Fabricated References: Check Citations Before You Submit

On 2 November 2026, Phys.org reported on a Norwegian statistician’s concern that software sold in China promises publishable research in under two hours, and that the output is flooding the literature. The same month, the problem of fabricated references keeps surfacing in journals: references that look authentic but point to papers that do not exist, or to real DOIs attached to the wrong article. For Indian scholars writing theses and papers, a reference list is now something to verify, not just format. This guide explains what recent cases show and how a citation check fits into a pre-submission routine.

What the recent cases show

The Phys.org report covers a study in the Journal of Clinical Epidemiology titled, in translation, “Paper mill or paper mine?”. Its authors examined more than 700 papers built on the Global Burden of Disease database. According to the article, most were by Chinese researchers, shared similar structures, and for 99% of them no source code was provided, so the results cannot be reproduced. The researchers stressed that they were not claiming the results were wrong; their concern was relevance and volume. The same report cites the journal Accountability in Research, which saw a 60% rise in submissions in 2026 (Phys.org, 2 November 2026).

Fabricated references have also led to retractions. Per a secondary report, Springer Nature retracted a 2024 review in Discover Artificial Intelligence on 3 October 2026. The publisher found that generative AI had been used without disclosure and that several citations were contextually incorrect. The report notes that the publisher said it “no longer has confidence in the reliability of the article’s contents” (BioEngineer report on the Springer Nature retraction).

An earlier case shows how the errors happen. A letter in Intensive Care Medicine carried 15 references. Retraction Watch could match only five, and reference 11 cited an article that did not exist in the journal. The authors told the publisher the fake references “resulted from the use of generative AI to convert the PubMed IDs of cited articles into a structured reference list.” The letter was retracted in November 2025 (Retraction Watch, 28 January 2026). The lesson is that a reference can be partly right, with a correct author and a plausible title, and still be wrong in ways that matter.

The pressure behind this is volume. Degen, one of the Phys.org study’s authors, told The Verge that the system is “already at the limit,” and Riebler notes that reviewers are already scarce. When editors and examiners cannot read every submission closely, a reference that is only partly right can pass unnoticed. That shifts the checking work to the author, who is the only person who knows the list was assembled and by what tools.

Why fabricated references look real

Most fabricated references are not invented from nothing. They are a mix of errors that a tired writer will not catch by eye:

• A real paper with the wrong year, pages or author order, which is what Retraction Watch found in several of the letter’s references.
• A real DOI attached to a different paper, often after an automated tool fills in identifiers from a list.
• A plausible title for an article that was never published, usually in a journal the author cites often.
• A retracted source that still circulates in drafts and reference managers.

None of these is caught by reading the bibliography aloud. They are caught by checking each entry against an authoritative record. Crossref is a useful example: its public REST API gives access to scholarly metadata for DOIs without sign-up, and a single DOI record can be compared with what your reference says (Crossref REST API documentation).

How a citation check works in InstaBizIntel

InstaBizIntel is a browser-based statistics, BI and research platform with an AI analyst. Its citation checker is part of the research-integrity toolkit. Per the product’s feature set, it:

• checks up to 150 references per run;
• verifies each reference against Crossref and OpenAlex;
• flags fabricated references, real DOIs that resolve to the wrong paper, and retractions.

A check of this kind confirms that a source exists and that its metadata matches what you wrote. It does not read the source to decide whether it supports the sentence it sits beside. That judgement, and the decision to cite a paper at all, remains with the author. The platform’s job is to remove the mechanical risk so that your time goes to the reading.

The same integrity toolkit includes statcheck-style recomputation of p-values, a GRIM test for impossible means, an overclaim linter, and an AI-use disclosure generator aligned with GAMER, UGC-India and Vancouver styles. These are useful when a journal or university asks how AI was used in drafting.

A pre-submission checklist for references

Run this routine before you send a thesis chapter, a paper or a report to a supervisor or journal:

• Export your reference list from your reference manager, so the check runs on the final entries, not an old draft.
• Run a citation check and read every flag. Treat a DOI that resolves to a different title as a serious error, not a typo.
• Confirm retractions before relying on a source, and replace it or note the retraction in your text.
• Read every cited sentence against its source. A real reference can still be misquoted.
• Keep a disclosure statement for any AI tool used for drafting, editing or literature search, naming the tool and the task.
• Share your data and code where your institution or journal asks for them. The Phys.org study found that most of the papers it examined lacked code, which makes checking impossible for editors and examiners.

Frequently Asked Questions

What is a fabricated reference?

A fabricated reference is a citation that cannot be matched to a real publication, or that is attached to the wrong one. It may combine a real author with an invented title, or carry a real DOI that belongs to another paper. Errors can come from careless writing, from AI tools that reformat citations, or from deliberate misconduct.

Can a citation checker tell whether a source supports my argument?

No. A check of existence, metadata and retraction status is different from a reading of the source. You still need to read each cited source and confirm that it supports the sentence it is attached to.

How much does InstaBizIntel cost?

The Premium plan is ₹800 per 30 days, inclusive of 18% GST, which is ₹677.97 plus GST. There is a 15-day full trial with no card required. The semester and annual passes are not yet available and are not offered for purchase.

Why check references if my reference manager already formats them?

Reference managers format citations; they do not confirm that the source exists or that its metadata is current. A formatted entry can still carry a wrong DOI, the wrong year or a retracted paper.

Sources

• Phys.org: Chinese software that produces ready-to-publish research papers in under two hours threatens peer-review system (2 November 2026)
• BioEngineer: Springer journal retracts AI review after undeclared generative AI use (October 2026)
• Retraction Watch: Medical journal publishes a paper on AI with a fake reference to itself (28 January 2026)
• Crossref: REST API documentation for retrieving metadata

To check your own reference list, visit the InstaBizIntel product page, or read more research-integrity guides on the InstaDataHelp blog.