YouTube to Reels
mining a back catalogue instead of starting from zero

Unlike a podcast, YouTube long-form often has visual teaching moments — screen shares, whiteboard explanations, product demos — that need to remain visually coherent when reframed to vertical, not just audio moments that happen to sound good. Prioritise moments where a single idea is fully explained visually within the frame, since the reframe has to preserve whatever is on screen, not just the speaker's face.

Favour the single most counter-intuitive or specific claim in each video over the video's actual thesis statement — a stranger scrolling responds to 'most advice on X is backwards, here's why' far more than to a balanced summary of a topic, even if the summary is what the full video is actually about.

Reframing 16:9 footage to 9:16 with a static centre crop routinely cuts off on-screen graphics, gestures and demo elements that were shot for a wide frame — this is the most common reason auto-reframing tools produce unusable output for anything beyond a static talking head. Active speaker tracking that follows movement, combined with manual repositioning when a graphic or demo needs to stay in frame, is required for any video with visual teaching content.

For screen-share-heavy sections, consider a split-frame layout (speaker top third, screen content bottom two-thirds) rather than trying to reframe the screen share alone into a vertical crop, which usually makes on-screen text illegibly small.

Process a long-form video's most-watched or most-engaged section first (check YouTube Studio's audience retention graph before selecting anything) rather than starting from the video's beginning — retention data tells you exactly where viewers actually stayed engaged, which is a strong proxy for what will work as a standalone clip.

Cap output at 8-15 shorts per long-form video even if more moments technically qualify — publishing every possible clip from one video floods a channel's short-form feed with material from a single source and can read as repetitive to anyone who watches more than two in the same session.

Watch every reframed clip specifically checking that no visual demo, graphic, or gesture central to the explanation is cut off by the vertical crop — this is a distinct failure mode from audio or caption QA and needs its own dedicated pass. Confirm captions don't overlap with existing on-screen graphics from the original video, which happens often when captions are added generically without accounting for what's already in frame.

Check that the clip still makes sense without its original video's title or thumbnail context, since a viewer on Reels or TikTok has no idea what long-form video this came from and won't seek it out unless the clip itself earns that curiosity.

Open YouTube Studio, sort your published videos by average view duration or retention percentage, not total views — high retention indicates content that held attention, which is your best raw material for clips regardless of the video's overall size. Watch the audience retention graph for your top 5-10 videos and note where retention holds flat or spikes rather than declining; those sections are your clip candidates.

Extract those sections with a free tool, using active reframing (most modern editing tools including CapCut offer speaker-tracking auto-reframe) rather than a static crop, and manually adjust any clip involving a screen share or demo. Publish 3-5 per week rather than dumping a backlog at once, and track which topics from your existing catalogue perform best short-form — that data should directly inform what you make more of long-form too.

Using a fully automated reframing tool with no manual review, which reliably cuts off demos, graphics and gestures central to the original explanation. Selecting clips based on which section of the long-form video the creator personally likes best, rather than what YouTube's own retention data shows actually held viewer attention.

Publishing a clip that assumes the viewer has context from the original video — a joke or reference that only lands if you watched the first ten minutes will fail completely as a standalone piece. And exhausting a single long-form video by publishing every conceivable clip from it in one week, which saturates the feed with one source and reduces the perceived variety of the channel.