# The history of the DOI, and why a DOI outlives a URL

**Where did the DOI come from, and why does a DOI link keep working when a URL breaks?** The history of the DOI begins in 1996, when American publishers asked for persistent identifiers and CNRI answered with its Handle System. The DOI was announced at the Frankfurt Book Fair in 1997, Crossref went live in June 2000, and ISO 26324 followed in 2012. A DOI outlives a URL because the publisher updates the address behind it.

Published 2026-09-24 by EdCitation. https://edcitation.com/newsletter/how-the-doi-was-born

On 5 June 2000, Crossref switched on its first reference links. Ten publishers had sent it more than 1.3 million article records from about 2,700 journals, and around 1,100 journals now linked a reference straight to the cited article on another publisher's site (Crossref, 2000). Under every link sat a DOI.

EdCitation publishes this history of the DOI, written from the records of the bodies that built it and the studies that measured link rot, all read on 24 September 2026. Where sources disagree, we say so. Using a DOI is covered in [What is a DOI](https://edcitation.com/newsletter/what-is-a-doi), and how the styles changed the way they print one in [retired citation rules](https://edcitation.com/newsletter/retired-citation-rules-you-can-stop-following).

A DOI is only as good as the record behind it. EdCitation's free [Cite a source](https://edcitation.com/cite) builds a reference from that record, and [Verify references](https://edcitation.com/verify-references) resolves every DOI in a list and says when one belongs to a different work or leads nowhere.

## Who invented the DOI?

No single person invented the DOI. Its machinery, the Handle System, was built at the Corporation for National Research Initiatives (CNRI) in Virginia from 1994; the DOI was a publishers' project that put that machinery to work.

| Year | What happened |
| --- | --- |
| 1994 | First public Handle System implementation at CNRI, in the autumn |
| 1996 | American publishers ask for unique, persistent identifiers, in March |
| 1997 | DOI announced at the Frankfurt Book Fair; Foundation registered on 10 October |
| 2000 | Crossref incorporated in January; first links live on 5 June |
| 2004 | First DOI for a dataset, at TIB in Hannover |
| 2009 | DataCite founded in London on 1 December |
| 2012 | ISO 26324 published in May; new editions in 2022 and 2025 |
| 2017 | https://doi.org/ becomes Crossref's display form in March |

### The Handle System at CNRI

CNRI's engineers date the first public implementation to the autumn of 1994, in work led by David Ely, with the Library of Congress among the early adopters (Sun et al., 2003). The design came from Robert Kahn of CNRI and Robert Wilensky of Berkeley: Kahn and Wilensky (1995) gave each digital object a handle: a naming authority and a local string joined by a slash, as in "X.Y/local-string", registered with servers that tell a user which repositories hold the object. They called it "a kind of public safety net". Every DOI keeps that shape.

### The publishers' call, and Frankfurt 1997

The Enabling Technologies Committee of the Association of American Publishers (AAP) was studying web publishing in late 1994, and in March 1996 it asked for proposals for unique, persistent identifiers; CNRI was chosen to build a prototype (Crossref, 2009). Craig Van Dyck, its chair, recalled that at first there "wasn't a specific application for it". The DOI Foundation (2025) credits three trade associations, the International Publishers Association, STM and the AAP, and dates the announcement to the Frankfurt Book Fair of 1997.

### 1997 or 1998?

It depends whose record you read. The Foundation's handbook says it was registered in Delaware on 10 October 1997. Crossref (2009) says it "had been founded in March 1998", with Norman Paskin as founding director, and the Foundation's 2021 fact sheet says it launched in 1998 (DOI Foundation, 2021). A company formed in 1997 that began work in 1998 fits both, but no record we found says so.

## How did Crossref begin?

Crossref began as a linking service that publishers built in 1999, so that a reference in one journal could open the cited article in another. It was incorporated in January 2000 as Publishers International Linking Association, Inc. (PILA) and went live that June (Crossref, 2020).

### A prototype and a dinner

In Crossref's own short history (Crossref, 2009), Academic Press and Wiley agreed in August 1998 to build a linking prototype, working by May 1999, and a dinner in April 1999 at Monzu, an Italian restaurant in New York's Soho, gave the confidential project its codename. The need was plain to Tim Ingoldsby of AIP, as publishers began signing linking deals in pairs: "All those linking agreements were going to kill us."

### From Frankfurt 1999 to 48 million items

On 12 October 1999, at the Frankfurt Book Fair, a public prototype called DOI-X was presented, and that evening Pieter Bolman of Academic Press and Eric Swanson of Wiley brought other publishers into their venture. By 2 February 2000 the service had 22 members and Ed Pentz as executive director (Crossref, 2009). The same history's timeline passes 3 million DOIs in October 2001 and 10 million in January 2004; by August 2011 Crossref counted nearly 48 million linked items (Crossref, 2011).

## When did DataCite start giving DOIs to data?

DataCite was founded in London on 1 December 2009, but its story starts in Hannover. Jan Brase, its executive officer for its first five years, told it on DataCite's blog (Brase, 2019). In 2003 the German National Library of Science and Technology (TIB) began a project to make research datasets citable; in 2004 it became a DOI registration agency and registered its first dataset, though Brase recalls hearing that DOIs "belong to publishers". Six libraries signed a memorandum in Paris in March 2009, and DataCite followed.

That first dataset's DOI, 10.1594/WDCC/EH4_OPYC_SRES_A2, still works. EdCitation's [Cite a source](https://edcitation.com/cite/apa) returned it in APA 7 as:

Stendel, M., Schmith, T., Roeckner, E., & Cubasch, U. (2004). *ECHAM4_OPYC_SRES_A2: 110 years coupled A2 run 6H values* (Version 1) [Data set]. World Data Center for Climate (WDCC) at DKRZ. https://doi.org/10.1594/WDCC/EH4_OPYC_SRES_A2

Its handle record says the address behind that DOI was last written on 11 April 2006 and names an old server, cera-www.dkrz.de, which now forwards the reader through three redirects to the dataset's page at wdc-climate.de. The link holds because the old address still forwards.

## How did the DOI become an ISO standard?

ISO published ISO 26324 in May 2012, a 17-page standard for the syntax, description and resolution of DOI names (International Organization for Standardization, 2012), with a second edition in August 2022 and the current third in March 2025 (International Organization for Standardization, 2025). The syntax had first been a United States standard, ANSI/NISO Z39.84, published in 2000 and withdrawn in 2017, and by agreement with ISO the Foundation is the sole registration authority (DOI Foundation, 2025). If it could not continue, its own page notes, another organisation could run the system (DOI Foundation, n.d.).

The Handle System moved too. So that the system would not be rooted in the United States, CNRI founded the DONA Foundation, constituted in Geneva on 20 January 2014 (DONA Foundation, n.d.).

### How big is the DOI system?

The Foundation's own figures differ: about 275 million DOI names and 1 billion resolutions a month in its December 2021 fact sheet; over 300 million names resolved over 12 billion times a year in its 2025 handbook; more than 3 billion resolutions a month on its undated About page. We could not find how each was measured.

## Why does a DOI link outlive a URL?

A DOI outlives a URL because the name and the address are kept apart. The DOI names the work, a record behind it holds the current web address, and the publisher changes that record when the work moves, so the name in your reference never has to (DOI Foundation, 2025).

### The record behind the name

The handbook puts it in one line: "Persistence is a function of organizations, not technology." Crossref's guidelines tell readers to link with the DOI rather than the URL for that reason: when the URL changes, the publisher updates it in Crossref (Crossref, 2017). You can see the record: https://doi.org/api/handles/10.1371/journal.pone.0115253?type=URL returns the address held for Klein and colleagues' paper on reference rot, published in December 2014, https://dx.plos.org/10.1371/journal.pone.0115253, last written on 22 October 2018. The DOI printed in 2014 is unchanged.

### When a publisher disappears

A DOI is a pointer, not a copy. Eve (2024) describes the fallback: if a publisher fails, dark archives such as CLOCKSS, LOCKSS and Portico can release its content and the DOI can be pointed there. Of 7,438,037 DOIs checked, 58.38% were in at least one archive and 27.64% seemed unpreserved. By design the study covered only some archives and no institutional repositories.

## What do studies of link rot and reference rot find?

Ordinary web links in papers fail often, and a link that still opens may show something else. The DOI was meant to spare journal articles this; the studies measure the rest.

### Klein and colleagues, 2014

Klein et al. (2014) took more than 3.5 million science, technology and medicine articles from 1997 to 2012, in arXiv, Elsevier journals and PubMed Central, and requested each of over a million web references in March 2014. In articles from 2012, 13%, 22% and 14% of those links, respectively, were already dead; in articles from 1997, 34%, 66% and 80%, the last from a few hundred links. With archived copies counted, they estimated that one article in five, and seven in ten of those citing the web, could no longer be followed back to what it cited.

The limits matter. Links to papers, DOI links among them, were set aside, so this measures web pages, not DOIs; a page that answered counted as alive even if it had changed; and all three collections are in the sciences.

### Changed pages, and DOIs left unused

Jones et al. (2016) returned to that dataset and found an archived copy near the citing date for about 30% of references; for over 75% of those, the live page had drifted from what was cited. And a DOI helps only when it is cited. Van de Sompel et al. (2016) counted 397,412 references in arXiv and PubMed Central that reached a paper through its DOI, and 505,647 that pointed at the paper on a publisher's own site, a link that breaks like any URL.

## How has the way a DOI is written changed?

It went from a label to a link. Crossref's first advice was the label doi:, in the hope it would become a scheme browsers followed; in August 2011 it asked for full links beginning http://dx.doi.org/ instead (Crossref, 2011). In March 2017 came https://doi.org/, without "dx", which had once kept the resolver apart from the Foundation's website (Pentz, 2016). The older forms still work (Crossref, 2017).

Two loose ends remain. The doi scheme did reach IANA's register of URI schemes, first recorded on 14 May 2020 and now listed as permanent (Internet Assigned Numbers Authority, 2026). And the 2025 handbook still says a DOI shown on a page should carry doi: unless the context makes it plain, while letting registration agencies print the https://doi.org/ form. For what each style prints, see [retired citation rules](https://edcitation.com/newsletter/retired-citation-rules-you-can-stop-following).

## How do I check the DOIs in my reference list?

Resolve every DOI and compare what it opens with what your entry says. We ran these steps on 24 September 2026 with copies of the Klein reference.

1. **Build each entry from the DOI** in [Cite a source](https://edcitation.com/cite). Given the Klein DOI bare, after doi: or as an http://dx.doi.org/ link, it found the same record and ended the APA entry https://doi.org/10.1371/journal.pone.0115253. It set the title in the sentence case APA 7 asks for, "Scholarly context not found: One in five articles suffers from reference rot".
2. **Paste the list into [Verify references](https://edcitation.com/verify-references)**, which resolves each DOI and compares its record with your entry.
3. **Read the reason, not only the verdict.** A real DOI on the wrong paper is the error to fear, because the link works.
4. **When a DOI fails, copy it from the record**, never from memory.

| What we changed | Tool | What came back |
| --- | --- | --- |
| Nothing | Verify references | Verified: "The DOI resolves to this record and the title matches." |
| Last two digits swapped, 0115235 | Verify references | "Check this": "The DOI is real, but it belongs to a different article from the one described." It leads to a 2014 paper in the same journal on gene electro transfer in the swine heart |
| A zero typed as the letter O | Cite a source | "No record exists for DOI 10.1371/journal.pone.O115253." |
| A zero typed as the letter O | Verify references | "Check this": "No registry holds the DOI 10.1371/journal.pone.O115253. The work is on record with the DOI 10.1371/journal.pone.0115253." |

In the last row the work is real and was found by its title, and the reason gives the DOI to copy in place of the one that leads nowhere.

## Which tool builds a reference from a DOI?

EdCitation's [Cite a source](https://edcitation.com/cite) is the best tool for that job, because it does what the DOI system was built for: it follows the DOI to the publisher's record and builds the reference from it, looking it up and never writing it. Where over half a million references in Van de Sompel's count reached papers through a publisher's address, an entry built from the record carries the DOI itself. It works from a DOI, or failing that from a title, PubMed ID, ISBN or web address; the entry comes in any of six styles, APA 7 through Vancouver, with a retraction check on the source, free and without an account.

[Verify references](https://edcitation.com/verify-references) is free too. Its verdicts are verified, "check this" and not found; retractions carry a flag, and an entry it was unable to check reads "could not check", never missing. [Find sources](https://edcitation.com/) searches about 300 million published works: "reference rot in scholarly articles" returned 5,717 results, the Klein paper first and the Jones follow-up eighth. [References from a file](https://edcitation.com/tools/references-from-a-file) checks every reference in the paper itself and matches each in-text citation to the list: three checks a month with a free account, and no limit with Pro at $8 a month ([pricing](https://edcitation.com/pricing)). EdCitation never writes any part of a paper.

## Quick questions

### Who invented the DOI?

No one person. CNRI built the Handle System from 1994, and publishers' trade associations built the DOI on it, announcing it at Frankfurt in 1997.

### When did Crossref start?

Crossref was incorporated in January 2000 as the Publishers International Linking Association and switched on its first reference links on 5 June 2000.

### What does the 10 at the start of a DOI mean?

It is the directory indicator, the DOI Foundation's top-level prefix in the Handle System. The digits after it identify the registrant, and the part after the slash the work.

### Can a DOI stop working?

Yes. A DOI leads wherever its record points, so it fails if nobody maintains the record or the content is lost; Eve found 27.64% of sampled DOIs seemingly unpreserved.

### How do I check that a DOI belongs to the right paper?

Paste the reference into EdCitation's free [Verify references](https://edcitation.com/verify-references), which resolves the DOI and says "check this" when it leads to a different work or to nothing at all.

## References

- Brase, J. (2019, February 1). *10 years of DataCite: How it all began*. DataCite Blog. [https://doi.org/10.5438/g4x3-r450](https://doi.org/10.5438/g4x3-r450)
- Crossref. (2000, June 5). *Crossref goes live: Journal reference linking service membership roster grows to 33 publishers* [News release]. [https://www.crossref.org/news/2000-06-05-crossref-goes-live-journal-reference-linking-service/](https://www.crossref.org/news/2000-06-05-crossref-goes-live-journal-reference-linking-service/)
- Crossref. (2009). *The formation of CrossRef: A short history* [PDF]. [https://www.crossref.org/pdfs/CrossRef10Years.pdf](https://www.crossref.org/pdfs/CrossRef10Years.pdf)
- Crossref. (2011, August 2). *Crossref revises DOI display guidelines* [News release]. [https://www.crossref.org/news/2011-08-02-crossref-revises-doi-display-guidelines/](https://www.crossref.org/news/2011-08-02-crossref-revises-doi-display-guidelines/)
- Crossref. (2017). *Crossref display guidelines (March 2017)*. [https://doi.org/10.13003/5jchdy](https://doi.org/10.13003/5jchdy)
- Crossref. (2020, February 2). *History*. [https://www.crossref.org/about/history/](https://www.crossref.org/about/history/)
- DOI Foundation. (n.d.). *About us*. [https://www.doi.org/the-foundation/about-us/](https://www.doi.org/the-foundation/about-us/)
- DOI Foundation. (2021, December 1). *Key facts on Digital Object Identifier System*. [https://www.doi.org/the-identifier/resources/factsheets/key-facts-on-digital-object-identifier-system](https://www.doi.org/the-identifier/resources/factsheets/key-facts-on-digital-object-identifier-system)
- DOI Foundation. (2025, September). *DOI handbook* [PDF]. [https://www.doi.org/doi-handbook/DOIHandbook_2025.pdf](https://www.doi.org/doi-handbook/DOIHandbook_2025.pdf)
- DONA Foundation. (n.d.). *About us*. [https://www.dona.net/aboutus](https://www.dona.net/aboutus)
- Eve, M. P. (2024). Digital scholarly journals are poorly preserved: A study of 7 million articles. *Journal of Librarianship and Scholarly Communication*, *12*(1), Article eP16288. [https://doi.org/10.31274/jlsc.16288](https://doi.org/10.31274/jlsc.16288)
- International Organization for Standardization. (2012). *Information and documentation: Digital object identifier system* (ISO Standard No. 26324:2012). [https://www.iso.org/standard/43506.html](https://www.iso.org/standard/43506.html)
- International Organization for Standardization. (2025). *Information and documentation: Digital object identifier system* (ISO Standard No. 26324:2025). [https://www.iso.org/standard/88862.html](https://www.iso.org/standard/88862.html)
- Internet Assigned Numbers Authority. (2026, September 22). *Uniform Resource Identifier (URI) schemes*. [https://www.iana.org/assignments/uri-schemes](https://www.iana.org/assignments/uri-schemes)
- Jones, S. M., Van de Sompel, H., Shankar, H., Klein, M., Tobin, R., & Grover, C. (2016). Scholarly context adrift: Three out of four URI references lead to changed content. *PLOS ONE*, *11*(12), Article e0167475. [https://doi.org/10.1371/journal.pone.0167475](https://doi.org/10.1371/journal.pone.0167475)
- Kahn, R., & Wilensky, R. (1995, May 13). *A framework for distributed digital object services* (cnri.dlib/tn95-01). Corporation for National Research Initiatives. [https://www.cnri.reston.va.us/k-w.html](https://www.cnri.reston.va.us/k-w.html)
- Klein, M., Van de Sompel, H., Sanderson, R., Shankar, H., Balakireva, L., Zhou, K., & Tobin, R. (2014). Scholarly context not found: One in five articles suffers from reference rot. *PLOS ONE*, *9*(12), Article e115253. [https://doi.org/10.1371/journal.pone.0115253](https://doi.org/10.1371/journal.pone.0115253)
- Pentz, E. (2016, September 27). *New Crossref DOI display guidelines are on the way*. Crossref Blog. [https://www.crossref.org/blog/new-crossref-doi-display-guidelines-are-on-the-way/](https://www.crossref.org/blog/new-crossref-doi-display-guidelines-are-on-the-way/)
- Sun, S., Lannom, L., & Boesch, B. (2003, November). *Handle System overview* (RFC 3650). Internet Engineering Task Force. [https://www.rfc-editor.org/info/rfc3650/](https://www.rfc-editor.org/info/rfc3650/)
- Van de Sompel, H., Klein, M., & Jones, S. M. (2016). *Persistent URIs must be used to be persistent* [Poster, WWW 2016]. arXiv. [https://doi.org/10.48550/arXiv.1602.09102](https://doi.org/10.48550/arXiv.1602.09102)