Match & match_phrase queries

match is the workhorse full-text query: it analyzes your input the same way the field was indexed, then OR-combines the resulting terms; match_phrase additionally requires those terms adjacent and in order.

Why it matters

These two cover the bulk of real search-bar traffic. match maximizes recall (“any of these words”), while match_phrase enforces exactness (“this exact wording”). Understanding operator, minimum_should_match, and slop lets you dial precision vs recall without resorting to brittle term queries on the wrong field.

How it works

Both run the field’s analyzer on the query string, producing terms looked up in the inverted-index.

Query	Default combine	Position-aware?	Key knob
match	OR (should)	no	operator, minimum_should_match
match_phrase	all, in order	yes	slop
match_phrase_prefix	phrase + last term prefix	yes	max_expansions

• operator: "and" — flips match to require every term, raising precision.
• minimum_should_match: "75%" — keeps OR behavior but demands a fraction of terms match.
• slop: 2 — allows up to 2 position moves, so match_phrase "quick fox" (slop 1) matches “quick brown fox”.
• Scoring — match sums per-term BM25; match_phrase requires positions stored in the index (index_options must keep them).

Example

{ "match": { "title": { "query": "wireless noise cancelling",
                          "operator": "and" } } }
// indexed terms must include all of [wireless, noise, cancelling]

{ "match_phrase": { "title": { "query": "noise cancelling", "slop": 1 } } }
// matches "noise cancelling" and "noise (active) cancelling"

With the default OR match, a doc containing only “wireless” still matches but scores lowest; operator:"and" drops it entirely.

Pitfalls

• match on keyword — no analysis happens, so it behaves like term; mixed-case input then misses.
• match_phrase_prefix is expensive — the last term expands to up to max_expansions (default 50) terms per shard; avoid for true autocomplete, use a completion suggester or edge n-grams.
• Stopword surprises — english analyzer drops “the”/“of”, so match_phrase "war of the worlds" matches “war worlds” unless you account for position gaps.
• slop is not edit distance — it measures token moves, not character typos; use fuzziness on match for typos.

See also

• full-text-search
• term-level-queries