In brief:
1. He'd play through a section of the game, coming up with certain key points to speak about in that section.
2. He'd then re-play through that section, recording the movements, then rendered out the movements in slow-mo so that everything would be smooth always.
3. Then he'd break it up into pieces. He has 5 seconds before the next thing so he tries to think of something that fits in that.
4. Once each piece is recorded with multiple takes, then they're put together and, with various tweaks to the audio, he'd put it together into single video and release.