, or quoting someone, a plausible fabrication is worse than an error message. You can act on it without ever noticing.
How to tell you got an invented summary
- It can't point to a moment. Ask where in the video something was said. A summary built from real content can locate things. One built from a description will hedge, offer a vague "around the middle", or produce a timestamp that doesn't hold up when you check it.
- It contradicts itself when you push. Ask the same question again, phrased differently. Ask it to quote the exact sentence. Content that came from a transcript stays stable. Content assembled on the fly gets rewritten each time, and the details drift.
- Everything it says applies to any video on that topic. If the summary would be equally true of ten other videos with similar titles, it probably came from the title. Look for specifics: names, numbers, examples, the tangent halfway through, the part where the speaker changed their mind.
- It misses what actually mattered. If you have seen the video and the summary skips the one surprising thing in it, that is not a summarising style problem. That is a video-was-never-read problem.
You can also ask directly: "Did you access the transcript, or are you inferring from the title and description?" You often get a straight answer. Don't rely on it alone, but it costs nothing.
The options that actually exist
Paste the transcript yourself
Open the video, open the transcript panel, copy it, paste it into ChatGPT. This works. It is the most reliable free method, because the model is now reading real words that were really said.
The costs are real. It is tedious for every video. Long videos hit message limits and have to be split. The copied text usually loses timestamps or turns them into noise, so you can't jump back to verify anything. And you are doing it again for the next video.
Third-party summariser extensions and websites
Plenty of Chrome extensions and sites will summarise a YouTube video for you. Many of them work.
The trade-off is where the transcript comes from. Most pull it through their own servers or a third-party transcript API, so the video URL and often your usage travel through infrastructure you don't control. Quality varies with whatever model runs behind the scenes, and you usually can't see the source text, so you are back to trusting a summary you can't check.
How AskAVideo handles it
AskAVideo is a plugin you add inside ChatGPT. It reads the transcript panel of the video you already have open in Chrome, then hands that text to ChatGPT. Nothing is guessed from the title.
Two things follow. First, ChatGPT is answering from what was actually said, so you can ask ChatGPT questions about a YouTube video the way you would ask someone who watched it. Second, every answer cites the exact timestamp it came from. You click it, the video jumps there, and you hear the sentence yourself.
The timestamp is the point. It turns a summary you have to trust into a claim you can check in four seconds. That is also what makes it usable to fact-check a YouTube video with ChatGPT, where an unverifiable answer is worthless. If you want the mechanics, we wrote up how ChatGPT analyzes YouTube videos in more detail.
What it doesn't do
Being clear about the edges is more useful than a feature list.
- It needs a public video with captions. If the video has no captions or auto-captions, there is no transcript to read. Private and unlisted videos are out.
- It needs Chrome on a desktop. The transcript is read from a real browser tab, so there is no mobile version and no way to use it from the ChatGPT app on your phone.
- It only knows what was said. Speech, not sight. If the speaker says "as you can see on this chart" and the number only exists on screen, that number is not in the transcript and AskAVideo will not know it.
- It doesn't make ChatGPT watch video. Nothing does. It gives ChatGPT the words, with sources attached.
You can install AskAVideo in ChatGPT and test it on a video you have already seen. That is the honest way to check any of this: pick something you know well, ask questions with real answers, and see whether the timestamps hold up.
Frequently asked questions
Why does ChatGPT sometimes summarize a YouTube video correctly?
Usually because it found real text about the video somewhere: a detailed description, an article covering the same talk, or a transcript republished on another site. Sometimes the topic is conventional enough that a generic summary happens to be accurate. Neither means it watched anything, and neither is repeatable.
Does a paid ChatGPT plan let it watch YouTube videos?
No. Paid plans change models, limits and available tools, but none of them add video playback. The gap is not a subscription tier.
Why does ChatGPT say the YouTube link isn't working?
The most common causes are browsing being unavailable in that chat, YouTube blocking the fetch, or the video being private, age-restricted or region-locked. A refusal is more useful than the alternative, because at least you know nothing was read.
Can't ChatGPT just read the YouTube subtitles?
Captions are not sitting in the page text the way normal page copy is; the player serves them separately. Something has to open the transcript panel and hand that text over, whether that is you copying it or a plugin reading it from your browser tab.
Is pasting the transcript in myself good enough?
For one video, yes. It is free and it works. It stops being practical when you do it often, when videos are long enough to split across messages, or when you need timestamps to verify claims later, which raw pasted text usually loses.