Short video is the format Claude is oversized for, in a good way
Most transcription advice is about fitting a 90 minute podcast into a context window. Short-form video inverts the problem. One clip is trivially small, and the interesting question is never "what did this video say" but "what do these forty videos have in common". That comparison needs all of them in the prompt at the same time, which is exactly the shape Claude is built for.
Convert each clip at TikTok Transcript, then concatenate the Markdown files under one heading per video with the view count in the heading text. Claude reads ## 412K views, posted 2026-06-14 as a real section boundary and will happily sort, group and contrast across them. Ask it to rank the openings by specificity, or to find every clip where the call to action came before the payoff.
The Projects pattern for a competitor library
Claude Projects is the right home for a corpus you keep adding to. Convert a clip, download the .md, and upload it to the Project knowledge base by hand. To be explicit: there is no MDisBetter connector, MCP server or Claude integration, so this step is a manual upload each time. In exchange, every conversation in that Project starts with the whole library loaded, and you can ask "how has this account changed its opening line over the last six months" without re-pasting anything.
Long-form sources belong in the same Project. Use YouTube to Markdown for the creator's long videos and Video to Markdown for anything you have as a local file, so short and long content sit side by side in one consistent format.
What the transcript cannot tell Claude
The transcript is the audio, and only the audio. Text typed onto the screen is not read, the visual edit is invisible, and a clip that is music over silent b-roll comes back empty. When the message lives in the on-screen text rather than the voiceover, screenshot the frame and run it through Image to Markdown instead. Being clear with Claude about which signal it is holding stops it from confidently explaining edits it never saw. More on the trade-offs in how to transcribe a TikTok video.