Search

The search parameter finds results matching a given text search. Search requests cost $1 per 1,000 calls (vs. $0.10 per 1,000 for list+filter requests); rerank adds $1 per 1,000. See pricing.

# Works with "dna" in title, abstract, or fulltext, or tagged with the DNA keyword
https://api.openalex.org/works?search=dna

What gets searched

Each entity type searches different fields:

Keywords in search

A works search also finds papers by their keywords. When a phrase of one to six words in your search is the name of a keyword, or one of its synonyms, a work matches if the phrase is in its text or the work carries that keyword. Every other word must still be in the text. So remote work productivity finds papers tagged remote-work that mention productivity, even if they say “telework” instead, but never a paper tagged only productivity. A search never returns fewer works than the same words in the text alone.

# Title, abstract, or keywords: what openalex.org searches by default
https://api.openalex.org/works?search.title_abstract_keywords="antimicrobial resistance"

# Title and abstract text only
https://api.openalex.org/works?search.title_and_abstract="antimicrobial resistance"

The first search finds 241,167 works, the second 118,920: the keywords add papers that say “antibiotic resistance” or “drug-resistant bacteria”, papers in other languages, and papers with no abstract. Boolean and phrase searches work the same way, one term or quoted phrase at a time. The filter form is title_abstract_keywords.search, and in OQL it’s title-abstract-keywords has (…). The broad search (search=, or the fulltext.search filter) also matches keywords.

Text processing

OpenAlex uses stemming and removes stop words to improve results:

• Words like “the” and “an” are removed
• A search for “possums” also returns “possum”
• Searches match whole words only (“lun” won’t match “lunar”)

Exact (unstemmed) search

Use search.exact to search without stemming:

https://api.openalex.org/works?search.exact=surgery

Note: Works also take scoped search parameters: search.title, search.title_and_abstract and search.title_abstract_keywords, each with an .exact form. Each one can appear once per request, and several are combined with AND. search.semantic can’t be combined with any other search.

Boolean search

Use AND, OR, NOT (uppercase) for complex queries. Surround phrases with double quotes for exact matching:

# Works about "elmo" AND "sesame street" but NOT "cookie" or "monster"
https://api.openalex.org/works?search=(elmo AND "sesame street") NOT (cookie OR monster)

Note: Words not separated by boolean operators are treated as AND.

Large Boolean queries

Because the query travels in the request URL, the whole request URL is limited to about 4 KB (roughly 4,000 characters). A very long search= value — typically a Boolean query with many OR terms, common in systematic reviews — can exceed this and return a 400 error:

{
  "error": "Request URL too long",
  "message": "Your request URL is 4612 bytes, over the 4094-byte limit (roughly 4 KB, mostly the 'search' value). Split a large Boolean query into smaller chunks, request each separately, and combine the returned IDs client-side."
}

This is a fixed limit, not a usage/credit cost — splitting the query does not lose any results. To run a query larger than ~4 KB, split the OR list into chunks, request each chunk separately, and take the union of the returned IDs client-side. The result is exactly the same set of works:

# Instead of one query that is too long:
search=(termA OR termB OR ... OR termZ) AND climate

# Split the OR list and union the IDs from each request:
search=(termA OR ... OR termM) AND climate
search=(termN OR ... OR termZ) AND climate

Keep each chunk’s full request URL under ~4 KB. Combining results is exact because (X AND (a OR b OR c OR d)) equals (X AND (a OR b)) ∪ (X AND (c OR d)). Sending your api_key in the Authorization header instead of the URL frees a little extra room, but for a large Boolean query you will still need to split it.

Tip: Each chunked request is billed independently, just like any other request. Splitting is only needed to stay under the URL length limit — it is not a way to reduce cost.

Phrase and proximity search

Phrase matching

Use double quotes to search for an exact phrase. Multi-word searches without quotes rank results higher when words appear close together.

# Exact phrase (few results)
https://api.openalex.org/works?search="fierce creatures"

# Words anywhere (more results)
https://api.openalex.org/works?search=fierce creatures

Proximity search

Append ~N to a quoted phrase to find the words within N positions of each other, without requiring an exact match:

# "climate" and "change" within 5 words of each other
https://api.openalex.org/works?search="climate change"~5

This works for phrases of any length, not just two words. N is the total number of position moves allowed to line up all the terms — a single budget shared across the whole phrase, letting the words be reordered or spread apart as long as the total displacement stays within N. Widening N returns more matches:

# the three words near each other, in any order
https://api.openalex.org/works?search="climate change policy"~10

To require two separate phrases near each other — rather than loosening one phrase — join two quoted operands with ~N. Each operand keeps its own words adjacent, and the two operands must appear within N words of each other, in either order:

# "machine learning" and "neural network" within 5 words of each other,
# each phrase kept intact
https://api.openalex.org/works?search="machine learning"~5~"neural network"

Wildcards and fuzzy search

Wildcards

Use * to match zero or more characters and ? to match exactly one character:

• Trailing wildcard: machin* matches “machine”, “machines”, “machinery”
• Single-character wildcard: wom?n matches “woman” and “women”

The search term must have at least 3 characters before the wildcard. Leading wildcards (e.g., *ology) are not supported.

Hyphenated and slashed terms are split into separate words by the search engine, so the 3-character rule applies to the last part. e-cigarette* matches “e-cigarette” and “e-cigarettes” (it runs as the adjacent words e + cigarette*), and x-ray* and t-cell* work the same way. covid-19* is rejected, because only 19 precedes the wildcard.

Wildcards require the exact (unstemmed) search — use search.exact (or the fulltext.search.exact filter), not the default search. The default search is stemmed, which strips the literal text before the wildcard at index time, so a wildcard there would return wrong results; OpenAlex rejects it with a 400 rather than return misleading matches.

https://api.openalex.org/works?search.exact=machin*

Quotes around a single word make no difference: search.exact="machin*" behaves exactly like search.exact=machin* (a one-word “phrase” is just the word). Wildcards also work inside a multi-word quoted phrase, where they match as an adjacent phrase — see Phrase matching (e.g. search.exact="smart* phone").

Tip: To combine stemmed and truncated terms in one query — common in systematic-review search strategies — use the filter form so each term keeps its own matching: a stemmed fulltext.search filter alongside a wildcard fulltext.search.exact filter, e.g. ?filter=fulltext.search:treatment,fulltext.search.exact:psoriat*.

Fuzzy search

Append ~N to a term to allow up to N character edits (insertions, deletions, or substitutions). The edit distance N can be 0, 1, or 2:

# Matches "machine", "machin", and other close variants
https://api.openalex.org/works?search=machin~1

The search term must have at least 3 characters before the ~. This is useful for catching typos and spelling variations.

Relevance score

Search results include a relevance_score property and are sorted by it (descending) by default. The score is based on:

• Text similarity to your search term
• Citation count, capped: citations raise the score by at most half (a paper with 100 citations gets 25% more, one with 5,000 about 50% more), so a famous paper that barely matches can’t outrank a close match
• A small bonus for works carrying a keyword your search names

Rerank

Add rerank=true to a works search and the top 100 results are reordered by how well each one answers your search, judged by a decision model from each work’s title, venue, year and type. Result 101 onward is exactly what you get without rerank.

https://api.openalex.org/works?search.title_abstract_keywords=remote work productivity&rerank=true

• Reranked results carry rerank_score, the model’s probability that the work is a relevant result. relevance_score stays the text score, so inside the top 100 it is no longer in order.
• meta.reranked says whether the page’s order came from the reranker. It is false on pages past result 100, and when the reranker didn’t answer in time; you then get the normal order.
• Pages and cursors work as usual. With cursor=*, the first 100 results come in reranked order, then the walk continues in the normal order from result 101. It returns the same works as a walk without rerank, with no repeats and no gaps.
• The reranked order is computed once and kept for 24 hours, so paging through it is stable.
• Rerank adds 10 credits to the request: a search costs 10 credits, a reranked one 20 ($2 per 1,000 calls). The same add-on applies on the OQL door. openalex.org and the Claude and ChatGPT connectors rerank by default. See Example costs.
• rerank=true needs a search sorted by relevance: with another sort, group_by, sample or search.semantic it returns a 400.

Semantic search

If you want to match by meaning rather than keywords — or if you’re searching with a long input like an abstract or grant description — semantic search is a better fit. It uses AI embeddings to find conceptually related works even when the wording differs.

https://api.openalex.org/works?search.semantic=machine learning in healthcare

See the Semantic Search guide for examples, supported filters, and limits.

Legacy: filter-based search

Warning: Deprecated. The filter=field.search: syntax still works but is no longer recommended. Use the search query parameter instead.

The .search filter suffix searches a specific field rather than all searchable fields at once:

# Authors with "Einstein" in their name (deprecated)
https://api.openalex.org/authors?filter=display_name.search:einstein

# Works with "cubist" in the title only (deprecated)
https://api.openalex.org/works?filter=title.search:cubist

The fulltext.search filter searches a work’s title, abstract, and full text together — the same fields as the search query parameter. (The older default.search filter is a deprecated alias of fulltext.search: it still works and behaves identically, but fulltext.search is preferred.) Variants like .search.exact (unstemmed) and .search.no_stem also exist for some fields.

These filter-based searches cost the same $1 per 1,000 requests. For autocomplete use cases, use the autocomplete endpoint instead.

Searching author bylines with raw_author_name.search

raw_author_name.search finds works where someone appears by a specific name-as-published — exactly the string in the byline of the paper. It’s the one filter-based search that does not have a search parameter equivalent, so it stays the right tool for matching a person to their works.

Warning: It matches across all author names on the work, not within a single byline. Each unquoted token can match a different author. A search for raw_author_name.search:john smith returns a work whose authors are John Doe and Alice Smith — neither person is named “John Smith,” but the tokens appear somewhere in the byline list.

To scope a search to a single person (one byline), wrap the name in double quotes — the query becomes a phrase match against one raw_author_name at a time:

# Work-scoped: "john" and "smith" appear in *any* bylines (93,304 hits)
https://api.openalex.org/works?filter=raw_author_name.search:john smith

# Byline-scoped: works with an author named "John Smith" (2,727 hits)
https://api.openalex.org/works?filter=raw_author_name.search:"john smith"

Allow middle names and initials with proximity (~N)

Append ~N to a quoted name to allow up to N intervening tokens. This catches middle names and middle initials without dropping the byline scope:

# "Jane Smith" only — 1,661 hits
https://api.openalex.org/works?filter=raw_author_name.search:"jane smith"

# Also matches "Jane M Smith" — 3,779 hits
https://api.openalex.org/works?filter=raw_author_name.search:"jane smith"~1

# Also matches "Jane Marie Smith" — 4,571 hits
https://api.openalex.org/works?filter=raw_author_name.search:"jane smith"~2

Wider slop also picks up co-author noise on common names — start narrow and widen only when you need more recall.

Handle name variants with OR

The filter value supports Lucene OR between quoted phrases. Use one OR’d query (not multiple raw_author_name.search: clauses — those AND together and return nothing) to combine name forms:

# Comma-reversed byline ("Priem, Jason" → "Priem Jason")
https://api.openalex.org/works?filter=raw_author_name.search:"jason priem" OR "priem jason"

# First-initial form
https://api.openalex.org/works?filter=raw_author_name.search:"j priem" OR "priem j"

For a step-by-step walkthrough that uses these patterns to audit and correct an author profile’s works, see the Audit an Author Profile’s Works recipe.