Verified against Claude · 2026-07-23
Generate accurate chapter timestamps from a raw transcript
Turns a raw, timestamped transcript into real chapter markers cut at genuine topic-change points, with search-usable titles — a job auto-chapters frequently gets wrong on transcripts with tangents or false starts.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are generating YouTube chapter markers from a raw, timestamped transcript. Chapters must be cut at genuine topic-change points in the actual content, not at mechanically even intervals, and each chapter title must work as a standalone search snippet — this is specifically needed because YouTube's own auto-chapters feature frequently misplaces cuts around tangents, false starts, and re-explanations, which is exactly the kind of judgment call this prompt is for.
RAW TRANSCRIPT WITH TIMESTAMPS
[0:00] Alright so today... [0:38] okay let's actually talk about the three settings that matter... [4:12] now before I show you the results, wait, I should back up... [4:40] so the three settings again are... [9:05] let's look at what actually changed...
VIDEO TOTAL LENGTH
14:32
TARGET SEARCH PHRASES FOR THIS VIDEO
DaVinci Resolve color grading settings
KNOWN STRUCTURE NOTES
The section starting around 4:12 has a false start where the speaker restarts the explanation — the real explanation doesn't begin until about 4:40
CHAPTER RULES
Identify a new chapter boundary only where the actual subject genuinely changes — a new sub-topic, a new step in a process, a shift from explanation to demonstration — never at a fixed interval like every two minutes regardless of content. If the speaker restates or circles back to an earlier point rather than introducing something new, that restatement belongs inside the existing chapter, not as a new one; a chapter list that fragments a single continuous explanation into three near-duplicate chapters is worse than useful, since it makes the jump-to menu look cluttered and imprecise. If the transcript contains a false start, a correction ("wait, let me redo that"), or an aside that gets cut in editing, do not create a chapter around it and do not let it shift your read of where the real topic boundary is — read past it to where the actual content resumes. The first chapter must start at 0:00 exactly, per YouTube's own requirement for chapters to register at all, even if the true content start is a few seconds later. Every chapter must be at least 10 seconds long; if the true topic segments are shorter than that, merge them into the nearest adjacent chapter rather than producing a chapter YouTube will refuse to register as its own entry. Title each chapter with a specific description of its actual content in 3-6 words — something a person scanning the jump-to menu, or seeing it as a snippet in a search result, could use to decide whether that segment is what they're looking for — never a generic label like "Continued" or "More Tips." Where a chapter's content matches one of the target search phrases closely, prefer phrasing that captures it naturally, without forcing an awkward match.
OUTPUT FORMAT
A list of timestamp-title pairs in the exact mm:ss (or h:mm:ss for videos over an hour) format YouTube auto-detects, one per line, starting at 0:00. After the list, one line for any place you merged what looked like two topics into one chapter, and why, and one line for any place you deliberately ignored a false start or aside in the transcript when placing a boundary.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
YouTube's automatic chapter generation runs on the video's own transcript and tends to place boundaries at surface-level cues — a pause, a filler-word cluster, an even time interval — which is precisely why it misfires on exactly the material this prompt is built to handle: a false start, a mid-sentence correction, or a speaker circling back to restate a point earlier in the transcript all look, to a shallow pattern-matcher, like the start of something new, when a careful read of the actual content shows they aren't. Forcing chapter boundaries to require a genuine subject change, and explicitly instructing the model to read past a false start to where content actually resumes, targets that exact gap — this only works because the model is given the real transcript to reason over content, not a signal-only heuristic guessing from pauses and word frequency the way YouTube's automated feature does. The minimum 10-second chapter length rule isn't a stylistic preference — YouTube enforces its own minimum duration for a chapter to register as a distinct, clickable entry at all, and a chapter list that ignores this produces markers that either get silently merged by YouTube's player or, worse, fail to render as chapters entirely, which means a set of timestamps that look correct in a text list can quietly not exist as usable navigation on the actual video. Requiring the 0:00 start exactly, regardless of a few seconds of true dead air, mirrors YouTube's own documented requirement that the first chapter must begin at zero for the rest of the list to register — a chapter list starting at 0:04 because that's technically where content begins is a well-intentioned mistake that breaks the whole feature. The specific, non-generic titling rule matters for a second reason beyond the in-player jump menu: well-titled chapters are what YouTube surfaces as direct timestamp links inside Google search results for the video, so a chapter titled "More Tips" is not just unhelpful in the player's own menu, it is a wasted opportunity for that exact segment to independently rank and get clicked from a search results page for whichever specific thing it actually covers.
Verified against
Claude Sonnet 4.6 · 2026-07-23
ChatGPT GPT-5.1 · 2026-08-04
Changelog
- 2026-07-23 — Initial publish, verified against Claude (Sonnet 4.6) and ChatGPT (GPT-5.1).
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
