A-roll is the footage that carries the words, usually a person talking to camera. B-roll is the footage you cut to while the words play: the product in use, the close-up, the reaction. Most ads are A-roll audio with B-roll on screen.
A-roll carries the words. B-roll carries the picture.
A-roll is the creator talking to camera: the testimonial, the explanation, the claim. B-roll is everything you cut to while that audio plays: the product in use, the texture close-up, the before and after, the reaction.
Most ads are A-roll audio with B-roll on screen. The creator says "it absorbed in ten seconds" (A-roll) while you show the serum on her cheek (B-roll). The viewer hears one thing and sees the proof of it at the same time.
This is why an editor's brief has two halves. "The take where she says X" is an A-roll search. "A dropper close-up on a clean background" is a B-roll search. A library that only has transcripts can answer the first and not the second, which is why the visual description of each moment matters as much as the words.
Let's say you cut a cookware ad, 15 seconds. A-roll: "I stopped using oil completely." B-roll over it: egg sliding off the pan, then the creator wiping the pan with a paper towel. Two B-roll segments, one A-roll line, no scrubbing if all three are tagged.
Yes. A talking-head take often has a B-roll moment inside it, like the creator holding up the product. In a segment library the two get tagged separately even if they came from the same file.
The 10-15 actions specific to your category. Cookware: searing, food release, wiping clean. Skincare: dispensing, texture on fingertips, the mirror check. Give creators that list and the B-roll comes back usable.