The thirty-second answer
Caching happens at several distinct layers and they are worth separating: the browser, a content delivery network at the edge, an application-level shared cache, the database's own buffer pool, and precomputed results. Decide what may be stale and for how long — that is the real design decision, and it is answered per data type rather than globally. The standard pattern is cache-aside with deletion rather than update on write, because a cache can then only ever be stale or empty rather than wrong. What goes wrong is predictable: stampedes when a hot key expires and every request misses at once, an empty cache sending full traffic to a database that cannot take it, a single hot key saturating one node regardless of cluster size, and stale data outliving its usefulness because an invalidation was missed.
Say this first, then expand into the areas below as the interviewer steers.