How to Get ChatGPT, Claude or Gemini to Understand a Video
AI assistants read text, and a video's words are audio inside a player. Give the assistant the transcript instead: it works for every platform. How to do it, with prompts that work.
Short answer
The reliable way to get an AI assistant to understand a video is to give it the video's transcript as text. Whether an assistant can open a given video link depends on the product, the platform and login walls, and it changes often. A transcript works with every assistant, for every platform, and lets you check each answer against timestamps.
More and more people ask an AI assistant to "watch this video and summarise it". Sometimes it works. Often it quietly does not: the assistant summarises the title and description, or guesses from the page around the video, and the answer sounds right without coming from what was said.
Why a link is not enough
An assistant works with text. A video's words are not text on the page; they are audio inside a player. Whether an assistant can get at that audio depends on three things it does not control:
- The product. Some assistants can read YouTube captions through built-in tools. Few can fetch the audio of a TikTok, Instagram, Facebook or Vimeo video. What each one can do changes often.
- The platform. Many videos have no captions at all, so there is no text to read even when the page opens.
- Login walls. Some platforms serve automated readers a different page. We measured Facebook serving a logged-out reader a decoy page instead of the video for 7.4% of transcriptions in one week. See why a Facebook video won't transcribe.
The result is an answer you cannot verify, about a video the assistant may never have heard.
The reliable method: transcript first
1. Get the transcript. Paste the link into SubGrab. It uses the platform's captions when they exist and AI transcription when they do not, for YouTube, TikTok, Instagram, Facebook, Vimeo, Twitch and more.
2. Pick the format. Plain text is easiest to paste. Keep timestamps when you want to check the answers: an SRT line carries its time. See SRT vs VTT vs TXT.
3. Paste it with one clear instruction. Tell the assistant to answer only from the transcript, and to quote with timestamps.
Prompts that work with a transcript
- Summary you can check: "Summarise the key points of this transcript. After each point, give the timestamp where it is said."
- Answer only from the source: "Answer using only this transcript. If it does not say, reply that it does not say."
- Find the moment: "Where does the speaker first mention pricing? Quote the line and its time." See searching inside a video for the exact moment.
- Turn it into something: "Rewrite this as a 600-word article in the speaker's own words, keeping every figure exactly as stated."
How long is too long
Speech runs at roughly 150 words a minute. A one-hour talk is about 9,000 words, which most assistants take in one go. A ten-hour stream is around 90,000 words; split it by hour, or start from a summary. See transcribing a 10-hour livestream VOD.
Why this beats "watch this for me"
A transcript makes the assistant checkable. Every claim it makes points at a line you can read and a timestamp you can play. If you only want the short version, SubGrab's AI summary gives key points, a timeline and quotes straight from the transcript.
Frequently asked
Can an AI assistant watch a video from a link?
Sometimes, and it varies. Some assistants can read YouTube captions through built-in tools; most cannot fetch the audio of a TikTok, Instagram, Facebook or Vimeo video, and login walls stop automated readers. Pasting the transcript works in every case, which is why it is the dependable route.
How long a video can I give an assistant as a transcript?
Speech runs at roughly 150 words a minute, so a one-hour talk is about 9,000 words, which most assistants handle comfortably. A ten-hour stream is around 90,000 words; for that, work an hour at a time or start from a summary.
Should I keep timestamps in the transcript I paste?
Yes, when you want to check answers. A subtitle format such as SRT keeps a timestamp on every line, so you can ask the assistant to quote with times and then jump to that moment in the video to confirm it.