For reading and quoting: TXT
Clean TXT is the words alone, one flowing text — best for reading straight through, pasting into documents, or word-counting. Note that auto-generated captions carry no punctuation, so clean text from a long ASR track reads as one continuous paragraph; creator-written captions read normally. Timestamped TXT puts [MM:SS] at each line — best when you need to cite the moment something was said. Both-in-one stacks the clean text first, then the timestamped copy in the same file, so you never wish you'd picked the other one.
For subtitle work: SRT and VTT
SRT is the lingua franca of subtitle files — numbered cues with millisecond ranges. Video editors (Premiere, Resolve, CapCut), media players, and translation tools all accept it. VTT is its web-native sibling: the format HTML5 video and most web players expect. Rule of thumb: editing software → SRT; embedding on a website → VTT. Both are always timestamped by nature.
For notes: Markdown
A heading plus timestamped bullets, rendering natively in Obsidian, Notion, and Logseq. If the destination is a knowledge base rather than a video editor, this is the one — full guide here.
For code: JSON
Each line as data — { "start": 0, "end": 4.2, "text": "…" } with video and track metadata attached. For scripts, RAG pipelines, and anything that would otherwise have to parse SRT syntax with a regex — full guide here.
Can't decide?
The default (TXT, both-in-one) is right for most people: readable now, citable later, one file. And because every format is generated from the same cached captions, requesting the same video in another format later is instant — no choice here is permanent.
Related: download a whole channel or read the FAQ.