03 · What You Need to Know
Why AI Can Produce Correct Titles With Incorrect Authors or Publication Dates
A Real Article Title Does Not Guarantee Accurate Bibliographic Metadata
Academic references contain several pieces of information that collectively identify a publication. A correct title is useful, but it does not automatically validate the remaining fields.
Generative language models may reproduce a familiar article title while generating associated details from learned patterns or incomplete contextual information. The resulting reference may include an incorrect author, an omitted coauthor, or a publication year associated with another record.
These errors are particularly easy to overlook because researchers often treat a successful title search as confirmation of the entire citation.
Walters and Wilder (2023) examined 636 citations generated by GPT-3.5 and GPT-4 across 42 multidisciplinary topics. Among citations corresponding to real publications, 43% of GPT-3.5 references and 24% of GPT-4 references contained at least one substantive bibliographic error.
Their analysis included incorrect author names, publication dates, article titles, journal titles, and other bibliographic details. Errors in numerical fields, including dates and volume or page information, were especially prominent. The researchers also reported that errors involving authors and titles were comparatively uncommon in their sample. The original study provides the detailed results.
These findings establish that correct source identification and correct metadata generation are not equivalent. They do not establish a universal error rate for current AI systems, particularly those using different models or retrieval mechanisms.
How Can AI Attribute an Article to the Wrong Author?
Author errors can take several forms. An AI-generated citation may omit a contributor, substitute another researcher's name, reverse author order, or attach the correct initials to the wrong surname.
Some errors may reflect confusion among researchers who publish on similar topics. Others may result from inaccurate reproduction of bibliographic information.
For example, an article written by two researchers might be attributed to only the first author. Alternatively, the AI might combine the correct title with an author who published a different article in the same research area.
These are not merely stylistic differences. Authorship identifies the individuals or organizations credited with producing a scholarly work.
Crossref recommends complete and accurate contributor metadata, including names, roles, and identifiers such as ORCID iDs where available. Its documentation explains that contributor information supports citation matching, discoverability, and appropriate attribution.
Why Does Author Order Matter?
In many disciplines, author order carries meaning. First authorship may indicate a leading contribution, while last authorship can carry particular significance in some research fields. Other disciplines use alphabetical ordering or different conventions.
Consequently, changing the author sequence can misrepresent how a publication attributes credit, even when all the correct names appear somewhere in the reference.
Author order also affects citation formatting. In author-date systems, the first author's surname commonly determines how a work is cited and alphabetized. An incorrect first author may therefore produce incorrect in-text citations and reference-list placement.
Crossref's contributor documentation distinguishes the first contributor from additional contributors and notes that systems reusing deposited metadata commonly preserve contributor order.
Researchers should reproduce the author sequence from the authoritative publication record rather than allowing AI to rearrange names.
Why Can a Publication Have More Than One Legitimate Date?
Not every disagreement about a publication year is an AI error. Scholarly articles may pass through several publication stages, each associated with a different date.
For example, an article might become publicly available online in December 2023 and subsequently appear in a journal issue dated February 2024. Both dates may be legitimate descriptions of different publication events.
The National Library of Medicine distinguishes electronic publication dates, print publication dates, and ahead-of-print dates in its bibliographic records. Its PubMed documentation also explains that the interval between online-first publication and issue assignment can extend beyond a year.
| Date Type |
What It Describes |
Why It May Differ |
| Online-first publication date |
When an article first becomes available online before or independently of issue assignment. |
The article may appear online before receiving final issue details. |
| Issue or print publication date |
The date associated with a particular journal issue or print publication. |
The issue may be assigned later than the online release. |
| Preprint posting date |
When a manuscript version is posted on a preprint server. |
The preprint may precede a subsequently published journal article. |
| Correction or update date |
When a publication receives a subsequent update or correction. |
It does not necessarily replace the original publication year. |
The appropriate year depends on the publication version, the authoritative bibliographic record, and the citation style being followed.
Researchers should not automatically replace an earlier year with a later one merely because a newer date appears on a database page.
How Do You Distinguish a Wrong Year From a Legitimate Publication-Date Difference?
Start by identifying the publication version you intend to cite. Then compare the publisher's recommended citation with the dates displayed in the article record.
If the publisher identifies the version of record as published in 2023, a 2024 date supplied by AI may be incorrect even if another related event occurred in 2024.
However, if the publisher provides separate online and print publication dates, the apparent disagreement may require interpretation under the relevant citation rules.
Crossref recommends that publishers supply relevant publication dates, including both online and print dates where applicable. Its metadata guidance also cautions against retaining only the most recent publication date because doing so can misrepresent the publication history.
When the dates remain ambiguous, consult the publisher's citation recommendation, the article's version-of-record information, and the relevant style manual.
Can an AI-Generated Citation Have the Correct Title and DOI but Still Be Wrong?
Yes. A DOI can correctly identify the publication while the surrounding citation contains inaccurate metadata.
Suppose the AI provides a genuine DOI and article title but assigns the wrong first author. Resolving the DOI confirms which publication is being referenced, but it does not make the AI-generated author information correct.
This illustrates why a genuine DOI does not automatically validate an entire citation.
The researcher should compare all relevant fields with the original record rather than stopping once the DOI resolves.
Can Bibliographic Databases Contain Conflicting Metadata?
Yes. Different services may display records differently, and some metadata may be incomplete or incorrect.
For example, one database may display abbreviated author initials while the publisher lists full names. Another may use an online publication date while a different service emphasizes the issue publication date.
These differences are not necessarily evidence that either record is fabricated.
Crossref acknowledges that bibliographic metadata require ongoing maintenance and correction. Its guidance emphasizes accurate contributor information and publication dates because errors can affect citation matching and discoverability.
When discrepancies arise, the publisher's version-of-record information is usually an appropriate starting point. Relevant indexing records can then help explain differences or identify possible metadata errors.
What Role Can ORCID Play in Verifying Author Identity?
ORCID provides persistent identifiers intended to distinguish researchers, including those with similar names or changing institutional affiliations.
An ORCID iD can help establish whether a publication is associated with a particular researcher, especially when the identifier appears in publisher-supplied metadata or an authenticated record.
However, the presence of an ORCID iD does not independently prove every authorship claim. ORCID records may contain information added through different mechanisms, and completeness varies.
Use ORCID as supporting evidence alongside the publisher's author list and authoritative bibliographic records.
Why Are These Errors More Than Formatting Problems?
Incorrect authors and years can affect attribution, source retrieval, and the integrity of a manuscript's references.
An incorrect first author can cause an in-text citation to identify the wrong researcher. An incorrect publication year can make a source appear more recent or older than it actually is, potentially distorting the chronology of a research field.
These problems differ from punctuation, capitalization, or other citation-style issues. They concern the underlying identity of the publication.
That distinction matters when evaluating whether AI can format academic references reliably. A perfectly formatted reference can still contain incorrect metadata.
04 · A Practical Example
A Real Article With an Incorrect AI-Generated Publication Year
Hypothetical AI Output Using a Real Publication
Checking the Authors and Year of a Study on ChatGPT Citations
Suppose an AI assistant provides this reference:
Walters, W. H., & Wilder, E. I. (2024). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, 14045. https://doi.org/10.1038/s41598-023-41032-5
The article title, authors, journal, volume, article number, and DOI identify a genuine publication. However, the publication year supplied in this hypothetical citation is incorrect.
Locate the article.
Search the exact title or resolve the DOI. The publisher's website identifies the corresponding article in Scientific Reports.
Verify the authors.
The publisher lists William H. Walters and Esther Isabelle Wilder, in that order. The AI-generated author names and sequence match.
Check the publication year.
The publisher identifies September 7, 2023, as the publication date and lists the article in volume 13, article number 14045.
Correct the reference.
Replace the incorrect year, 2024, with 2023. Retain the verified author information and DOI.
The corrected reference is:
Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5
The publisher's record confirms these details, and the corresponding PubMed record independently identifies the same authors, year, and DOI.
This example demonstrates that a reference may identify the correct publication while containing an incorrect year. The appropriate response is to correct the metadata, not to classify the genuine article as fabricated.
