Which transcript format for which job

Markdown is for reading and linking. Headings, bold, bullet points - it renders in Obsidian, Notion, a static site, or GitHub without extra work, and it stays plain enough to diff and version.

Use it when the transcript is going into a personal knowledge base or a wiki page next to other notes. It is the format you write links and citations against, not the format you feed to a model.

Plain text has no markup to strip and no formatting for a model to trip over. Paste a plain-text transcript into a chat window or a prompt and every token is transcript, not syntax.

It is also the smallest format, which matters when you are up against a context window. If the job is "summarize this" or "answer questions about this," plain text is the least friction.

SRT and VTT exist for one job: timed text under a video. Both encode start and end times per line, and both are read natively by video players, editors, and most platforms you would upload footage to.

SRT is the older, simpler format and the one most editing software expects. VTT adds styling and positioning and is the web standard - the format YouTube and browser video players read directly.

JSON is for code, not for reading. Segments, timestamps, and speaker labels come out as structured fields a script can loop over, filter, or feed into another system - a search index, a database, a second model call.

If you are building something on top of transcripts rather than reading one, start here. It is the only format that keeps speaker and timing information machine-addressable instead of embedded in text.

Every transcript in Vocateca - local file, YouTube video, podcast episode, Instagram video - exports to Markdown, plain text, SRT, VTT, and JSON from the same source. Per-segment timestamps and speaker labels carry through to every format that supports them.

You do not pick a format when you transcribe. You pick it when you know what the transcript is for.