Why we ran this
When you paste a DOI into a citation generator, the tool does not read the article. It asks Crossref for the record the publisher deposited when the DOI was registered, then arranges that record in your style. If the record says the first author is “J.”, the citation says “J.”. If the record has no page range, neither does the citation.
Students are told to double check generated citations, but rarely what to check or how often something is wrong. We wanted a number, so we measured the records themselves rather than any particular tool.
How we checked
| Step | What we did |
|---|---|
| Source | The public Crossref REST API, filtered to records of type journal article |
| Sampling | Crossref's random sample option, 100 records per request, ten requests, duplicates removed |
| Sample | 1,000 unique DOIs from 317 publishers and 918 journals |
| Pulled | 31 October 2026 |
| Years covered | 1845 to 2026, median year 2013 |
| Measured | Every element a reference list entry uses: authors and first names, title, journal, volume, issue, pages, article number and date |
Two limits are worth stating plainly. The sample is random across everything registered as a journal article, so older material is better represented than in a typical student reference list, which leans recent; the section on publication era below shows how much that matters. And a record is not the article: a field missing from the record may be printed on the paper itself, which is exactly why a quick look at the article fixes most problems.
Citation data by the numbers
| What we checked | Share of records | What it means for your citation |
|---|---|---|
| Hold every element APA needs | 84.0% | Author, title, journal, volume, page range or article number, and year are all present |
| Also hold an issue number | 74.5% | The same, plus the issue that APA adds when a journal uses issues |
| No author at all | 9.1% | The generator has nothing to put first; check whether the article is truly unsigned |
| Authors given as initials only | 18.9% | Fine for APA, but MLA, Chicago and Turabian want first names as printed |
| At least one author missing a first name | 2.9% | Usually a single name deposited in the wrong field |
| Title entirely in capitals | 6.2% | Has to be retyped in sentence case or title case |
| Formatting code inside the title | 4.3% | Italic species names, subscripts and symbols arrive as tags |
| Subtitle stored in a separate field | 5.3% | A tool that reads only the title field drops it |
| Volume present | 96.1% | Rarely missing |
| Issue present | 87.1% | Missing in one record in eight |
| Page range present | 88.1% | The rest use an article number, or nothing |
| Article number instead of pages | 10.0% | APA writes Article and the number in place of pages |
| Neither pages nor an article number | 5.7% | Usually an online first version; check the publisher page |
| Publication year only, no month | 13.1% | Fine for most journal articles in most styles |
Authors: missing, abbreviated or not people
The author list is the first thing a reader sees and the part most often thin. 9.1 percent of the journal article records had no author at all. Many of those are genuinely unsigned items such as editorials, notices and front matter, but a generator cannot tell the difference, so it simply starts the citation with the title.
A further 18.9 percent gave every author as initials only, as in “J. R.” rather than “James R.”. That costs nothing in APA, which uses initials anyway. It does matter in MLA, Chicago and Turabian, which write first names as they appear in the source. A generator cannot expand an initial it was never given, so the fix is to open the article and copy the names from the byline.
Two smaller groups round this out. In 2.9 percent of records at least one author had no first name at all, usually a full name deposited in the wrong field, and 1.9 percent listed an organisation, such as a research consortium, as an author. ORCID identifiers, which tie a name to one specific researcher, appeared in 14.4 percent of records overall but in 31.9 percent of those published since 2015.
Titles: capitals, code and split subtitles
6.2 percent of titles were deposited entirely in capital letters. No citation style keeps them that way, and no generator can reliably guess which words are proper nouns, so these have to be retyped: sentence case for APA, where only the first word, the first word after a colon and proper nouns take a capital, and title case for MLA, Chicago and Turabian.
Even a normal looking title can need work. Most publishers deposit titles in headline style, so an APA reference built straight from the record carries capitals APA does not want. Here is a real record from our sample:
Another 4.3 percent of titles contained formatting code: tags such as <i> for an italic species name or <sub> for a chemical formula. Those italics and subscripts are real and belong in your citation, but they need to be applied as formatting rather than pasted as text. And 5.3 percent of records kept the subtitle in a separate field; a tool that reads only the main title quietly drops everything after the colon.
Volume, issue and pages
The journal name was present in every record and the volume in 96.1 percent. The weaker fields were the issue, present in 87.1 percent, and the page range, present in 88.1 percent. Only 76.7 percent of records had all three of volume, issue and pages together.
Many of the gaps are not errors. 10 percent of records used an article number instead of a page range, which is normal for online journals that publish articles one at a time. APA handles this directly: you write the word Article followed by the number in the place where the pages would go. The records to watch are the 5.7 percent with neither pages nor an article number, which are often online first versions that will gain page numbers later. If you read the final version, cite the final version.
Dates
Every record had at least a year. 47 percent carried a full day, month and year, 39.9 percent a year and month, and 13.1 percent the year alone. For journal articles that is rarely a problem, because APA and Chicago need only the year for a journal article, MLA adds the month when the journal gives one, and Chicago allows a month or season.
Old records and new records go wrong in different ways
Splitting the sample by publication year shows that the problems shift rather than disappear. Older records are thin on names. Newer ones are thin on issues and pages, because online first publishing assigns a DOI before an issue exists.
| Measure | Before 2000 | 2000 to 2014 | 2015 to 2026 |
|---|---|---|---|
| Records in the sample | 299 | 249 | 452 |
| Initials only names | 39.1% | 17.3% | 6.4% |
| No author | 15.4% | 9.2% | 4.9% |
| Issue present | 94.6% | 92.4% | 79.2% |
| Pages or article number | 98.7% | 93.6% | 91.8% |
| Title in all capitals | 6.0% | 3.6% | 7.7% |
| At least one ORCID | 0% | 0% | 31.9% |
| Full day, month and year | 24.7% | 43.0% | 63.9% |
For a typical student reading list, which is mostly recent, that means the likeliest fixes are a missing issue or a missing page range, while anyone working with older literature should expect to type out first names more often than not.
How many authors a paper has, and why it matters
The number of authors decides how a citation is shortened, and it has grown. Across the whole sample the median paper had two authors, but 46 percent had three or more, rising to 56.9 percent for papers published since 2015 against 29.1 percent before 2000. The largest author list in the sample had 60 names.
| Authors on the paper | Share | What it triggers |
|---|---|---|
| No author listed | 9.1% | Follow your style's rule for an unsigned work |
| One author | 26.6% | Every style lists the name in full |
| Two authors | 18.3% | Both names, in every style, in text and in the list |
| Three to six authors | 35.9% | APA, MLA and Chicago shorten the in-text citation to et al. |
| Seven to twenty authors | 9.5% | IEEE, AMA, Vancouver and Chicago also shorten the reference list |
| Twenty one or more | 0.6% | APA lists the first 19, an ellipsis, then the last author |
Each style draws the line in a different place, and the lines differ between the text and the reference list. We set every rule out side by side, with this sample's numbers, in our guide to et al. in every citation style.
A five point check for any citation built from a DOI
The findings above reduce to five quick checks. They take about a minute per source with the article open beside the citation.
What this means for any citation tool
A generator that reads the publisher's record will not invent a journal, a year or a DOI, which is the failure that matters most and the one AI chatbots are known for. What it cannot do is add information the publisher never sent. FreeCitation reads the Crossref record and formats it with the open Citation Style Language files, as described on our methodology page, so the gaps measured here are the gaps to look for in its output too.
Download the data
Each row records the DOI, the publication year, the journal and publisher, the number of authors and a yes or no for every measure in this article. The underlying Crossref metadata is open, so the sample can be repeated with the method above.
Frequently asked questions
Where does the information in a DOI citation come from?
From Crossref, the registry that publishers deposit their records with when they register a DOI. Each record holds the authors, title, journal, volume, issue, pages and dates the publisher chose to send. A citation generator reads that record and arranges it in your style, so the citation can only be as complete as the record.
Why does my citation show initials instead of first names?
Because the publisher sent initials. In our sample, 18.9 percent of records gave every author as initials only, and the share was 39.1 percent for papers published before 2000. APA uses initials anyway, but MLA, Chicago and Turabian ask for first names as they appear in the source, so check the article itself and type them in.
Why is the article title in capital letters?
Some publishers deposit titles in all capitals, which happened in 6.2 percent of the records we checked. No style wants that. Retype the title in sentence case for APA, where only the first word, the first word after a colon and proper nouns are capitalised, or in title case for MLA and Chicago.
What should I do if a journal article has no page numbers?
Look for an article number, which many online journals use instead of pages. In APA you write the word Article followed by the number where the page range would go. In our sample 10 percent of records had an article number and 5.7 percent had neither pages nor an article number, which usually means an online first version, so check the publisher page.
Can I trust a citation generator?
Trust it for what the record says and check it for what the record leaves out. A generator that reads Crossref will not invent a journal or a year, but in our sample 16 percent of records lacked at least one element an APA reference needs, and about one in three had a capitalised title, embedded code or initials only names. A one minute look at the article fixes all of these.
Does a missing author mean the paper has no author?
Not always. In our sample 9.1 percent of journal article records listed no author, many of them editorials, notices and older items, but some are simply incomplete deposits. Open the article. If it really has no author, follow your style's rule for an unsigned work and move the title into the author position, and never invent a name.
Can I use this dataset?
Yes. The CSV with one row per DOI is free to download and reuse under the Creative Commons Attribution 4.0 licence. Please credit FreeCitation.org and link to this page. The underlying Crossref metadata is itself openly available, so anyone can repeat the sample with the method described above.
The bottom line
The data behind a DOI is reliable about what it contains and silent about what it leaves out. In a random sample of 1,000 Crossref journal article records, 84 percent held everything an APA reference needs, 18.9 percent gave authors as initials only, 9.1 percent listed no author, 6.2 percent had capitalised titles and 5.7 percent had neither pages nor an article number. A generated citation is a strong first draft. Keep the article open beside it for one minute and check the author, the names, the title, the locator and the issue.