How to build a clip library you can actually re-use
Four hundred exported clips in a folder is not a library. A library is one row per clip, carrying the hook line, a short tag set, and a record of everywhere it has already run.
8 minute read
A folder is not a library
A folder answers one question: where is the file. A library answers a different one: what do I have. Those sound similar and behave nothing alike, which is why people with three hundred finished clips on a drive still open the editor and cut a new one. Nothing in the file system tells you which clip covered pricing objections, which one already ran on LinkedIn in February, or which two are near-duplicates of each other. The cheapest thing that fixes it is a spreadsheet with one row per finished clip. Not per project, not per source recording, per clip — because the clip is the unit you publish, reuse and eventually retire. If you never build anything more sophisticated than that sheet, you have already solved most of the problem.
One row per clip, and the fields that earn their place
Keep the column list short enough that filling a row takes under a minute, or it will not get filled. Eight fields cover almost every real use: an ID, the source recording and its timecode, a link or path to the master file, the verbatim opening line, duration, a short tag set, which aspect ratios you have rendered, and where it has been posted. Add a free-text notes column for the thing that does not fit. Resist everything else. Fields for sentiment, predicted performance, campaign codes and internal ratings feel useful when you design the sheet and are empty within a month, which is worse than not having them because they make the library look unreliable. A library people trust is one where every column is actually populated.
The hook is the index, not the title
Clip titles are invented after the fact and they go stale fast, because the title you wrote in March describes how you were framing the topic in March. The spoken opening line does not change, and it is how your memory actually stores a clip — you remember the line, not the file name. So put the first sentence in the row verbatim, exactly as it is said on camera, rather than a summary of it. Two things fall out of that. Searching the library becomes text search over the thing people actually said, which finds clips you had forgotten existed. And when you go looking for a hook to reuse in a script or a caption, the phrasing is already there in the voice it was delivered in, rather than paraphrased into something flatter.
Three tag axes, and a vocabulary you lock
Tag sprawl kills libraries faster than neglect. The failure mode is documented well outside video: Fotoware's guidance on digital asset management makes the case for a controlled vocabulary, a fixed, authorised list of terms, precisely because free typing produces five names for one thing and a search that finds none of them. Its example is a room that gets tagged foyer, hallway, corridor, vestibule and entrance by five different people. Pick three axes and no more. Topic, using whatever content pillars you already publish under. Job, meaning what the clip does for the viewer — teaches, argues, proves, tells a story. And format fit, which is really about length and whether it needs context to make sense. Write the permitted values down once, use a dropdown rather than free text, and revisit the list quarterly rather than mid-session.
The posted log is the field that pays for the whole thing
If you keep only one thing beyond the clip itself, keep the record of where it has run. Four values per posting: platform, date, the public URL, and one result number that you will actually look at again. This is the column that answers the question that made you build the library — have I already used this — and it answers two more for free. It tells you which platform has never seen a clip that did well elsewhere, which is usually several of them. And it gives you a route back from a comment or a DM to the master file, which otherwise means scrubbing your camera roll. Keep it as a growing list rather than a single date field, because the whole point is that a clip gets posted more than once.
When a clip can run again
The honest answer is that nobody has published a study establishing a correct gap, so treat any specific number you see as a housekeeping convention rather than a finding. What is measured is how short the window is. Metricool's 2026 YouTube study, built from 799,718 videos across 71,177 accounts, found 83% of interactions arrive within the first ten days of publishing. Its 2026 TikTok study, covering 2,314,756 posts from more than 92,000 accounts, found creators published 72.10% more videos year on year — the feed refills faster than anyone can scroll it. Between those two, a clip that ran once in March is functionally invisible by May to everyone who missed it, and most people missed it. A ninety-day cooldown per platform is a reasonable default. The real test is not elapsed time, it is whether the clip still reads as current.
Variants are siblings, not separate clips
One moment usually ships as several files: a 9:16 render for TikTok, Reels and Shorts, a 16:9 or square version for LinkedIn, sometimes a shorter cut with a harder opening. Filing each of those as its own library row is the most common way these systems start lying to you, because the posted log fragments across four rows and you lose the ability to see that the moment has already run everywhere. Keep one row for the moment, and a variants column listing the renders that exist. Platforms genuinely differ in what they reward — Metricool's 2026 studies covering the same two-month window found Instagram views up 26.56% year on year while TikTok views fell 31.30% — so per-platform variants are worth making. They are just not separate assets.
Fill it at cut time, not later
Every clip library that dies, dies of the same thing: cataloguing was a separate job from cutting, so it happened last and then stopped happening. The fix is to write the row while the clip is in front of you, when the hook line, the source timecode and the topic are all on screen anyway. Thirty seconds per clip during review costs nothing. Reconstructing the same information from a folder of exports three months later costs an afternoon, and you will not do it. This is also the argument for cutting in batches: reviewing twenty clips in one sitting means twenty rows written in the same ten minutes. FrameOS is built around that pass — it surfaces candidate moments with their transcript and timecodes, so what you need in the row is already next to the clip when you approve it. 300 credits for 3 days · no card.
FAQ
What should a clip library contain?
One row per finished clip, carrying an ID, the source recording and timecode, a link to the master file, the verbatim opening line, duration, a small fixed tag set, the aspect ratios you have rendered, and a running log of where and when it has been posted. Keep the column list short enough that a row takes under a minute to fill, or it will not get filled.
How should I tag video clips so I can find them later?
Use three fixed axes and a locked list of permitted values: topic, the job the clip does for the viewer, and format fit. Digital asset management guidance from Fotoware makes the case for a controlled vocabulary rather than free typing, because free tags produce several names for the same thing and searches that return nothing. Use dropdowns, and review the list quarterly rather than adding terms mid-session.
How often can I repost the same clip?
No study establishes a correct interval, so any specific number is a convention. What is measured is that the window is short: Metricool's 2026 YouTube study of 799,718 videos found 83% of interactions arrive in the first ten days. Ninety days per platform is a workable default. The binding test is whether the clip still reads as current, not how long it has been.
Do I need a proper tool or is a spreadsheet enough?
A spreadsheet is enough for most people and is the version that actually gets maintained. Move to a database or asset manager when two things become painful: several people needing to add rows at once, and wanting to preview clips without opening files. Until then, extra software mostly adds a place for the library to be out of date.
Sources
Related pages
Keep reading
More Clipping & Repurposing →How to livestream and clip it afterwards
Twitch deletes most streamers' past broadcasts after seven days and Facebook deletes live video after thirty. What you record locally, and how, decides what you can still post next week.
8 minute readHow to turn a blog post into a video
A 1,500-word article read aloud is roughly eight minutes of video. Here is how to work out which single idea in it earns the runtime, and why most lists do not survive the jump.
8 minute readHow to repurpose a newsletter into video
Your newsletter is better video material than your blog, because it already has a voice and a format. Here is which issues convert, and what the Substack data says about who this actually helps.
8 minute readTurn one long video into a week of posts
300 credits for 3 days · no card.
Start free