Marketing & GrowthFree
Webinar Content Repurposing Clip Index
Cross the retention curve against the timestamped question log, and the moments sort themselves into clips, explainers and the ones that cost you the room.
Trewithen's summit ran three sessions, three hours of published recording, 604 people live out of 1,842 registered. The attendee report gives a concurrency count for every minute, so the recording has a second timeline running under it. Segmented at every speaker and subject change the transcript gives 96 spans, and 31 of them held or gained viewers. Twelve of those 31 also carry a cluster of questions, which means the audience stayed because they were not following. Cut as clips they re-perform the confusion. Written up instead, they answer it.
The box nobody looks at is the other diagonal: 14 spans where questions clustered and the audience left anyway. That is 53.8% of every question cluster in the log, and a retention sort never reaches them. They are not clips. They are FAQ entries, and a named fix to next quarter's script. Another 51 spans carry no signal at all, 51.7% of the runtime. Nothing in either export says anybody wanted them, and the honest instruction there is to leave it alone.
Then clearance and duplicates. Four spans lose both formats, two lose the video and keep the words, and seven of the shortlist turn out to be three explanations given more than once. Thirty-seven assets survive, 11 of them clips. Written for whoever inherits three hours of recording and a fortnight. Pair it with the follow-up list while it still converts, and the article built from the transcript once the spans are chosen. The explainers go through the writing review before they ship.
Why the peak is ambiguous, and what settles it
The ambiguity is not a theory. YouTube's own retention documentation says a spike is a moment that was rewatched or shared. It can mean the audience watched that segment more than the ones before it, or that the content was not clear and they had to rewatch a section. Two opposite findings, one curve. The question log separates them, because confusion leaves a timestamped trace and appreciation does not. So the sign of the concurrency change is never read on its own.
Caption timings are offsets rather than times. The WebVTT specification defines a cue's timings as the start and end offsets of the cue block, measured from the beginning of one particular media resource. Cut captions from the raw capture, publish a trimmed recording, and every cue is wrong by the length of the pre-roll. Trewithen's sessions were trimmed by 2:05, 6:20 and 9:45, so no single correction fixes more than one of them. The offset is found by matching one spoken sentence across both files.
The index has a publishable form, and it has rules. YouTube's chapter documentation asks for a first timestamp of 00:00, at least three timestamps in ascending order, and a minimum chapter length of ten seconds. The shortest signal span here is 0:46, so the floor never binds. Two of the three sessions open on a span that carries no signal, which means a 00:00 chapter gets written rather than inherited. That entry is the one nobody remembers to author.
How it works
Align the clocks
Find one spoken sentence in both files, measure the offset per session, convert every timestamp.
Segment the transcript
Spans break at speaker and subject change, so a moment has real edges rather than round numbers.
Cross the two signals
Concurrency change against question density, which puts every span in exactly one of four boxes.
Clear, then deduplicate
Check what cannot be republished, then keep one take of each explanation given more than once.
What you get
- Every span with in and out points in published-recording time rather than caption time.
- The offset measured per session, with the spoken sentence it was checked against.
- Concurrency across each span from the attendee report, and the sign of the change.
- Questions and poll responses joined to the span they landed on or the 90 seconds after.
- A clearance verdict per span, separating what blocks the frame from what blocks the words.
- Repeated explanations collapsed to the take that held best, with the dropped takes named.
Common questions
Can a clip tool not just find the best moments?
It will return a ranked list, and the ranking is a black box. Trewithen's top span by viewers held is a two minute stretch where 375 people stayed and one question was asked. The span next to it held 362 and shed 9.3% while four questions came in. Both look like peaks. Only one is a clip.
Why does a couple of minutes of offset matter?
Because a moment is shorter than that. The median span here runs 1:50, and the smallest of the three offsets is 2:05, so an unaligned timecode lands past the end of the thing it was pointing at. 57 of the 96 spans are shorter than that smallest offset. Session C is out by 9:45.
Is a question cluster not a good sign?
It is a sign of interest and of unclear delivery at the same time, and the retention sign tells you which. Questions with the room holding means they wanted more. Questions with the room emptying means the explanation failed, which is worth knowing before you script the same section again.
What happens to the parts with no signal?
Nothing, deliberately. 51 of the 96 spans are 51.7% of the runtime and carry neither a held audience nor a question, so there is no evidence anybody wanted them. Producing assets from them is how a repurposing exercise turns into a month of work with nothing to show. Whether the event was worth running at all is a separate calculation, in the sponsorship evaluation.
Why check clearance before editing rather than after?
Because six of Trewithen's 31 peaks failed it. Three had a customer identifiable on camera with no written release, one was a guest whose agreement covered the live session only, and two had an unannounced price on the slide behind the speaker. Those last two still ship as written pieces.
We ran three sessions. Do we get three times the assets?
No, and that is the point of the pass. Seven spans in the shortlist were three explanations given in more than one session, so four takes get dropped and the best one survives. The migration explanation held 61% in session C against 43% in session A, so session C is the take.
What actually gets delivered?
A Sheet with one row per span carrying both timecodes, length, concurrency, questions, box, clearance verdict and duplicate group. A Doc with the clip list, the explainers drafted, and the FAQ entries. Quotes worth pulling go through the customer story tool.
Webinar Content Repurposing Clip Index
Fill in the form and your workspace opens with the work already underway.