The choice is usually made by habit or by what is easiest to shoot. It is worth making deliberately, because the two formats carry different material and break at different points.
What talking head carries
Argument. A claim that needs a person standing behind it, a position, a correction, anything where the authority of the speaker is part of the content. It is the only format where who is saying this is visible information.
It is also the format with the lowest production floor and the highest delivery floor. Nothing hides a flat read. There is no cut to a shot of the thing, no motion to carry a slow sentence — the face is the frame for the whole duration.
Where it breaks: in the middle. The opening usually survives on novelty and the ending on resolution; the sag is at fifteen to twenty-five seconds, when the viewer has understood the claim and the argument has not yet turned.
What B-roll carries
Process and evidence. Anything where seeing the thing is the point: the build, the before and after, the mechanism, the place. Cuts create their own momentum, which does part of the retention work that delivery has to do in a talking head.
Where it breaks: at the narration. B-roll-led video fails when the footage is doing all the work and the voice is describing what is already visible. The viewer is being told what they can see, which reads as padding even when the footage is good.
Talking head fails in the middle. B-roll fails in the voice. Neither fails at the edges, which is where most people look.
The hybrid, and its specific trap
Most strong short-form is a talking head with B-roll cut over it. The trap is cutting away at the wrong moment: leaving the face precisely when the speaker reaches the line that needed a face behind it.
A workable rule is to cut away on description and cut back on claim. The footage illustrates; the person asserts. Videos that do the reverse feel oddly weightless, and it is hard to name why from the outside.
The format decision that is actually about demotions
One practical consideration outranks the creative one. B-roll is frequently sourced — stock, screen recordings, clips from elsewhere — and sourced footage is where watermarks, letterboxing and re-compression enter a project. All three are named by Instagram as reasons it reduces visibility.
A talking head shot on a phone has none of those risks by construction. That is not an argument against B-roll; it is an argument for auditing the footage you did not shoot.
How to choose
Ask what the video has to prove. If it has to prove a claim, the person should be on screen. If it has to prove a result, the result should be. If it has to do both, lead with whichever is less believable without evidence — that is the one that needs the screen time.
Then check the choice against your own history rather than against the format that is fashionable this quarter. See why the bigger number is usually the worse lesson for why the fashionable one is misleading.