Advanced Search with CirrusSearch Extension in MediaWiki

CirrusSearch is MediaWiki's Elasticsearch/OpenSearch-backed search engine — the same one Wikipedia runs. The query language is what separates 'search that works' from 'search that finds things'. This is the hands-on syntax reference (installation and operation are covered in the companion guide on this blog). All keywords are lowercase and case-sensitive.

Basics: words, phrases, wildcards

• Plain words: parser extension — matches all terms, ranked by relevance
• Quoted phrases: "exact phrase" — disables stemming, matches exactly
• Wildcards: wik*edia (any characters), wik?pedia (one character); a wildcard cannot match the first character of a word
• Fuzzy: mediawiki~1 (word-level), "exact phrase"~2 (phrase proximity)
• Negation: -spam excludes pages containing the term

Filters

intitle:Foo            page title contains Foo
incategory:Books       pages in the Books category (or subcategories via deepcat:)
linksto:Foo            pages linking to Foo
hastemplate:Infobox    pages transcluding the template
inlanguage:de          page language filter
prefix:User            pages whose name starts with User
subpageof:Foo          subpages of Foo
contentmodel:wikitext  restrict by content model

Filters chain naturally, and every one restricts the candidate set — which is also the performance rule: add a filter before adding complexity.

Searching the source: insource

insource:"{{Infobox"          finds template calls
insource:/regex pattern/     regular expression match

insource: searches raw wikitext — the tool for 'which pages still use the old template' and other source-level audits. Plain regex needs the optional search-extra plugin; without it, quoted substrings still work.

Dates

creationdate:2025-09-01    pages created that day
lasteditdate:>today-30d    edited in the last month
creationdate:>=2024 creationdate:<2025   created during 2024

Date filters combine two filters for ranges; now-1h, today and now-1y style expressions are supported.

Files

filetype:png          file type filter
filemime:image/png    MIME filter
filesize:>20          size in kilobytes (20 = 20 KB)

More like this

morelike:Foo|Bar finds pages sharing significant terms with the given pages — research discovery and orphaned-content hunting. Available in the API too (action=query&list=search&srsearch=morelike:...).

Ranking and sorting

• prefer-recent:1y — decay older pages' scores
• boost-templates:"Template:Important|200%" — favor pages transcluding certain templates
• Explicit order via the results URL: &sort=last_edit_desc (also create_timestamp_desc, incoming_links_desc, random, title_natural_asc; API callers use srsort); sorting disables score-based keywords

Example queries

// All books by a specific author, newest first
incategory:Books "specific author"

// Templates still using the old infobox
hastemplate:Infobox old infobox

// Recently edited user pages about the API
prefix:User API lasteditdate:>today-90d

// Files larger than 5 MB named 'map'
filetype:png filesize:>5000 intitle:map

Pitfalls

• Keywords are case-sensitive — INCATEGORY: does not work
• Underscores vs. spaces: prefix:User Foo_Bar treats _ as part of the word — use quotes for multi-word values (incategory:"Foo Bar")
• The search suggestions (autocomplete) index updates roughly daily, while full-text results update within minutes — don't diagnose bugs across the two

The full reference is Help:CirrusSearch on mediawiki.org — including logical operators, proximity details and language-specific behavior. The 15 minutes learning the filters repay themselves in every search you stop clicking through.