WikiCite/ReCite

ReCite (Reusable/Reliable Citation system) proposes a curated reference database and a stateful fork of Citoid that reads from it — deliberately scoped to reusable reliability, as a foundation for larger efforts.

Infographic describing the ReCite system, Proof of Concept, August 2026 v3

Status: Proof of Concept (PoC) Design v3, August 2026 version -- actively being (re)designed and drafted

Wikipedia has no citation entity. A reference is text inside an article, re-typed in every article that uses it, unaware of what the community has already decided about its source.

Background

Discussions have long been held in Wikipedia communities about the value of having a centralized database or system regarding citations; the WikiCite project and proposals related to Shared Citations are examples of this.

However, these are very big proposals at a time when financial resources are shrinking.

Yet, citation centralization may have the ability to maintain or even increase Wikipedia’s value due to its relationship to supporting reliable information across Wikimedia’s information. But how exactly will centralized citations across Wikimedia become meaningful? Specifically:

• How might centralized citations be incorporated into editor workflows and Wikimedia content?
• How could centralized citations demonstrate their value within and beyond Wikimedia?

To help answer these questions, we’re proposing a smaller-scale Proof of Concept (PoC) for a Reliable/Reusable Citations system ("ReCite"), which includes a new database along with a fork of the Citoid software ("ReCitoid"). Focusing on the issue of reliability alone is helpful because even this more limited scope presents sufficient complexity. The ReCite resource, if successful, could then serve as the foundation for larger efforts such as Shared Citations as well as WikiCite.

At this time, there are several paths that a ReCite Proof of Concept could take. The right path will depend on problem priority and partner collaborations. This document outlines several possibilities for future decision making.

Problem scope: Two failures, One root cause

Citations are stored as strings, not as objects. Everything follows from that.

ReCite is focused on two key issues:

• The "Prime Citation" problem. One of the major problems related to reliable citation is the use of ancillary or instance-based links: citing a book itself, rather than citing the Amazon.com copy of the book. In this case, the citation might appear to look “unreliable” when looking at Amazon.com due to Amazon’s ratings in the English-language Reliable sources/Perennial sources list (WP:RSP).

• Reliability judgments have difficulty reaching the citation. ReCite would also aim to help propagate the judgments represented in places like (WP:RSP) in a structured format to be applied at the citation level. Examples of these judgments include or may include:
  • RSP (with flavors in 14 languages)
  • all sources that land on the RS noticeboard (with flavors in 15 languages)
  • smaller bespoke RS tables for individual projects ushc as https://meta.wikimedia.org/wiki/Cite_Unseen/sources w/ most/some of the above OR WikiSignals.org OR https://en.wikipedia.org/wiki/Wikipedia:Citation_Watchlist/Lists
  • tools with stored verification, such as https://en.wikipedia.org/wiki/User:Alaexis/AI_Source_Verification
  • all sources whose reliability is disputed (via edit summaries mentioning source reliability; via talk page sections mentioning it)
  • all sources subject to edit wars

In addition, the following is also a consideration of this project, in the background:

• Wikimedia Space/Data Savings. The number of times that a single citation appears across Wikimedia takes up a lot of data/space, as has been documented in the past. A citations database with a single ID per entity may provide the opportunity for some future data savings. This also needs to take into account storage issues related to the Wikidata graph split.

Core issue: absence of citation entity on Wikipedia

A core issue across Wikimedia is a centralized way of managing the citation or cited source itself, regardless of what Wikipedia article the citation appears in, what page number of the citation is referenced, or whether the LibraryThing versus Google Book version of the reference is used.

All of these aspects matter when it comes to the question of reliability: the act of citation itself is one that attempts to give legitimacy or validity to a work.

Citations are complicated. One way of looking at them is as a relationship (or an edge in a graph): a specific source of information cites another source of information.

However, when citations are employed, they acquire additional attributes that reflect scholarly standards, bibliographic structures, and location:

• Bibliographic. Several metadata standards exist in library science, arising out of different needs and contexts: consider the difference between citing an artwork or a book and a webpage. Examples of standards such as Dublin Core (defined in the mid-1990s) and MARC (defined in the 1960s) can be found here.
• Scholarly practice. Different citation styles exist for different professions. Examples include the ACM Style for certain computer science journals, or in the United States The Chicago Manual of Style for humanities and social science writing and Bluebook citation style for legal writing. These different styles not only help ensure that the appropriate information about the work is collected, but also to ensure scholarly rigor.
• Location within source. Such as page numbers or line numbers.

From a reliable citation management perspective, here’s what exists:

• Citation-related data: Wikidata currently has data related to citations but these are not systematically collected or organized from a citations-oriented perspective. Examples:
  • the article represented by https://www.wikidata.org/wiki/Q64706055 cites many other sources using property P2860 cites work
  • metadata related to this work, which may hint at usage on Wikimedia, includes https://en.wikipedia.org/wiki/Special:EntityUsage/Q20124?entity=Q20124 and https://www.wikidata.org/w/index.php?title=Q20124&action=info
• Citation Instances: Internet Archive has been building a citations database which is focused on every instance of a citation across Wikipedia articles (LINK)

Here's what's needed:

• A citation-oriented model that considers the standards of bibliographic modeling with scholarly practice including fields such as Citation item or Citation style but also takes into account the variance that can occur within a work across editions.
• A prime citation/source model that maps the relationships among a single work and its various copies, see Infographic for examples

Data Modeling

Currently, this project is positing the following objects for conceptualization/design:

• CitedInstances
• CitedSources/CitedSourceLocations
• SourceUnits
• SourceSeries

Refer to Data Modeling for more details about this (maybe) emerging data model and a collection of other data considerations.

Citation Scale

Addressing this problem, we want to keep citation scale in mind:

• English language: 29.3 million citations from 6.1 million English Wikipedia articles as of May 2020.
• Overall, in 2022, 200 million references that were extracted from over 40 million Wikipedia articles
• IA Citations - https://blog.archive.org/2026/04/23/gone-but-not-forgotten-recovering-the-dead-web/
• 350 million total, 40 million in Wikidata as of 2025 (need citation)

Projecting scale for ReCite:

• tk

Proof of Concept design: ReCitoid + Database(s)

ReCite: Giving Wikipedia’s citations somewhere to live — and carrying reliability judgments to the moment an editor cites.

We can create a ReCite PoC by mapping the relationships across several existing resources, creating database(s) and a different version of Citoid that is stateful: ReCitoid would link to a citations reference database (or two) that permits editors to view related reliability data.

Figure 2: An outline of how information for ReCitoid (Frontend, I) is fed from ReCite Data (Backend, II). Things in gold would be built by the PoC.

Let’s walk through the front-end and back-end information flow or possibilities in the ReCite system.

Front-end Approaches

There are three paths currently being explored, but ultimately the general idea here is that "ReCitoid" via Visual Editor will suggest a curated reference when possible or allow for the saving/suggestion of a new one that might need to be cleaned up.

Proposed "ReCitoid"-powered search within Visual Editor for ReCite project

More about each approach can be found at Front-end Approaches.

Back-end Approaches — Citation-related database(s) and Wikibase

There are several ways to solve the problems of prime citation and judgment association, exploration will help us to see what’s best. This Proof of Concept is interested in bringing two streams together: ReCite Core -- the organization of streamlined, cleaned citations -- and ReCite Attributes, a collection of Source Series judgments.

Through several phases, we'll try to bridge these two areas eventually into a Wikibase.

Implementing the PoC. Stage 2 presents two possibilities, either growing the initial Zotero database (1a) or creating a new database of pointers to Wikidata (1b). Applying Source Series judgments presents at least three possibilities for development.

Refer to Back-end Approaches for more information.

Trials and Experimentation

• Phase 1 Trial

How Citation Value Grows (Forward thinking)

In terms of demonstrating the value of reliable citations within and beyond Wikimedia (suggestions welcome):

• Standardizing clean, reliable citations. Through the use of ReCite even among a small number of Wikipedia editors, including clean, reliable citations becomes an easier process shared among many people.
• Data savings? There may eventually be some data savings in terms of potentially replacing reference data in Wikipedia articles, as discussed in the Shared Citations proposal (e.g. replacing the in-page citation data with appropriate markup pointing to the citation ID).

One thing that the PoC may allow for is the substitution of full citations with a linked reference – in other words, in a Wikipedia article, one would see something like “JE Hoare, 1999” with a link to the citation. This probably means there is also a set of full bibliography pages somewhere, where the link resolves to, and that the set of bibliography pages represent the latest/greatest ReCite data in full. (Then, in a way, the bibliography pages is what represents ReCite and not the database, which might be useful in a collaboration with Internet Archive.) But all this is TBD.

• Existence of reliable citations data collection as a signal. In a world where AI hallucinates regularly, it may be that having the collection in itself is helpful for verification purposes?
• Ecosystem of projects that utilize Wikipedia judgments. Creating a central location for reliability judgments at the citation level can support a greater ecosystem of projects that seek to utilize these judgements and further increased reliability in Wikimedia content:
  • Signals of source quality, such as Wikisignals.org: Being able to know how many times a given source (at the website domain level, for now) has been cited on Wikipedia (as a whole and on individual language projects) would be helpful, as this can indicate how much the Wiki community considers it to be a credible source.
  • Unstructured discussions of source and article quality: Curious if discussions across talk pages could be captured somehow through a tool like Arbiter? Would have to think about what the data model is there, and how judgements flow down to individual articles
• Claim-level possibilities. Creates possibilities for even further verification – claims have M:M relationships with citations → ReCite Data Modeling
• General Wishlist possibilities, but need to sift: Community wishlist focus area for reference management: https://meta.wikimedia.org/wiki/Community_Wishlist/FA12

Related Documents

(probably still being drafted)

• Testing metadata and other issues:
  • https://en.wikipedia.org/wiki/User:Connieatwork/ReCite_Test_Article_(EN)
  • ReCite Test Article (Meta)
• https://meta.wikimedia.org/wiki/WikiCite/ReCite/Front-end_Approaches
• https://meta.wikimedia.org/wiki/WikiCite/ReCite/Back-end_Approaches
• https://meta.wikimedia.org/wiki/WikiCite/ReCite/Data_Modeling
• https://meta.wikimedia.org/wiki/WikiCite/ReCite/Trials