What is the role of the payload index in query planning?

Beyond simply speeding up a lookup on an individual field, payload indexes feed directly into Qdrant's query planner, which uses information about which fields are indexed (and roughly how selective they are) to decide the most efficient overall strategy for executing a filtered vector search.

When a query includes a filter, the planner considers whether the filtered fields have indexes and how selective those filters are likely to be, choosing between narrowing the candidate set via the payload index first (efficient when the filter is highly selective) versus checking the filter condition inline during HNSW graph traversal (efficient when most points would satisfy the filter anyway, making upfront narrowing less valuable).

This is why Qdrant's own guidance emphasizes indexing fields that meaningfully narrow results, rather than indexing everything indiscriminately: an index on a low-selectivity field (one where almost every point matches any given filter value) provides little planning benefit while still costing memory and maintenance overhead, whereas an index on a genuinely selective field gives the planner a real opportunity to avoid unnecessary work.