Full-text search

Full-text search matches text fields by analyzing the query the same way the field was indexed, then ranking hits by relevance rather than returning a flat yes/no set.

Why it matters

It is the reason to reach for Elasticsearch over a SQL LIKE. Tokenization, stemming, stopwords, and BM25 scoring turn “noise cancelling headphones” into a ranked list where the best documents float to the top — handling word order, partial matches, and typos that exact comparison cannot.

How it works

The query string passes through the field analyzer, producing terms scored against the inverted-index.

Query	Spans fields?	Best for
match	one	single-field search bars
multi_match	many	search across title+body+tags
query_string	many	power-user syntax (AND, "...", field:)
simple_query_string	many	safe user input (never throws on bad syntax)

• multi_match types — best_fields (default, takes the single best field score), most_fields (sums all), cross_fields (treats fields as one big field, good for names).
• Recall knobs — operator, minimum_should_match, and fuzziness trade precision for coverage.
• Per-field boost — "fields": ["title^3", "body"] weights title matches 3×.
• Scoring — each term contributes BM25; the analyzer must match index and query side or hits vanish.

Example

{ "multi_match": {
    "query": "wireless headphones",
    "type": "best_fields",
    "fields": ["title^3", "description"],
    "operator": "and",
    "fuzziness": "AUTO" } }

A doc with both words in title outscores one matching only description; fuzziness:AUTO still catches “headphnes”.

Pitfalls

• query_string on raw user input — a stray ] or AND throws a 400; expose simple_query_string to end users instead.
• cross_fields requires shared analyzers — mixed analyzers per field silently fall back to per-field scoring.
• Over-boosting — extreme boosts (title^100) make every result title-driven, hiding body relevance.
• Stopwords drop signal — searching “to be or not to be” against an english analyzer leaves almost no terms.

See also

• match-match-phrase-queries
• relevance-scoring-bm25