A video on a card is watched differently from a video in a feed. The person holding it is standing still, phone in one hand, card in the other, usually with someone next to them. They will watch it two or three times in a row. That changes what makes a good clip.
Keep it short
Between five and twenty seconds. Long enough to hold a moment, short enough that the loop back to the beginning feels like a gift rather than a chore. A two-minute clip is not watched twice, and holding a phone over a card for two minutes is tiring.
Hold the phone still
The card is already moving in the viewer's hand — the AR layer keeps the video pinned to the paper, but if the footage itself is also swinging around, the effect reads as chaos. Brace your elbows, or rest the phone on something. A static shot of something moving beats a moving shot of everything.
Shoot toward the light, not away from it
Faces lit from the front. A window behind your subject turns them into a silhouette, and the video will be watched on a phone held at arm's length in someone else's lighting, which never improves it.
Say the thing
The clips people keep are almost never scenic. They are someone looking into the camera and saying the sentence they would have written in the card. If you are unsure what to film, film that.
Sound matters more than you expect
It plays out loud, in a room, often with several people leaning in. Wind noise and a distant voice ruin an otherwise perfect clip. Get close, and check the audio before you upload rather than after the card is printed.