Stop manually scrolling through 50-page transcripts. TopicSplit finds where topics actually change โ in seconds, offline, for free.
A 60-minute interview = 8,000+ words. Finding where one topic ends and another begins takes 30+ minutes of scrolling.
Manual reading misses subtle topic shifts. Vocabulary changes signal transitions your eyes skip over.
Uploading transcripts to AI tools leaks sensitive participant data. IRBs hate it. You need offline.
Qualitative research already takes hundreds of hours. Why spend 20% of it just splitting text?
From Otter.ai, Rev, Zoom transcription, or any text source. Paste the full transcript into TopicSplit.
Higher = more granular segments (good for micro-analysis). Lower = broader themes (good for high-level coding).
TopicSplit analyzes lexical cohesion โ where content-word overlap drops between sentences, that's a topic boundary. Export clean markdown with ### Topic N headers.
Each segment becomes a thematic unit you can tag, code, and organize in NVivo, Atlas.ti, Dedoose, or your qualitative framework of choice.
"I paste every interview transcript into TopicSplit before coding. It finds the topic boundaries I'd spend 20 minutes deciding. The output is clean markdown I can code line-by-line."
Paste a transcript. Get thematic segments in 5 seconds. No install, no account, no tracking.
๐ Launch TopicSplitYes. The algorithm is language-agnostic. It detects vocabulary shifts between sentences โ works in any language.
Yes โ TopicSplit Pro is a CLI batch processor. Drop a folder of .txt transcripts, get a folder of segmented markdown. Pay-what-you-want (min $5).
TopicSplit runs 100% in-browser. Your data never leaves your machine. No server, no logging, no third-party processing. Your IRB will approve because there's nothing to audit โ the code is MIT and client-side.
NVivo uses AI/NLP that ships your text to a server (or requires expensive local models). TopicSplit uses simple lexical cohesion โ transparent, offline, free. You see exactly why each boundary was placed.
Paste the plain text. TopicSplit ignores timestamps and focuses on content-word shifts. If timestamps break sentences, the algorithm adapts โ it works on whatever text you give it.
Yes. MIT license. The segments are your original words โ TopicSplit just groups them. No attribution required (but appreciated).
Every dollar goes back into building more tools that respect your data and your time.
โ Tip $1 on Ko-fi ๐ณ Tip $1 on PayPal