Why a TikTok URL is dead weight in a ChatGPT prompt

Two separate walls. The first is access: TikTok serves its video pages through a client-side app shell, so even when ChatGPT browsing fetches the URL it comes back with a title, a caption and a share count, not a script. The second wall is the one people miss. The captions you see burned across the frame while watching are pixels drawn into the video image. They are not stored beside the video as selectable text, there is no export button, and nothing can copy them. That is the whole reason TikTok will not give you the transcript no matter which menu you dig through.

So the model gets a caption and a hashtag list, then fills the gap with something that sounds right. Asking it to "summarise this TikTok" produces confident fiction, which is worse than an error.

The thirty second workflow

Open TikTok Transcript, paste the public link, convert. The audio is fetched and transcribed, punctuation and paragraphs are restored, and you get a Markdown document. Copy it to the clipboard or download the .md and attach it. Everything happens in the browser tab: there is no ChatGPT plugin, no extension and no connector to install, and none is needed because a short clip is a few hundred words that paste inline without touching your context budget.

For clips over a couple of minutes, or for the long-form videos on other platforms, the same output shape comes out of Video to Markdown and YouTube to Markdown, so a mixed prompt of TikToks and YouTube uploads reads as one consistent document set.

Prompts that only start working once GPT has the text

"Rewrite this hook five ways, keeping it under nine words." "This clip did 400K views and the last one did 9K. Compare the first three sentences of each and tell me what changed." "Pull every factual claim and flag the ones that need a source before I repost this." All of these are impossible against a link and trivial against a transcript, because the model is finally reading the thing you are asking about.