Why use shards and not replica shards?


Let's assume I have k indices, each with one shard, running on a single node. The performance is good.
I want to add more nodes, and for that - I'll need replicas.

What is the benefit of using shards and not replica shards, assuming index size fits well in a shard?
The ability to change the amount of replicas after index creation is great and I'm wondering why not use it.
I am assuming that replica shards can answer search queries as the documentation states.

When would I regret defining only a single shard per index, other than in case of data growth inside an index?