Attributes

This is the canonical dictionary of every attribute on a work object. Attributes shared with other entities (id, ids, display_name, created_date, updated_date) are documented once on Common attributes; works-specific notes on them are below.

Attributes

id

String. The OpenAlex ID for this work, e.g. https://openalex.org/W2741809807. See Common attributes.

ids

Object. All known external identifiers for this work, as URIs where possible; keys for unknown IDs are omitted. Work-specific keys: openalex, doi, mag (Microsoft Academic Graph integer ID), pmid (PubMed), pmcid (PubMed Central).

doi

String. The work’s DOI — the canonical external ID for works. A work can have more than one DOI (e.g. a preprint DOI and a published DOI); this field holds one, picked by source: a Crossref DOI first (the published version’s over a preprint’s), then DataCite (a concept DOI over its version DOIs), then a DOI printed on a PubMed, MAG, or repository record. Every DOI-registered copy also shows its own DOI as the doi of its location. Why a work can have several: DOIs.

Filter: doi: matches a work whose own doi or any location’s doi equals the value, so a secondary DOI (a preprint’s, a Zenodo version’s) finds the work it belongs to. Short (10.7717/peerj.4375) and URL (https://doi.org/10.7717/peerj.4375) forms both work, case-insensitively, and up to 100 can be OR-ed with |. doi:!… excludes works that have the DOI in either place. has_doi checks only the work’s own doi.

display_name

String or null. The work’s title. Identical to title; every entity carries a display_name. Null when the work has no title; see title.

title

String or null. The title of the work. Identical to display_name. Null when no source gives the work a title (an empty or whitespace-only title from a source counts as none): about 2.3M of the ~465M works.

alternate_titles

List of strings. The work’s other titles, as its publisher deposited them: most often the same title in a second language, such as the English title of an Arabic or Japanese article, or the reverse. Empty when there are none (most works). They come from the work’s Crossref record (the extra elements of its title list and its original-title), minus the title the work shows as title, repeats, and labels such as “English”; at most 5, in the order deposited. About 4.1M works have at least one.

Search: search=, title.search and title_and_abstract.search (and their .exact forms) look in them too, so a work is found by any of its titles. For example, W3164863413 has an Arabic title and its English title in alternate_titles, and /works?filter=title.search:interactive whiteboard female teachers finds it.

type

String. The work’s type (e.g. article, preprint, dataset, paratext). Every work has exactly one. See Work types for the full vocabulary and how the type is assigned.

is_traditional

Boolean. True if the work’s type is a traditional one, the kinds of works the scholarly record has long counted (articles, books, chapters, theses, conference papers and the like); false for every other type, such as datasets, software and reports. It comes from type, so every work has a value, and it changes only when the type does. Both lists are on Work types.

Filter: is_traditional:true keeps works of the traditional types, and is_traditional:false keeps all the others. Group with group_by=is_traditional. You can’t sort by it.

language

String. The language of the work’s metadata (title and abstract), as a lowercase code: the two-letter ISO 639-1 code where the language has one (en, zh), else its three-letter ISO 639-3 code (ceb for Cebuano). We detect it with the fastText language identifier (lid.176), reading the title and the start of the abstract; when that gives no answer, the work takes the language its source record declares. For a PubMed article whose indexers record a language other than English, that language comes first. It reflects the metadata language, not necessarily the full text, and is unset when neither gives a language. Support can correct a work’s language through a curation. See Languages.

publication_date

String. The day the work was published, as an ISO 8601 date. Where several dates exist, OpenAlex usually picks the earliest electronic-publication date. For a work with a Crossref record, that is the earliest publication date the publisher states in it (issued, published online, published in print, posted or approved), so an article published online in one year and in a print issue the next takes the online date. The day the DOI was registered counts only when the record states no online or posted date: it then stands in for when the work went online, as for publishers that deposit only a print date. A date that is wrong in the publisher’s record is fixed there, and OpenAlex follows. Applies to the primary_location version; other locations may be earlier.

publication_year

Integer. The year the work was published.

biblio

Object. Old-style bibliographic pointers, mostly useful in citation contexts. All strings (values like “Spring” or “Inside cover” turn up): volume, issue, first_page, last_page. When a publisher numbers its articles instead of printing page numbers, first_page holds the article number (e70493, 094517, giab008) and last_page is empty. biblio.volume is the volume’s number as printed in a citation; filtering, sorting or grouping by it is no longer supported. The book or proceedings volume itself is one of the work’s parent_works.

parent_works

List. The works this work is part of: a chapter’s book, a figure’s paper, a file’s dataset. Each entry is a dehydrated work (id, display_name) plus link_type, the kind of link, one of the four part-of link types: volume, component, data-series or part. For example, a chapter has [{"id": "https://openalex.org/W3109397498", "display_name": "Routledge Handbook of Critical Indigenous Studies", "link_type": "volume"}]. Until Monday 7 December 2026 each entry also carries the same value as relationship, the field’s old name (Deprecations). Empty when the work isn’t part of another work, or no link was found. Only part-of links are here: a preprint and its article are one work, and a paper’s cited works are its referenced_works.

The links come from the records themselves. A chapter, section or reference entry gets its book when the chapter’s Crossref record and the book’s agree on an ISBN, or the chapter’s DOI extends the book’s. A DataCite record gets a parent when it declares IsPartOf another DOI that is also a work. A work can have several parents, up to 10. Links that can’t be a hierarchy are left out: a record that claims more than 10 parents, two records that each claim to be part of the other, and chains more than 10 levels deep.

Filter: parent_works.id:W3109397498 lists a book’s parts, has_parent_work:true keeps works that are part of another, and parent_works.link_type:volume keeps one kind of link. Group with group_by=parent_works.link_type or group_by=parent_works.id. See Parent and child works. parent_works replaced the volume field on 14 November 2026 (Deprecations).

lineage

List. OpenAlex IDs of this work, then its ancestors, nearest first: its parents, then their parents, and so on. A work’s own ID is always first, so a work with no parent has a lineage of just itself. It works like lineage on institutions: the filter lineage:W3109397498 returns that work and everything under it, at any depth.

child_works_count

Integer. How many works name this work as one of their parent_works: a book’s chapters, a paper’s figures. Direct children only. List them with parent_works.id:<this work's ID>. Filter and sort by it, e.g. child_works_count:>100.

abstract_inverted_index

Object. The abstract as an inverted index — each word mapped to its positions. OpenAlex does not ship plaintext abstracts for legal reasons; reconstruct the abstract from the index. When a preprint and its published version are one work, this is the published version’s abstract, from its Crossref or PubMed record, unless that record holds only a fragment. Newer works are more likely to have one (over 60% of 2022 works vs. ~45% of pre-2000 works). Abstracts come from diverse sources, sometimes text-mined from the fulltext, so one can occasionally carry trailing text that isn’t part of the abstract (section headings, keywords); file a ticket if you spot this on a work.

{ "abstract_inverted_index": { "Despite": [0], "growing": [1], "interest": [2], "in": [3, 57, 73] } }

authorships

List. Authorship objects, each pairing an author with their institution(s) and role on the work. In a list of works, each work shows its first 100; the single-work endpoint (/works/{id}) and the snapshot show all of them, up to 5,000. See How many authors a work shows. Filter by the region or city of the authors’ institutions with authorships.institutions.geo.region and authorships.institutions.geo.city (looked up at query time).

is_authors_truncated

Boolean. In a list of works, true when authorships holds only the work’s first 100 authors, false when it holds them all. For the full list, fetch the single work (/works/{id}), which leaves this key out. Filter with authors_count:>100.

corresponding_author_ids

List. OpenAlex IDs of authors whose authorship has is_corresponding: true.

corresponding_institution_ids

List. OpenAlex IDs of institutions attached to a corresponding author’s authorship. When no author is marked corresponding, but every author has an affiliation string matched to an institution and they all share one institution, it holds that one institution, and corresponding_author_ids stays empty (we know where, not who). About 1 work in 10 gets its value this way.

countries_distinct_count

Integer. Number of distinct author country codes across the work’s authorships.

institutions_distinct_count

Integer. Number of distinct institutions across the work’s authorships.

institutions

List. Empty on every work for now ([]). It is reserved for institutions linked to the work as a whole rather than through an author (such as a dataset’s hosting organization), and nothing fills it yet. The institutions of a work’s authors are in authorships[].institutions; institutions_distinct_count counts them, and the filter institutions.id (the same as authorships.institutions.id) finds works by them.

primary_location

Object. The location holding the best (closest to the version of record) copy — for a journal article, the published full text at the publisher’s DOI URL. See Locations for the object shape. Its dehydrated source carries the source’s list memberships, so you can narrow works to journals on an external list: primary_location.source.listed_in:doyens (see listed_in for the list ids; the same key works under locations. and best_oa_location.). Filter/group_by. You can also filter by attributes of the primary source and its publisher that the work doesn’t store, looked up at query time: primary_location.source.country_code, .is_global_south, .is_ojs, .is_high_oa_rate, .is_preprint_repository, .summary_stats.h_index, .summary_stats.2yr_mean_citedness, and primary_location.source.host_organization.country_code (see Filter by a related entity’s attributes).

locations

List. Every unique place this work lives, each a location object. In a list of works (filter, search and sample results), each work shows its first 100 locations; the single-work endpoint (/works/{id}) and the snapshot show all of them. Filters and group_bys on locations.*, repository and doi look through every location, so a work can match on one its list entry doesn’t show. See Works with many locations.

locations_count

Integer. The number of locations: all of them, including in a list of works that shows only the first 100.

best_oa_location

Object. The best openly available location, or null if none. “Best” is scored by: must be OA; publisher beats repository; publishedVersion beats acceptedVersion beats submittedVersion; a direct PDF link beats none; and major repositories (PubMed Central, arXiv) rank above others.

open_access

Object. The work’s access status, an OpenAccess object:

  • is_oa (Boolean) — true if a free-to-read full text exists somewhere (OpenAlex uses a broad definition: readable without paying or logging in).
  • oa_status (String) — one of diamond, gold, green, hybrid, bronze, or closed. The bronze-vs-gold distinction depends on the journal’s openness — see Sources.
  • oa_url (String) — the best OA URL (closest to the version of record); may be a PDF or a landing page.
  • any_repository_has_fulltext (Boolean) — true if any location is both OA and hosted by a repository. Surfaces “shadowed green” OA that oa_status hides once a publisher-hosted copy exists.

See Open access for how these fields combine.

apc_list

Object. The work’s list-price article processing charge as advertised by the journal (value, currency, value_usd). The listed price, not necessarily what was paid; apc_list.value of zero indicates a diamond-OA journal.

apc_paid

Object. OpenAlex’s best estimate of the APC actually paid (value, currency, value_usd): the article’s own charge from OpenAPC when OpenAPC has it (then always in EUR), otherwise the list price for gold and hybrid works. No field says which of the two a value came from, so you can’t filter to OpenAPC values only.

primary_topic

Object. The top-ranked topic for the work, with its subfield, field, and domain. Same as the first entry in topics. null when the work has no topic (the classifier found it not classifiable, or it has no title or abstract). See how topics are assigned.

topics

List. The work’s three most likely topics, best first, each with a score and its subfield/field/domain; empty when the work has no topic. score is the classifier’s calibrated probability that the topic is the work’s main one, so the 2nd and 3rd scores are often close to 0.

keywords

List. Short phrases saying what the work is about (keywords), in the order the tagging model named them (not sorted by score), each with a score: the tagging model’s confidence, from 0 to 1. Empty when the record says nothing about its subject.

concepts

List. Legacy concept tags with a score. Concepts are a superseded classification retained for continuity; topics are the current primary classification. Ancestors of assigned concepts are also included, so you may see low or zero scores.

sustainable_development_goals

List. The UN’s 17 Sustainable Development Goals the work addresses, as objects (id, display_name, score), tagged by a classifier OpenAlex trained on language-model judgments of titles and abstracts. Every goal with a score of 0.4 or above is shown. score is the classifier’s confidence from 0 to 1, not a probability: tags scored about 0.4 to 0.7 were right roughly half the time when checked, tags above 0.9 about 8 times in 10. Until October 2026 this field came from the Aurora classifier; see SDGs for the switch.

study_designs

List. How the research inside the work was done, as study design objects (id, display_name): Randomized Controlled Trial, Clinical Trial, Observational Study, Case Report, Systematic Review, Meta-Analysis or Study Protocol. A work can have several, and parents are included: every randomized controlled trial is also a clinical trial, and every meta-analysis is also a systematic review. Every value comes from automated tagging of the title and abstract, held to at least 99% precision for randomized controlled trials and at least 95% for every other value; PubMed’s own tags are not copied. Only works with an abstract and a research-carrying type are tagged, so an empty list means “not tagged, or none of these seven”, never “not a study”. Publication formats (editorial, letter, review, guideline) are in type, not here. Filter with study_designs.id:randomized-controlled-trial; group by study_designs.id.

mesh

List. MeSH tag objects. Present only for works sourced from PubMed; an empty list otherwise.

cited_by_count

Integer. The number of works that cite this work — the count of successful reference matches pointing at it (see Citations and references).

reviewed_by_count

Integer. The number of works that review this work: peer reviews, preprint reviews, expert recommendations and book reviews linked to it by a review link. 0 when nothing reviews it. List the reviews with /links?filter=target:<this work's ID>,link_type:review. Filter and sort by it, e.g. reviewed_by_count:>0 for works that have at least one review. It counts linked reviews; it is not a mark of quality, and publishers differ in how many records they mint per review. It changes the work’s updated_date when it changes, as cited_by_count does.

cited_by_api_url

String. The API query for the works that cite this work: https://api.openalex.org/works?filter=cites: followed by the work’s ID. Following it lists the citing works in the API’s default corpus; add corpus=all to include the expansion corpus too. Available only through the API, not in the snapshot, where you can build it from id.

counts_by_year

List. cited_by_count for each of the last ten years, binned by year (years with zero citations omitted; citations older than ten years excluded). See Common attributes, including why it can drift from a live count.

cited_by_percentile_year

Object. The percentile rank of this work’s citation count against other works from the same year, as a min/max range.

fwci

Float. The Field-Weighted Citation Impact: citations received divided by citations expected for works of the same type, year, and subfield. 1.0 is world average.

How fresh. Recomputed every night for every work, along with citation_normalized_percentile and cited_by_percentile_year. The snapshot carries each night’s exact values. The API (lists, filters, sorts, group-bys and single-work requests) carries them to within 1% of the night’s fwci, 0.01 of the percentile value and 1 point of cited_by_percentile_year: a work that moves further is updated the morning after the night. After a night that shifts many works at once, the search results can take a day or two to catch up.

citation_normalized_percentile

Object. The same information as fwci expressed as a percentile (value, is_in_top_1_percent, is_in_top_10_percent). value is the share of works of the same type, publication year and subfield with fewer citations in the FWCI window (the publication year plus the three following years). Works with the same count share one value, so a work with no citations in that window reads 0 and is never in the top 10% or top 1% (from 10 November 2026; before, tied works were spread out by work ID, and about 1.7M uncited works read top 10%). Recomputed every night; a change in value alone does not move updated_date, a change in either flag does (from 30 October 2026).

referenced_works

List. OpenAlex IDs of the works this work cites (this work ➞ others). See Citations and references.

referenced_works_count

Integer. The length of referenced_works.

List. OpenAlex IDs of up to 10 works closest in meaning to this one: the nearest neighbours of its title, abstract and venue among the embeddings of every titled work. Paratext, errata, retractions, reviews and retracted works are never listed, nor works with this work’s own title. A list is shorter, or empty, when too few works qualify (common for generic titles such as “Preface”). Recomputed for every work in October 2026; until then this was a list of recent papers sharing the most topics with this one. That recomputation did not move updated_date: if you keep a copy current by updated_date, reload related_works once from a full snapshot made after 28 October 2026, or from the API. New works get their list the day after they are added, and their updated_date moves when it lands.

funders

List. Dehydrated funder objects for this work. Replaces the removed grants property. Filter on funders.id, or by the funder’s country with funders.country_code (looked up at query time).

awards

List. Award/grant objects (id, display_name, funder_award_id, funder_id, funder_display_name) linking the work to specific grants. An award’s DOI is on the award itself. Replaces the removed grants property.

has_content

Object. Whether downloadable full text exists in each format: pdf (Boolean) and grobid_xml (Boolean, TEI XML). More than 60 million works have has_content.pdf: true. See Fulltext.

has_fulltext

Boolean. True if OpenAlex holds the work’s full text for searching, so fulltext.search can match it. This is not the same as a downloadable file: about 4.3M works have searchable text but no file in has_content, and about 3.5M have a PDF whose text isn’t searchable yet. To find works you can download, filter has_content.pdf:true.

content_urls

Object. URLs for downloading full-text content, present when has_content indicates the format is available: pdf and grobid_xml, each a URL under content.openalex.org (or null). Fetching requires your API key (a free one is fine); each download is metered against your usage, and the free daily allowance covers a batch of them (see Example costs). Available only through the API, not in the snapshot.

is_retracted

Boolean. True if the work is known to be retracted: the Retraction Watch database lists its DOI as a retracted paper (and not as reinstated), or Crossref or PubMed records a retraction of it, or its title starts “Retracted”. We read Retraction Watch every day, as Crossref publishes it, so a retraction added there usually shows here the next day. A retracted paper with no DOI is flagged only through Crossref, PubMed or its title. Where these sources are wrong about one work, our curators correct it with a curation, which wins over them.

is_pulled

Boolean. True if the work’s publisher, host or authors retracted, withdrew or removed it: every retracted work (except a retract-and-replace and a Cochrane review withdrawn as out of date), plus works their publisher’s Crossref record withdraws or removes and preprints or repository records their host stamps withdrawn or removed. Pulled works are left out of works lists, searches and counts unless you add include_pulled=true; a filter on is_pulled or is_retracted brings them back too. Looking a work up by ID always shows it. See Pulled works.

is_paratext (deprecated)

Boolean. True if the work is paratext (covers, tables of contents, mastheads, and similar). Now derived directly from type — true exactly when type is paratext. Filter with type:paratext instead.

is_xpac

Boolean. True if the work belongs to the expansion corpus — the ~190M works added in the Walden update (primarily datasets and single-repository records), formerly called “XPAC” (Expansion Pack). Data quality is lower but improving, and the set is static: no works have been added to it since the cutover. Expansion works are excluded from API results by default; include them with corpus=all. See Expansion corpus for core/expansion/all and the deprecated legacy controls.

indexed_in

List. The indexes this work is indexed in. Possible values: arxiv, crossref, datacite, doaj, pubmed. doaj is the article-level DOAJ attribute: it means DOAJ’s article index holds a record of this work. DOAJ also has a journal list, and OpenAlex has a separate journal-level attribute for it: is_in_doaj on the source (primary_location.source.is_in_doaj as a works filter). A work in a DOAJ journal lacks indexed_in:doaj when DOAJ’s article index holds no record of that article; see DOAJ: journal list and article index.

created_date

String. The date the work was added to OpenAlex (YYYY-MM-DDTHH:MM:SS; only the day is recorded, so the time is always 00:00:00). It never moves later: a work whose record is rebuilt keeps the earliest date OpenAlex served it with. Many older works show 2016-06-24, the Microsoft Academic Graph load date. Restoring these dates in October 2026 did not move updated_date for 213.7 million works: if you keep a copy current by updated_date, reload created_date once from a full snapshot made after 28 October 2026. See Common attributes.

updated_date

String. The ISO 8601 UTC timestamp of the last change to the work object, including citation-count increases. It does not move when only fwci, the percentile value, cited_by_percentile_year (recomputed every night) or has_content change. See Common attributes.

Deprecated fields

  • host_venue and alternate_host_venues — removed; use primary_location and locations. Filtering or grouping on them errors.
  • grants — removed; use funders and awards.
  • concepts — superseded by topics (still present; see above).

Last updated

View as Markdown