Editorial illustration of a document-research workspace

GUIDE / RESEARCH

Verify a research reference and its version

A reference found in a search engine, AI answer or citation graph is a starting point. Check the exact document, its version and the passage that actually supports the claim.

01 / GUIDE / RESEARCH

Identification is not validation

A DOI identifies a resource; ORCID distinguishes people and ROR organisations. These identifiers alone do not assess methods or findings. PubMed provides citations and abstracts with links to available text; HAL and repositories may hold different document types and versions. First write down the claim to check and the relevant period.

Guides
Complementary identifiers: DOI for a resource, ORCID for a person and ROR for an organisation. Relationships between them still need checking in documents.
Complementary identifiers: DOI for a resource, ORCID for a person and ROR for an organisation. Relationships between them still need checking in documents.

02 / GUIDE / RESEARCH

Six checks before reusing a reference

Check the consulted object without treating citation counts or open access as a truth label.

01

Document identity

Compare DOI or another identifier, title, authors and publisher. A title or suffix error may identify a different object.

Record to retain · Exact identifier and destination page
02

Version read

Distinguish preprint, accepted manuscript, published version, data and software. Record the file version, not just metadata.

Record to retain · Type, version, file and date
03

Author and affiliation

Check ORCID where available and compare affiliations with the document. Use ROR to resolve an ambiguous organisation.

Record to retain · Identifiers and relevant dated affiliation
04

Corrections and updates

Read linked notices and the publisher page; retain reported corrections, withdrawals and changes instead of leaving an old conclusion intact.

Record to retain · Notice, date and link
05

Passage and method

Open the text and locate the table, figure or paragraph. Record population, period and conditions required to understand the measure.

Record to retain · Precise locator and scope
06

Data and protocol

Follow declared repository or registration links; compare versions, files and planned methods with the reported result.

Record to retain · Repository, version, licence and deviations

03 / GUIDE / RESEARCH

Four steps for verification

Keep a short chain between question, text and conclusion.

  1. 01

    Recover the object

    Search DOI or exact title in Crossref, DataCite or the relevant index, then open the publisher or repository resource.

  2. 02

    Read the right version

    Record type, date and version; check update notices. If only an abstract is accessible, retain that limitation.

  3. 03

    Test support for the claim

    Compare the passage with your question. Correlation, findings in a limited population and general conclusions are not interchangeable.

  4. 04

    Document without inflating evidence

    Keep identifier, URL, version, passage and access date. Do not count two records of one work as independent confirmations.

04 / GUIDE / RESEARCH

Four common situations

Methodological reading cases without invented findings or researchers.

01

An AI suggests a reference

Check existence and the precise passage. A real reference may be assigned to the wrong author or fail to support the generated sentence.

Read the dossier
02

One title appears in several databases

Compare identifiers and versions. PubMed, Europe PMC and a repository may describe one work with different files or metadata.

Read the dossier
03

Researchers share a name

Compare ORCID, publications, dates and affiliation; retain separate hypotheses until attributes support a match.

Read the dossier
04

A result cites a protocol

Open the relevant dated registration. Compare planned criteria, population and findings; absence in one register does not establish absence everywhere.

Read the dossier

A useful reference supports another check

Cite the precise page and version read. A portal title, search engine or access date does not replace the document date. Missing information stays missing; retain the uncertainty in your record.

Compose a public-source reference

06 / FAQ

Questions about publications and identifiers

Does a DOI guarantee peer review?

No. It may identify data, software or preprints, among other objects. Read the document type and editorial process.

Does PubMed always provide full text?

No. PubMed holds citations and abstracts; links may lead to PubMed Central or the publisher. Retain the version actually opened.

Does an ORCID affiliation establish current employment?

No. Check the field’s provenance and dates and the institution’s information; the public record may be partial.

Does no correction mean everything is reliable?

No. Checking updates complements reading methods and limits; it is not comprehensive scientific validation.

How should I cite only an abstract?

State that you consulted the abstract and restrict the claim to what it shows. Do not imply that you reviewed files or the complete method.

07 / Discuss a context

Keep a reference another reader can retrace

A useful record identifies the object, the version read and the conclusion’s scope.