Short-form video: Reels, Shorts and TikTok

How to build a hook that survives the first second

A hook is a promise made fast enough that a thumb stops moving. It has to work on three channels at once: what the viewer sees, what they hear, and what the on-screen text says. If any one of those is slow, the other two rarely save it.

The visual hook is the first frame. It should already show motion, a face mid-expression, or a strange object. A clip that opens on a static logo or a person settling into their chair has spent its most valuable second on nothing. The spoken hook is the first sentence, and it should state the payoff, not introduce it. "Here is why your ads stopped converting" works. "Hi everyone, welcome back, today we are going to talk about advertising" does not, because the promise arrives four seconds late.

The text hook is the caption card in the opening frame. It exists because most viewers arrive on mute. It should be short enough to read in under a second, so roughly three to seven words. We usually write it as a specific claim, an unexpected statement, or a question the viewer already has.

• Open on motion or a face, never on a logo or a settling shot
• State the payoff in the first spoken sentence, do not introduce it
• Put a three to seven word text hook in the opening frame
• Cut the first two seconds of most raw takes, they are almost always throat clearing
• Avoid promising something the clip does not deliver, retention collapses at the reveal

Reading a retention curve and what each shape tells you

Retention graphs are the only honest feedback short-form gives you, and the shape matters more than the average. Every platform shows some version of it: what percentage of viewers were still watching at each second.

A steep drop in the first two or three seconds means the hook failed. The content itself is untested, because almost nobody saw it. Fix the opening before touching anything else. A gradual, even decline is normal and healthy, and it means the clip is holding attention roughly as expected. A sudden cliff in the middle usually marks a specific moment: a long pause, a tangent, a section where the visual stopped changing, or the point where the viewer realised the promised payoff was not coming.

A curve that flattens or ticks upward near the end is the best signal there is. It means people rewatched or stayed for a conclusion, and platforms reward that heavily. The practical move is to trim ruthlessly toward wherever the cliff sits. If viewers leave at 18 seconds, a 40 second clip is a 20 second clip with 20 seconds of damage attached.

Safe zones: where platform interface elements cover your video

Every vertical platform lays its own buttons, captions and profile details over the video. Anything you place under those elements is invisible in the feed even though it looks fine in your editing software. This is the single most common technical mistake in short-form.

The pattern is consistent across TikTok, Reels and Shorts even though exact pixel values differ and change with app updates. The right edge carries the action buttons: profile, like, comment, share, sound. The bottom carries the account name, caption text and often a music ticker. The top carries the status bar and sometimes a following or for-you tab. The middle of the frame is the only reliably clear area.

We work to a conservative safe zone rather than chasing exact figures per app, because the apps move them. In practice that means keeping essential text and faces inside roughly the central 80 percent horizontally and away from the top and bottom bands. It costs a little composition freedom and saves you from re-exporting a whole batch when a caption sits under the share button.

• Keep critical text and faces away from the bottom quarter of the frame
• Keep the right edge clear, that is where every platform stacks its buttons
• Leave the top strip clear for the status bar and platform tabs
• Preview on an actual phone in the actual app before publishing a batch
• Design one safe layout that works across all three platforms rather than three versions

Aspect ratios, resolution and reframing from horizontal footage

Vertical short-form is 9:16, which in practice means 1080 by 1920. Square, 1:1, still appears in some feed placements, and 4:5 is the tallest ratio that reliably works in an Instagram feed post as opposed to Reels. If you only make one size, make 9:16, because it can be cropped down but not out.

Converting horizontal footage to vertical is a reframe, not a crop. A straight centre crop of a 16:9 interview usually cuts off the subject or leaves them jammed to one side, because they were not framed for it. Proper reframing tracks the subject across the shot, adjusting position over time so faces stay in the upper third where the eye naturally goes.

When the shot genuinely cannot be reframed, for example two people sitting apart, the options are a split-screen stack, a blurred background fill with the original shot centred, or cutting between the two speakers so only one is on screen at a time. The last option is nearly always the best, because it fills the frame and adds pace at the same time. The blurred fill is the weakest, because it wastes half the screen.

Caption styling that works on mute

Short-form captions are a design element, not an accessibility afterthought. Most viewing happens with sound off, so the captions carry the message. That means legibility and rhythm matter more than accuracy of transcription style.

The rules that hold across platforms: a heavy sans-serif weight, generous size, a stroke or drop shadow so text survives a light background, and one to three words per line rather than full sentences. Word-by-word or short-phrase captions timed to speech read faster than a static block, because the eye is not scanning a paragraph while also watching the picture. Emphasis colouring on one keyword per phrase pulls attention where you want it, but it stops working if every other word is coloured.

Position matters too. Captions sitting slightly above centre stay clear of the platform interface and stay near the face, so the viewer is not moving their eye between two parts of the screen. Automatic transcription is a good starting point and a poor finishing point. It gets brand names, industry terms and numbers wrong often enough that every clip needs a human pass.

• Heavy sans-serif, large size, stroke or shadow for contrast
• One to three words per line, timed to the speech rhythm
• Position slightly above centre, clear of the interface and close to the face
• Colour one keyword per phrase at most
• Always proofread automatic transcription, especially names and numbers

Turning one long video into a month of clips

Repurposing long-form is the cheapest way to sustain a short-form schedule, but it only works if you select for self-contained moments rather than slicing at even intervals. A clip has to make sense to someone who has no idea what came before it.

The selection test is simple. Does this segment contain one complete idea, does it start on a strong sentence, and would it still land with no setup? A one-hour podcast typically yields somewhere between eight and fifteen segments that pass. Everything else is context that only works in the full episode. Cutting at arbitrary sixty second marks produces clips that open mid-sentence and end nowhere, which is why so much repurposed content underperforms.

Once selected, each clip still needs the full treatment: a new hook at the front, tightened pacing with pauses and filler removed, vertical reframing, styled captions, and its own text hook. That last step is where most repurposing effort fails. A raw excerpt with burned captions is not a short-form video, it is a trailer for a video nobody asked for.

• Select for complete ideas, never for fixed time intervals
• Give each clip its own hook, do not rely on the parent video
• Cut pauses and filler words, long-form pacing is too slow for a feed
• Reframe to vertical with subject tracking, not a centre crop
• Batch the whole set in one session so styling stays consistent