Libi
DocsDownload
← libi · blog
13 january 2026·5 min· editing· short-form· workflow· creators

The B-roll formula: when, how often, and where to cut

B-roll isn't decoration. It's a retention tool with a measurable cadence — and most creators are overdoing it.

Pull up a recent MrBeast video and count visual changes for one minute. You'll hit somewhere between 8 and 20 — a cut, an angle swap, a B-roll insert, a zoom. Retention-editing guides recommend a visual change every 15-25 seconds and a hard cut every 3-7 seconds for long-form YouTube. Casey Neistat is on record about trying to make sure something visually interesting happens every 30 seconds or less. These aren't aesthetic choices. They're retention mechanics with numbers attached.

Most creators get the numbers wrong in the same direction: too much B-roll, picked too generically, cut at the wrong moment.

Why B-roll exists at all

B-roll is not there to look cinematic. It's there because a static talking head loses attention faster than a moving image. The retention curve is the product, the visuals are the lever.

OpusClip ran a controlled test on an 8-minute video and published the numbers: the no-B-roll version held 66% to the end. The 40% B-roll coverage version held 83%. The 70% coverage version dropped back to 73%. There's a curve, and it peaks somewhere between 35-50% coverage. After that, you're hiding the speaker more than supporting them.

The same study showed CTAs delivered straight to camera converted 40% better than CTAs laid over B-roll. People trust faces. Cover the face during the ask, lose the sale.

Cadence by format

Short-form (Reels, Shorts, TikTok): a visual change every 2-4 seconds. Vidpros' clip-length guide goes shorter still, recommending 1-2 second clips for TikTok and Reels and citing data that average clip lengths around 2.5 seconds correlate with materially higher completion rates. The Hormozi style is jump cuts on every breath plus a B-roll layer on top — because dead air on a 60-second video is fatal in a way it isn't on a 12-minute one.

Long-form YouTube: a visual change every 8-15 seconds, but cuts can sit longer. B-roll inserts in OpusClip's tests landed in the 3-8 second range — long enough to register, short enough not to lose the audio thread.

Talking-head explainer (Ali Abdaal style): jump cuts on the A-roll handle most of the pacing; B-roll layers on top of specific nouns and demonstrable concepts, not on every line.

Cinematic vlog (Casey Neistat style): the music track is picked first, B-roll cuts land on beats. The cadence is whatever the song says it is.

The four kinds of B-roll, ranked

  1. 01Hand inserts and prop close-ups. You pour the coffee. You hold the product. You point at the chart. These are the highest-performing because they're literal — they show what the audio is describing, in the same room, with the same lighting.
  2. 02Environment shots — the room, the city, the laptop screen. These ground the viewer in place. Useful, cheap, abundant.
  3. 03Reaction or second-angle shots of yourself. A dolly shot from another camera. Pure pacing tool.
  4. 04Generic stock footage. The drone shot of a city you've never been to. The smiling stock model.

That last category is the contrarian piece.

Generic stock B-roll is worse than no B-roll

This is where I disagree with most AI editing tools. OpusClip's own marketing concedes that generic stock footage rarely performs as well as custom B-roll. My read of the data is stronger: bad stock B-roll actively costs you watch time. Wistia is direct about it — if the connection isn't immediately clear, B-roll becomes disruptive. A clichéd handshake clip under a sentence about "partnership" is a tell. The viewer registers it as filler, and filler reads as low-effort, and low-effort gets swiped.

Tools that auto-insert B-roll from stock libraries every 3 seconds optimize for a metric (coverage) that doesn't match the metric you actually care about (retention). A talking head that holds attention beats a talking head buried under irrelevant skyline shots.

Where to cut

There are three reliable cut points. Cut on a claim — the moment the speaker says the noun, show the noun. Wistia's principle: B-roll should be tied directly to the surrounding narrative, and timed so the visual lands as the speaker mentions it. "I bought a new camera" — show the camera on the word "camera," not two seconds later.

Cut on a beat shift — the speaker moves from setup to payoff, from problem to solution, from past tense to present. The change in narrative is where a visual change feels earned.

Cut on a breath — the natural micro-pause between sentences. This is where Hormozi-style edits live. The breath gets removed and replaced with a B-roll insert that bridges the audio. Done well, the viewer never registers that air was cut.

Avoid cutting mid-clause. Avoid cutting just because you've gone 5 seconds without a change — that's the cadence rule eating the meaning rule, and meaning wins.

One takeaway

If you take one number from this: aim for 35-50% B-roll coverage on long-form, with each insert lasting 3-8 seconds, cut on the noun. Anything past 60% coverage and you're hiding your strongest asset — your face — for diminishing returns.

Sources