DOI, PMID or URL — which identifier belongs in your reference?

Most reference entries carry whichever identifier the database happened to show when the reference was copied. They are not interchangeable, and the one you pick decides whether your citation still works in ten years.

This is the short, practical version: what each identifier is, which to prefer, and what to do when a source has none.

The three you will actually meet

  DOI          the work itself
               assigned by the publisher, registered
               centrally, permanent by design
               10.xxxx/journal.2023.0447

  PMID         the RECORD of the work in PubMed
               assigned by a database, stable, but
               biomedical literature only
               38104772

  URL          a location, not a work
               guaranteed by nobody; changes when a
               site is reorganised
               https://example.org/reports/2022/...

The distinction in the middle column is the whole thing. A DOI names the paper. A PMID names an index card describing the paper. A URL names a shelf the paper was on last time anyone looked.

A DOI is a promise by the publisher that the identifier will keep pointing at the work. A URL is a description of where a file sat on an afternoon.

The order of preference

DOI, always, when one exists. It survives the journal changing publisher, the platform being retired and the site being redesigned — all of which happen over the life of a doctorate. It is also the strongest evidence a reference checker has, which matters practically: an entry with a DOI can be resolved to its source and confirmed, and an entry without one can only be searched for.

PMID as a companion, not a replacement. In biomedical fields a PMID is genuinely useful — PubMed is stable, its records are curated, and some journals ask for it. But it identifies a database record rather than the paper, and it means nothing outside biomedicine. Carry both where the style allows; carry the DOI if forced to choose.

URL only when nothing else exists. Reports, government statistics, standards, working papers, datasets on institutional pages — much of the grey literature has no identifier at all, and a URL is the honest answer. It just needs the supports that a DOI would otherwise provide: an access date, and ideally an archived snapshot.

What to do when there is no DOI

Two moves, in this order.

First, check whether one exists. A surprising share of entries lack a DOI simply because it was not in the view the reference was copied from — the paper has one, the entry does not. A lookup by title and author against the bibliographic databases will usually find it in seconds, and adding it to the entry costs nothing and gains permanence.

Second, if there genuinely is none, cite the URL properly: full address, access date, and the version or edition if the document carries one. Do not substitute a search-results link or a login-walled address that names nothing to a reader without a subscription.

Three formatting traps

Styles disagree about presentation, and the disagreements are exactly the kind that a proofreader will not catch because each version looks reasonable.

  https://doi.org/10.xxxx/abc.2023.0447   a resolvable link
  doi:10.xxxx/abc.2023.0447               the older prefix form
  DOI: 10.xxxx/abc.2023.0447              a label, not a link
  10.xxxx/abc.2023.0447                   the bare identifier

  all four name the same work
  your style guide wants one of them
  and consistency matters more than
  which one, unless the guide is explicit

The traps: a DOI broken across a line by the word processor and silently gaining a hyphen; a trailing full stop from the end of the sentence being absorbed into the identifier; and a DOI copied with tracking parameters attached from a publisher's link. All three produce an identifier that fails to resolve while looking entirely normal on the page.

The others you will occasionally meet

Four more turn up often enough to be worth recognising, and each answers a narrower question than a DOI does.

  arXiv ID     a preprint on arXiv
               2301.04471 — versioned, so v1
               and v3 are different documents

  PMCID        the FULL TEXT in PubMed Central
               different from a PMID, which is
               only the record

  ISBN         an edition of a book — genuinely
               useful, because editions are what
               books vary by

  ORCID        an AUTHOR, not a work
               belongs in author disambiguation,
               never in place of a reference id

Two of those repay attention. An arXiv identifier is versioned, so citing one without its version number is the preprint equivalent of citing a book without its edition — and the version you read may not be the version now served. An ISBN is the cleanest way to pin which edition of a book you used, which is otherwise the hardest thing to state unambiguously.

ORCID is the one people misplace. It identifies a person, so it belongs in the discussion of who wrote something, not in the field that says which work you are pointing at.

Why this matters more than it sounds

An identifier is not decoration on a reference. It is the thing that lets anyone — your examiner, a reader, a checker — get from your bibliography to the actual work without guessing.

It is also the difference between a reference that can be verified and one that can only be approximately matched. A DOI resolves to a page that either names your paper or does not, and that is a fact. Without one, confirming an entry means searching by title and author and judging whether the result is the same work — a judgement that is usually right and is never certain. The failure mode when it goes wrong is in the DOI that pointed at the wrong paper, and the slower decay of URLs is in link rot and DOI rot.

What this does not claim

A missing DOI is not an error. Plenty of legitimate sources have never had one — older papers, books from small presses, most grey literature, and anything published before the system existed. An entry without an identifier is not a defect and the report does not treat it as one, or as any kind of finding of misconduct.

And an identifier that resolves proves only that the identifier is correct. Whether the paper says what you cited it for is a question no identifier answers, and no reference checker is in a position to ask.