Algorithm, Search & Distribution
Candidate generation
Also called: retrieval stage, candidate pool
Recommending from a catalogue of billions happens in two passes. The first pass, candidate generation, pulls a few hundred plausible videos for a given viewer using signals like watch history and what similar viewers watched. Only that shortlist goes forward to the ranking stage, where each one gets scored in detail.
The practical consequence for a creator is that a video has to be retrievable before it can be ranked at all. If nothing connects it to an existing audience or an existing viewing pattern, it never reaches the stage where a strong title and thumbnail could earn the click.