Which join is used in data blending?

A blend behaves like a left outer join from the primary data source.

Rows from the primary are all retained; matching aggregated values from the secondary are attached where the linking field matches, and appear as null where it does not. Rows that exist only in the secondary are not shown.

It is worth being precise in an interview that this is the *behaviour*, not the mechanism. No join is executed — each source is queried and aggregated separately, then the results are matched. That is why blending copes with different granularities where a real join would duplicate rows.