How Persistent Identifiers Work

The web is full of addresses. Addresses are not the same thing as identities. A book can move between shelves without becoming a different book. A person can change employer without becoming a different person. A research dataset can move to a new server without becoming a different dataset. A journal article can be hosted by a new publishing platform without ceasing to be the article that was originally cited.

Ordinary URLs often mix identity with location. They tell a browser where to retrieve something now. When a website is reorganised, a domain expires or a repository migrates, the location changes and the link may break. Persistent identifier systems try to separate the durable identity of a thing from its current network location.

The technical trick is not enough. The DOI Foundation makes the deeper point explicit: persistence is a function of organisations, policies and processes, not technology alone. A persistent identifier remains useful because somebody commits to maintaining the connection between the identifier, the thing identified, its metadata and its current location.

The persistent-identity loop

THING
→ DEFINE IDENTITY BOUNDARY
→ ASSIGN IDENTIFIER
→ REGISTER
→ ATTACH METADATA
→ RESOLVE OR LOOK UP
→ CHANGE LOCATION
→ UPDATE TARGET OR RECORD
→ PRESERVE IDENTIFIER
→ VERSION / RELATE
→ CORRECT
→ RETIRE OR TOMBSTONE RESPONSIBLY
→ FUTURE DISCOVERY

Every stage solves a different problem. Assignment gives a label. Registration prevents accidental collision. Metadata tells users what the label refers to. Resolution connects the identifier to a current service or location. Governance keeps the connection alive through organisational change. Version relationships prevent “the same thing” from becoming an unexamined assumption.

1. An identifier is a claim about identity

An identifier says: this token refers to this entity inside a defined system. The entity might be a person, article, dataset, book edition, organisation, specimen, software release, grant, instrument, place or concept.

The token does not have to resemble the thing. A number can identify a book. A URI can identify a concept. An alphanumeric string can identify a researcher. The identifier’s strength comes from the rules connecting token and entity, not from human-readable wording.

This connects directly to When Are Two Things the Same Thing?. Before assigning an identifier, a system must decide what counts as one entity and when a change creates a new one.

2. Identity boundaries are domain decisions

A paperback and hardback can contain nearly identical words but receive different ISBNs because the ISBN system identifies particular publication products and formats. A corrected journal article may retain one work-level scholarly identity while producing a new version relationship. A researcher keeps one ORCID iD across employer changes because the person, not the affiliation, is the identified entity.

There is no universal rule saying how finely reality should be identified. The correct granularity depends on the purpose of the identifier system.

3. Uniqueness is scoped to a namespace

An identifier must be unique inside the namespace whose rules give it meaning. The number 1234 can identify countless things in different systems without conflict. What matters is the combination of namespace and local identifier.

This is why prefixes and structured identifier schemes matter. They prevent one organisation’s token from colliding with another’s and allow software to determine which authority or resolver should interpret the identifier.

4. A URL identifies a location more naturally than a durable entity

HTTP URLs can be persistent if their owners maintain them responsibly, but ordinary web practice often embeds organisational structure in the path: department names, content-management IDs, dates, file extensions or temporary routing. When the site changes, the URL changes.

A persistent identifier system therefore adds a governance layer. The identifier can remain stable while its resolver record is updated to point to a new location. The user cites the durable identifier; the infrastructure handles movement.

5. Persistence is a promise, not a magical property of a string

Any organisation can create a long random string and call it permanent. That does not make it persistent. Persistence requires institutional continuity, documented assignment rules, maintenance processes and a way to update or redirect records as the world changes.

The DOI Foundation explicitly states that persistence depends on social infrastructure layered over technical infrastructure. That is one of the most important ideas in digital preservation: durability is organisational behaviour.

6. Resolution separates identity from current location

A resolvable identifier can be submitted to a resolver, which returns or redirects the user toward a current resource or service. If the resource moves, the registration record can be updated without changing the identifier itself.

The architecture looks like this:

PERSISTENT IDENTIFIER
→ RESOLUTION SERVICE
→ REGISTRY RECORD
→ CURRENT TARGET / METADATA / SERVICE

The persistent identifier therefore becomes an indirection layer. Indirection adds complexity, but it protects references from ordinary location change.

7. DOI identifies objects through governed registration

The Digital Object Identifier system is standardised as ISO 26324. DOI names can identify research articles, datasets, reports, books and many other objects. They are registered through DOI Registration Agencies that operate under system rules and maintain metadata and resolution targets.

A DOI such as 10.xxxx/xxxxx contains a prefix associated with a registrant structure and a suffix chosen under the registrant’s rules. The suffix does not need to encode human meaning. It needs to remain unique within the relevant DOI namespace.

Crossref’s DOI display guidelines recommend showing Crossref DOIs as full HTTPS links using https://doi.org/..., emphasising that a DOI is both an identifier and a persistent link.

8. A DOI should not be confused with ownership or truth

A DOI tells you that an object has been registered in the DOI infrastructure. It does not prove that every claim inside the object is correct, that the work was peer reviewed, or that the registrant owns every right associated with the content.

Identifiers improve citation and linkage. They do not replace source evaluation. See How Research Methods and Source Evaluation Work for the evidential layer that begins after identity has been resolved.

9. DOI persistence depends on metadata maintenance

If a publisher migrates an article from one platform to another, the DOI should stay the same while the registered URL is updated. This maintenance obligation is the practical mechanism behind persistence.

A DOI whose target is never maintained can still fail operationally even though its identifier string survives. Persistent infrastructure therefore needs monitoring, correction and organisational accountability.

10. Crossref and DataCite use DOI infrastructure for different communities and object types

Crossref is strongly associated with scholarly publications and the citation network surrounding them. DataCite focuses heavily on research data and other research outputs, although the boundaries are not absolute. Both operate as DOI Registration Agencies and both rely on rich metadata to make identifiers discoverable and useful.

DataCite’s Metadata Schema provides a structured set of properties for accurate and consistent identification, citation and retrieval. Version 4.7 was released on 3 March 2026, adding new resource and identifier values while preserving the larger metadata model.

11. ORCID identifies people across changing names and affiliations

An ORCID iD is a persistent digital identifier for a researcher or contributor. ORCID’s purpose is to distinguish people who may share names, publish under different name forms, change institutions or move across disciplines.

ORCID explains that the iD belongs to the person across a career. The individual’s record can link contributions and affiliations while the core identifier remains stable. This separates person identity from mutable descriptions of that person.

That separation is essential. An affiliation is not a person. A job title is not a person. A spelling variant is not automatically a different person.

12. Persistent identifiers can identify organisations too

Research infrastructure often needs stable organisation identity because institutions merge, rename, split and operate under aliases. Systems such as the Research Organization Registry, ROR provide identifiers and metadata for research organisations.

This allows a publication, grant or researcher record to link to an organisation through an identifier rather than relying on a free-text institution name that may be spelled differently in every database.

13. ISBN identifies publication products, not abstract literary works

The International Standard Book Number is a standardised identifier for books and related publication products. The International ISBN Agency coordinates the global ISBN system through national and regional agencies.

Since 2007, ISBNs contain 13 digits. The structure includes a prefix, registration group, registrant, publication element and check digit. A different edition or format can require a different ISBN because the identifier’s job is to distinguish publication products in the book supply and cataloguing ecosystem.

This demonstrates why identity granularity must match function. “The novel”, “the 2026 paperback edition” and “this particular physical copy” are different identity layers.

14. Identifiers do not all resolve through the web

ISBNs are identifiers, but the ISBN string itself is not inherently a universal web-resolution mechanism in the same way as a DOI URL. Libraries, booksellers and databases use ISBNs as keys inside their own systems.

This distinction matters. Identifier, persistent identifier and resolvable identifier overlap, but they are not synonyms. A system should describe the actual guarantees it provides.

15. The Handle System provides a general resolution architecture

The Handle System is a distributed infrastructure for assigning, managing and resolving identifiers called handles. DOI resolution is built on Handle System technology, but handles can also be used outside the DOI governance structure.

This distinction illustrates the difference between technical substrate and institutional scheme. The same underlying resolution technology can support identifier systems with different assignment policies, metadata requirements and governance.

16. ARKs are designed for durable references across memory institutions

Archival Resource Keys, ARKs, are persistent identifiers designed to support long-term access to information objects. The ARK Alliance describes ARKs as open, decentralised identifiers used by libraries, archives, museums, data centres and other institutions.

ARKs can identify digital, physical or abstract entities. This breadth makes them especially interesting for cultural heritage, where a collection may need identifiers for manuscripts, paintings, specimens, digital surrogates, vocabulary concepts and archival components.

17. Persistent identifiers become more valuable when they are linked

A DOI for a paper, an ORCID iD for an author, a ROR identifier for an institution and a DOI for the supporting dataset can be connected in metadata. The result is not merely four stable identifiers. It is a network of typed relationships.

PERSON ──authored──▶ ARTICLE
PERSON ──affiliatedWith──▶ ORGANISATION
ARTICLE ──usesData──▶ DATASET
DATASET ──fundedBy──▶ AWARD
ARTICLE ──cites──▶ ARTICLE

This is one bridge between persistent identifiers and Knowledge Graphs and Semantic Data. Stable node identity makes relationships more reliable.

18. Metadata turns an opaque token into a discoverable object

An opaque identifier can be perfectly unique yet useless to a human who does not know what it refers to. Metadata supplies title, creator, publisher, date, resource type, relationships, rights and other descriptive fields.

The identifier anchors the record. Metadata makes the record intelligible and discoverable. Together they support reliable linking.

This is why identifier infrastructure and metadata infrastructure should be designed together rather than as separate afterthoughts.

19. Identifier metadata must itself be versioned

Titles can be corrected. Organisations can rename. URLs move. Rights change. Relationships are discovered later. The identifier may remain stable while its metadata evolves.

A mature registry distinguishes the enduring identity of the object from mutable descriptive claims about that object. It also preserves enough change history to audit major corrections where required.

20. Versioning is one of the hardest identifier problems

Should a new version receive a new identifier? The answer depends on what users need to cite and distinguish. If two versions can produce meaningfully different results, separate identifiers may be essential. If the changes are minor metadata corrections, a stable identifier may be more appropriate.

Strong systems model relationships such as isVersionOf, hasVersion, isNewVersionOf or isIdenticalTo rather than pretending identity is binary.

The DataCite schema includes explicit relationship types that allow research objects to be connected across versions, components, collections and derivations.

21. Tombstones are better than silent disappearance

Sometimes an identified object must be withdrawn, deleted, retracted or made inaccessible. A persistent system should avoid simply returning “not found” with no explanation when lawful and ethical constraints allow a record to remain.

A tombstone page can state that the object existed, identify its former citation, explain its status and point to a replacement or correction. This preserves the scholarly and archival record without pretending the withdrawn object is still normally available.

22. Persistence is compatible with correction

A common misunderstanding is that persistence means nothing may change. The opposite is usually true. Persistent identity makes responsible change easier because the stable identifier allows metadata, locations and status to be updated without breaking every reference.

Persistence protects the identity thread while correction changes what is known or where the object lives.

23. Check digits catch transcription errors, not identity errors

Identifier systems such as ISBN and ORCID use check digits or related validation mechanisms so common transcription mistakes can be detected. A syntactically valid identifier can still refer to the wrong entity if it was assigned incorrectly or attached to the wrong record.

Validation therefore has two levels: does the string conform to the scheme, and does the scheme’s record correctly identify the intended thing?

24. Human-readable identifiers trade mnemonic value against stability

Identifiers that embed titles, dates, departments or organisation names can be easier to recognise. They can also become misleading when those properties change. Opaque identifiers avoid that semantic decay but are harder for humans to remember.

Many robust schemes therefore keep the identifier relatively semantically light and place descriptive meaning in metadata that can change safely.

25. Local identifiers and global identifiers solve different problems

A museum accession number can be an excellent identifier inside one institution even if nobody outside the institution can resolve it globally. A database primary key may be perfectly unique inside one database. Global research infrastructure requires identifiers that can be interpreted across organisational boundaries.

The right strategy is often layered identity:

LOCAL SYSTEM ID
↕ crosswalk
INSTITUTIONAL ID
↕ crosswalk
GLOBAL PERSISTENT ID
↕ relationships
OTHER GLOBAL IDS

Crosswalks preserve local operational needs while allowing broader discovery.

26. Duplicate identifiers are a data-governance problem

Two identifiers can accidentally be assigned to what should have been one entity. Conversely, one identifier can be reused for two entities if governance is weak. Both errors damage downstream linking.

Registries therefore need duplicate detection, merge policies, alias handling and deprecation rules. An identifier system is a living institution, not a one-time numbering operation.

27. Aliases should point toward identity, not multiply it invisibly

A person may have multiple names. An organisation may have a former name and a current name. A book may have translated titles. These are aliases, labels or variants, not necessarily separate entities.

Authority-control systems and persistent identifiers complement one another. The identifier provides stable identity; authority data records preferred and variant labels. This is the bridge to the Library’s authority-control layer.

28. Identifier spoofing and lookalikes create trust risks

Users can be misled by text that resembles a trusted identifier but resolves somewhere else. Phishing links, Unicode lookalikes and fabricated identifier strings exploit the fact that people often trust familiar syntax.

Applications should validate scheme syntax, use known resolver domains, retrieve registry metadata and avoid treating identifier-shaped text as verified identity without resolution.

29. Persistent identifiers make citations more machine-actionable

A free-text citation asks software to infer which object was meant. A citation carrying a DOI can be resolved directly to a registry record. An author carrying an ORCID iD can be distinguished from people with the same name. A dataset DOI can be linked to the paper that analysed it.

This reduces ambiguity and allows citation networks, discovery services and research graphs to be constructed more reliably.

30. Persistent identifiers improve FAIR data practices

The FAIR Principles emphasise that data and metadata should be findable, accessible, interoperable and reusable. Globally unique and persistent identifiers are central to findability because they give digital objects stable references that other metadata can point toward.

See Research Data Management and FAIR Principles for the wider lifecycle. Persistent identification is one layer of FAIR practice, not a substitute for metadata, licences or quality.

31. Libraries have always separated identity from shelf location

The digital problem is new in technology but old in principle. Libraries distinguish a work, an edition, an item and its current shelf location. Catalogues allow users to retrieve the intellectual object even when the physical copy is moved.

Persistent identifiers extend that logic across networked information. See How Libraries Work for the collection architecture beneath modern identifier systems.

32. Archives need identifiers that survive reorganisation

Archival collections can be rearranged intellectually, digitised, migrated to new systems or transferred between institutions. If identifiers are tightly coupled to one folder path or database row, references break during preservation work.

Stable archival identifiers allow descriptions and digital representations to move while citations remain meaningful. ARK infrastructure is one prominent approach, but the underlying principle applies regardless of scheme.

33. Persistent identity needs rights and privacy boundaries

A globally persistent identifier can make an entity easier to track across systems. For publications and public datasets this is often desirable. For people, sensitive records or restricted cultural materials, persistent linking can create privacy and governance concerns.

Identifier design should therefore distinguish public identity from public disclosure. An identifier may remain stable while access to associated metadata is restricted according to lawful and ethical rules.

34. Persistent identifiers do not guarantee permanence of the object

An object can be lost while its identifier survives. A website can disappear while a registry retains the citation. A physical specimen can be destroyed. A dataset can become legally unavailable.

The identifier preserves the historical fact of identity and can point to status information, but preservation of the object itself requires separate storage, conservation, redundancy and rights arrangements.

This separates the identity layer from the preservation layer described in How Archives Work and How Data Management Works.

35. Identifier quality can be audited

A mature identifier system can test whether:

the identifier resolves;
the registry metadata is complete enough for its purpose;
the target represents the intended entity;
aliases and duplicate records are controlled;
version relationships are explicit;
withdrawn objects have meaningful status pages;
rights and access states are current;
the assigning organisation remains accountable;
machine interfaces expose the identifier consistently.

Persistence should therefore be monitored as an operational quality attribute.

36. A practical identifier-selection protocol

Define the entity. Person, work, version, physical item, organisation, dataset or concept?
Check whether an established community scheme already owns the job.
Choose the right granularity.
Prefer identifiers whose governance matches the required lifetime.
Record namespace and scheme, not only the local token.
Attach authoritative metadata.
Model versions and relationships explicitly.
Keep resolution targets maintained.
Preserve status after withdrawal where appropriate.
Crosswalk local IDs to global IDs rather than deleting local identity.

37. Persistent identifiers in the eduKate Library

A large knowledge estate eventually needs more than URLs and titles. Articles are revised. canonical owners move. research collections produce editions. datasets and source records need stable references. AI retrieval works better when entities can be linked through durable identity instead of fuzzy text matching alone.

This article therefore owns the general persistent-identity method. It does not replace the Library’s own accession rules or any private runtime. It provides the public conceptual layer required to understand why global knowledge systems use identifiers, namespaces, registries, resolution and metadata.

The internal route is deliberately modular: Libraries own collection infrastructure; Standards own conformance; Knowledge Graphs own entity relationships; Scholarly Publishing owns publication workflow. Persistent identifiers connect those owners without absorbing them.

38. Identity is infrastructure when change no longer breaks reference

The deepest value of a persistent identifier is not that the string lasts forever. It is that a community can continue referring to the same entity while locations, names, organisations, technologies and descriptions change around it.

That stability makes correction possible. It makes citation cumulative. It allows metadata to improve. It lets archives migrate. It allows research graphs to grow. It gives machines something more reliable than a guess about whether two names refer to the same thing.

39. World Return from persistent identity

The World Return of persistent identifiers is continuity. A future reader can follow a citation after a website redesign. A dataset remains identifiable after repository migration. A researcher remains connected to work after changing institution. A museum object can carry one identity across digitisation projects. A correction can point backwards without erasing the original scholarly trail.

Persistent identifiers turn knowledge from a collection of temporary addresses into a network of durable references.

Sources and further reading

Continue through eduKate

A durable identifier is not a permanent location. It is a maintained agreement about identity. The identifier survives because the community preserves the rules, metadata, resolution and correction machinery that allow the same thing to be recognised through change.

How Persistent Identifiers Work | From Names and URLs to DOI, ORCID, ISBN, ARK, Resolution, Metadata and Long-Term Identity