Citation generators save time, and every published test of them has found errors. Since 2005, librarians and lecturers have tested web generators, reference managers such as Zotero and EndNote, phone apps and database cite buttons. No tool was right every time, and the same faults keep returning: capitals, author names, missing numbers and links.
EdCitation publishes this roundup. For each study we give the tools, the sample and the errors as the paper states them, and whether we read the full text or only the abstract (all on 2 October 2026). Our own test of today's free generators is the best free APA citation generators, tested; what each kind of tool is for is in citation generator, reference manager or reference checker. We also gave six DOIs from these studies to EdCitation's free Cite a source and quote what came back, faults included.
Are citation generators accurate?
Not reliably. Scheinfeld and Chung (2024) found none of 60 APA 7 references from the PubMed and Ovid MEDLINE cite tools fully correct, Ho (2022) found no free web generator error-free, and Kratochvíl (2017) found no reference manager that formatted every reference correctly for any of 15 medical journals.
A generated reference is a draft. Nor can the figures be lined up across studies: Scheinfeld et al. (2025) reviewed 32 studies of APA reference accuracy, 14 of them tests of tools, and found 64 different ways of measuring accuracy.
The studies, in one table
Oldest first. "Read" says how much of the paper we could open.
| Study | Tools tested | Sample | What it found | Read |
|---|---|---|---|---|
| Kessler and Van Ullen (2005) | EasyBib, NoodleBib, EndNote | Student papers, per the 2025 review | Web generators as accurate as EndNote; users still need to know the style | Abstract |
| Kessler and Van Ullen (2006) | Citation help in databases | Citation examples in database help | 94% of examples had errors, 4.3 per example | Abstract |
| Gilmour and Cobus-Kuo (2011) | CiteULike, Mendeley, RefWorks, Zotero | 14 articles from 7 databases, 5 styles | From 0.8 errors per citation (RefWorks) to 2.88 (CiteULike) | Full text |
| Van Ullen and Kessler (2012) | Citation help in databases, again | Not stated in the abstract | 3.4 errors per citation; retrieval statements worst | Abstract |
| Van Ullen and Kessler (2016) | Android and iOS citation apps | Print books, MLA and APA | Most apps inaccurate; EasyBib the best | Abstract |
| Kratochvíl (2017) | EndNote, Mendeley, RefWorks, Zotero | 1,084 references, 15 medical journals and NLM | From 575 errors (Zotero) to 937 (RefWorks) | Full text |
| Ho (2022) | ZoteroBib, CiteMaker, Cite This For Me | 18 APA 7 references from 6 student sources | 18, 31 and 40 errors | Full text |
| Scheinfeld and Chung (2024) | PubMed and Ovid MEDLINE cite tools | 60 APA 7 references | 0 of 60 fully correct | Full text |
| Scheinfeld et al. (2025) | Review | 32 studies, 14 of them of tools | 64 different accuracy measures | Full text |
Two more tests could not be opened, so no figure from them appears: Homol (2014), on EndNote Basic, RefWorks, Zotero and EBSCO Discovery Service, and Laing and James (2023), on the EBSCO and Summon discovery tools.
What did the tests of free web generators find?
Errors in every tool, mostly in formatting.
Kessler and Van Ullen (2005): the first comparison
Read in abstract only. Kessler and Van Ullen (2005) compared free or cheap web generators (EasyBib and NoodleBib, as the 2025 review lists them) with EndNote, judged them as accurate, and warned that using them well still takes some knowledge of the format.
Ho (2022): three generators, six real sources
Ho took six sources from first-year students' portfolios, made an APA 7 reference for each with ZoteroBib, CiteMaker and Cite This For Me, and checked all 18 against the APA Style site. ZoteroBib made 18 errors, CiteMaker 31 and Cite This For Me 40, formatting (capitals, punctuation, italics, indents) being the largest category for each. None set a hanging indent; all three left "Article" off an article number, mishandled an author shown on the article with the title Dr, and cited Malay names wrongly (Ho argues these belong in full, since Malay authors have no surname).
Van Ullen and Kessler (2016): phone apps
Read in abstract only. Van Ullen and Kessler (2016) generated MLA and APA references for print books with free or cheap Android and iOS apps. Most produced citations the authors judged unacceptably inaccurate, many offered few functions, and EasyBib was the best.
Is Zotero more accurate than EndNote?
In the one test we read that ran both, yes: Kratochvíl (2017) counted 575 errors from Zotero and 882 from EndNote over the same 271 references. That test ran in December 2015 and January 2016, so it shows a gap then, not a verdict now.
Kratochvíl (2017): four managers, sixteen medical styles
Kratochvíl typed 17 publications by hand into EndNote X7.4, Mendeley, RefWorks and Zotero, so no error came from imported data, and generated 1,084 references for 15 medical journals and the NLM style. Zotero made 575 errors, Mendeley 679, EndNote 882 and RefWorks 937; Zotero had the fewest for 11 journals and NLM. EndNote left out the URL in 120 of 135 references to online documents (Zotero in 34). The key finding is where errors start: one record came out right in one journal's style and wrong in another, so many faults lay in the style files.
Gilmour and Cobus-Kuo (2011): four managers, five styles
Gilmour and Cobus-Kuo (2011) imported 14 articles from seven databases into CiteULike, Mendeley, RefWorks and Zotero, left the data as imported, and produced lists in five styles. RefWorks averaged 0.8 errors per citation, Zotero 1.3 and CiteULike 2.88; Mendeley is 1.51 in their table and 1.08 in their text, and the table's figures give 1.51. In APA, 6 of 14 RefWorks references were error-free, 5 from Zotero, 3 from Mendeley, none from CiteULike. EndNote was not tested.
Are database cite buttons accurate?
No more than the other tools. Kessler and Van Ullen (2006) found errors in 94% of the citation examples databases offered, 4.3 per example; the follow-up, Van Ullen and Kessler (2012), found 3.4 per citation and gives the earlier figure as 4.4. Both read in abstract.
Scheinfeld and Chung (2024) scored 14 elements of PubMed's and Ovid MEDLINE's APA 7 references for the top 30 PubMed articles of 2010 to 2020. PubMed's 30 carried 81 errors (average 2.7) and Ovid's 171 (average 5.7). PubMed put 22 of 30 journal names in sentence case; Ovid lost italics in all 30 through its copy button and left out 23 issue numbers. Neither handled article numbers in APA's form. Punctuation was reliable.
Which errors do citation generators make most often?
Four kinds recur: capitals, author names, missing elements and the wrong type of source. Across their 32 studies, Scheinfeld et al. (2025) found authors the error category most often looked for (24 studies).
Capitals
Generators print a title as the record holds it. Gilmour and Cobus-Kuo (2011) found managers left titles as imported, right in sentence case only when the database supplied it; Kratochvíl (2017) found RefWorks lowering proper names in two journals' styles only. Scheinfeld and Chung (2024) quote the National Library of Medicine's Amanda Sawyer on the limit: "there is no way to systematically capitalize proper nouns".
Authors
Ho (2022) found "Dr" carried into references and Malay names cited wrongly; Kratochvíl (2017) found RefWorks cutting Klaus-Peter to K; and Gilmour and Cobus-Kuo (2011) found all four managers applied limits on listed authors, such as IEEE's three, inconsistently or not at all.
Missing numbers, DOIs and links
Kratochvíl (2017) found missing data the commonest fault in all four managers: URLs, access dates and editions, as well as the issue numbers and "Article" the database tools dropped. Retrieval statements were the worst part of database citations (Van Ullen and Kessler, 2012).
The wrong kind of source
A tool that makes you pick the type passes your mistake on. Ho (2022) notes that Cite This For Me offers no type for a conference paper published in a serial; entered as conference proceedings, it came out shaped like a book chapter. Students' own versions of these mistakes are in what studies say about student referencing mistakes.
What did Cite a source make of DOIs from these studies?
We gave EdCitation's Cite a source six DOIs from the samples above on 2 October 2026 and read the APA 7 output against each paper. Three came back right; three need a hand correction, each fault repeated on a second run.
| DOI, and where it came from | What Cite a source gave (APA 7, quoted) | To correct by hand |
|---|---|---|
| 10.1136/bmj.i2231, Scheinfeld and Chung's sample | … De Ponti, F., Bégaud, B., & Pariente, A. (2016). … BMJ, 353, Article i2231. | Nothing. Cite a source added "Article" and the accent, both missing from the entry the paper prints |
| 10.1177/2056305120912488, Ho's source 1 | … A survey study in Singapore. Social Media + Society, 6(2), Article 2056305120912488. | Nothing |
| 10.1080/02673843.2019.1590851, Ho's source 2 | Keles, B., McCrae, N., & Grealish, A. (2020). A systematic review: The influence of social media … International Journal of Adolescence and Youth, 25(1), 79–93. | Nothing: word for word Ho's hand-made reference |
| 10.25215/0304.193, Ho's source 3 | Kulkarni, N., & Patki, S. (2016). A study of emotional intelligence, … International Journal of Indian Psychology, 3(4). | Add pages 12–27 as Ho read them |
| 10.1371/journal.pone.0010502, Gilmour and Cobus-Kuo's sample | Knapp, S. (2010). Four New Vining Species of Solanum (Dulcamaroid Clade) from montane habitats in Tropical America. PLoS ONE, 5(5), Article e10502. | Sentence case: Four new vining species of Solanum (Dulcamaroid clade) from montane habitats in tropical America |
| 10.1063/1.4966060, Ho's source 5 | Ghazali, S., … Alias, R. A. (2016). The impact of knowledge sharing through social media among academia. AIP Conference Proceedings, 1782, 030003. | Put "Article" before the number |
Crossref holds the second author of 10.25215/0304.193 in one field as "Dr Sairaj Patki"; in CiteMaker, in Ho's test, Dr became an initial, and Cite a source now leaves the title out. The Knapp title is registered in title case, and the APA and Vancouver entries kept part of it. On the AIP paper, Harvard and Vancouver named the publisher "Author(s)". That paper was at least set as a periodical, and its Malay names are inverted as the record splits them; follow your style guide there.
Where is EdCitation the right tool, and what does it not do?
For anything with a DOI, a PubMed ID or an ISBN, Cite a source is the best place to start, and these studies show why: several of their faults started before any formatting, in a database record pasted as it stood (Gilmour and Cobus-Kuo) or details typed in by hand (Ho). Cite a source looks the work up in the publisher's record (Crossref, DataCite, PubMed and Open Library are named in the site's footer), builds it in APA 7, MLA 9, Chicago 18 author-date, Harvard, IEEE or Vancouver, and screens it for retraction, free with no account. In our run it put "Article" before both article numbers the studies' tools had left bare. Read it all the same; the APA page serves APA alone.
Verify references, also free, checks a finished list against the records: each entry comes back verified, doubtful ("check this") or not found, retracted papers are flagged, and an entry it could not check is shown as that, never as not found. For the style itself, Mechanics QA in Pro, at $8 a month, checks the reference list's formatting against the APA 7 student paper, and References from a file checks every reference in an uploaded paper and matches each citation to the list.
EdCitation never writes any part of anyone's work: it finds, cites and verifies sources from the record, and the reading and the writing stay yours. A shared library with a Word plug-in and thousands of styles is a reference manager's job, compared in EdCitation, Mendeley and EndNote.
How do I check a generated reference before I hand it in?
Check where the studies found errors, in this order.
- Start from the identifier. Give Cite a source the DOI, PubMed ID or ISBN rather than a web address, so the entry is built from the publisher's record.
- Read the authors against the article. Look for a title such as Dr turned into an initial, a name without a surname inverted, a hyphenated forename cut short, and your style's limit on listed authors.
- Fix the capitals. APA and Vancouver want sentence case in article titles, with names, places and organisations kept capitalised.
- Check the numbers. Volume, issue, pages or article number, with "Article" before an article number in APA and every leading zero kept.
- Check the DOI or URL. Present, in the https://doi.org/ form, with no library proxy address and, in APA, no full stop after it.
- Check the whole list. Run it through Verify references, then set the hanging indent none of Ho's three generators did.
Quick questions
Are citation generators accurate enough to use without checking?
No. Every test we read found errors from every tool. Treat the output as a draft and check it against the source.
Is Zotero more accurate than EndNote?
In Kratochvíl's 2017 test, yes: 575 errors against 882 over the same 271 references, tested in late 2015 and early 2016.
Does Citation Machine make errors?
No published test we could read covers it. In EdCitation's own run on 23 September 2026 it lower-cased a proper noun and left out an article number.
Are the cite buttons in library databases reliable?
Not on their own. Scheinfeld and Chung (2024) found PubMed averaged 2.7 errors per APA reference and Ovid MEDLINE 5.7, mostly capitals, italics and missing numbers.
Can I check a whole reference list for free?
Yes. EdCitation's Verify references compares every entry with the publisher's record, flags retractions, and needs no account.
References
- Gilmour, R., & Cobus-Kuo, L. (2011). Reference management software: A comparative analysis of four products. Issues in Science and Technology Librarianship, (66). https://doi.org/10.29173/istl1521
- Ho, C. C. (2022). Free online citation generators: Which should undergraduates use with confidence? Voice of Academia, 18(2), 70–92. https://ir.uitm.edu.my/id/eprint/65509/
- Homol, L. (2014). Web-based citation management tools: Comparing the accuracy of their electronic journal citations. The Journal of Academic Librarianship, 40(6), 552–557. https://doi.org/10.1016/j.acalib.2014.09.011
- Kessler, J., & Van Ullen, M. K. (2005). Citation generators: Generating bibliographies for the next generation. The Journal of Academic Librarianship, 31(4), 310–316. https://doi.org/10.1016/j.acalib.2005.04.012
- Kessler, J., & Van Ullen, M. K. (2006). Citation help in databases: Helpful or harmful? Public Services Quarterly, 2(1), 21–42. https://doi.org/10.1300/J295v02n01_03
- Kratochvíl, J. (2017). Comparison of the accuracy of bibliographical references generated for medical citation styles by EndNote, Mendeley, RefWorks and Zotero. The Journal of Academic Librarianship, 43(1), 57–66. https://doi.org/10.1016/j.acalib.2016.09.001
- Laing, K., & James, K. (2023). EBSCO and Summon discovery generator tools: How accurate are they? The Journal of Academic Librarianship, 49(1), Article 102587. https://doi.org/10.1016/j.acalib.2022.102587
- Scheinfeld, L., & Chung, S. (2024). MEDLINE citation tool accuracy: An analysis in two platforms. Journal of the Medical Library Association, 112(2), 133–139. https://doi.org/10.5195/jmla.2024.1718
- Scheinfeld, L., Chung, S., Fena, C., Tran, C., Kretz, C., & Reisman, M. R. (2025). Assessing formatting accuracy of APA Style references: A scoping review. Evidence Based Library and Information Practice, 20(4), 159–191. https://doi.org/10.18438/eblip30800
- Van Ullen, M., & Kessler, J. (2012). Citation help in databases: The more things change, the more they stay the same. Public Services Quarterly, 8(1), 40–56. https://doi.org/10.1080/15228959.2011.620403
- Van Ullen, M. K., & Kessler, J. (2016). Citation apps for mobile devices. Reference Services Review, 44(1), 48–60. https://doi.org/10.1108/RSR-09-2015-0041