How would you test a search box / search functionality?

Clarify first

• Is search full-text or against a structured index (Elasticsearch, DB LIKE query)?
• Are results ranked by relevance, recency, or something else?
• Is there autocomplete or query suggestion?
• Does search span multiple entity types (products, users, articles) or a single type?

Functional

• Query returns relevant results; results match the search term
• Autocomplete suggestions appear after the configured keystroke threshold and are correct
• Clearing the search field returns all results or resets to the default state
• Pagination of results works correctly — page 2 has different results than page 1
• Filters or facets (category, date range) correctly narrow results when combined with search

Negative / error handling

• Empty query or whitespace-only → no results or all results (whichever is specified), no error
• Query with only special characters (/, *, ?, ", etc.) → handled gracefully, no server error
• SQL/script injection in search field (' OR 1=1 --, <script>alert(1)</script>) → treated as literal text, not executed
• Very long query string (e.g. 10,000 characters) → truncated or rejected with a clear message

Edge & boundary

• Single-character search (is it supported or is there a minimum length?)
• Exact-match vs partial-match behavior: searching "log" — does "login" appear?
• Case sensitivity: "Login" vs "login" return same results?
• Search while previous results are still loading (race condition — correct results shown)
• Unicode and emoji in the search query

Performance

• Response time for common queries under normal load
• Response time for wildcard or broad queries that return thousands of results
• Behavior under concurrent search requests

Compatibility

• Search works consistently across browsers; mobile keyboard "Search" key submits the query

Close: automate functional result assertions, known injection payloads, and boundary inputs (empty, max length). Keep manual for relevance quality — whether the right results appear at rank 1 is hard to assert programmatically and needs human judgment.