The first lecture asked what you're making when you record your grandmother, and whether it's worth the trouble. It answered: a conversation, and the object it leaves behind. And it thought them worth the trouble. Here is the trouble, in the order you'd take it.
Beforehand, you choose the person, and ask ahead of time. You tell her what it's for and who will hear it. You talk once without the recorder, write a page of topics, and rehearse with the machine.
On the day, the microphone is close, the room is quiet, and it's the two of you. Before the recorder goes on, you promise only what depends on you, as far as the law allows. Then you say the date, the place and the names. She says, in her own words, what you're doing and what she wants done with it. You ask short open questions, follow where she goes, wait through her silences, and supply nothing. If the story is hard, how much of it she tells is up to her.
Afterwards, you play back the end before you leave, and make a copy that day. She hears the recording before anyone else does. And you look after it: three copies, each one played, an index, and a sheet of paper with the family's papers.
Some things stayed open, among them how far an old memory's details can be trusted and who owns a recording. I don't think any of them is a reason to wait.
Two things ran through all of it. You are part of what she tells, and what you make is for someone who wasn't in the room. The first lecture's advice still stands: choose the person, and ask.