Can Generative AI Combine Real Studies Into a Fake or Nonexistent Paper?

Can AI Build One Fake Paper Out of Pieces of Real Research?

Sometimes an AI-generated citation is obviously fabricated. No matching title exists, the journal cannot be found, and the supposed authors appear to have nothing to do with the topic.

The more deceptive case is assembled from reality.

The first author may be genuine. The article title may resemble a real publication. The journal may exist. The findings may sound familiar because they actually appear somewhere in the literature. Yet those pieces do not belong to one paper.

Generative AI can conflate information associated with different sources into a single apparently coherent publication. The result is difficult to detect precisely because searching its individual components keeps returning real research.

Can AI Merge Real Research Into a Nonexistent Source?

How Can Several Real Sources Become One Fake Paper?

Source Conflation Is Different From Pure Fabrication

A purely fabricated citation may contain information with no identifiable relationship to a real publication. Source conflation is more subtle.

Here, generated information resembles or incorporates elements associated with genuine scholarship, but the elements are combined incorrectly.

The distinction is useful for verification, although the boundary is not always clean. A hallucinated reference may contain both invented details and recognizable fragments of genuine publications.

Evidence Shows That Fabricated References Can Contain Real Scholarly Components

Empirical studies of AI-generated citations have repeatedly shown that fabricated references need not be fictional in every respect.

Walters and Wilder found that many fabricated citations generated in their experiment contained the names of real journals, publishers, or organizations. Chelli and colleagues similarly reported hallucinated references whose titles and author lists bore similarities to genuine papers.

More recent educational work examining citation errors across several contemporary LLMs likewise observed fabricated citations formed from incorrect combinations of real authors, titles, journals, volume or issue information, and page numbers.

These observations matter because they explain why a fake paper may survive superficial fact-checking. The model is not necessarily inventing every component independently. Real bibliographic patterns and entities can appear inside the hallucination.

Authors From One Paper Can Be Attached to the Title of Another

Imagine two real papers on the same topic.

Paper A is written by researchers Garcia and Santos and examines faculty attitudes toward generative AI. Paper B is written by Lee and Ahmed and examines institutional policies on generative AI.

A conflated citation might attribute a title resembling Paper B to Garcia and Santos.

Now consider what happens during verification. You search Garcia and find a real researcher publishing about generative AI. You search the title keywords and find a real article on institutional policy. Both searches produce reassuring results, but neither verifies the generated citation.

The correct question is whether those authors wrote that exact paper.

A Title Can Blend Concepts From Several Real Papers

Article titles are semantically rich. When several papers discuss overlapping concepts, generative AI can produce a title that resembles the literature without reproducing any one genuine title.

The generated title might combine the intervention from one study, population from another, outcome from a third, and methodological language common across the field.

This can create what feels like the ideal source for your question.

Findings From Different Papers Can Become One Imaginary Study

Source conflation is not limited to bibliographic metadata. Substantive findings can also be blended.

Suppose one study reports increased engagement, another reports no significant achievement effect, and a third identifies concerns about academic integrity. A generated account might describe a single study that measured all three outcomes and reported all three findings.

Every substantive statement may have a recognizable counterpart somewhere in the literature. The hallucination lies in representing them as findings from one publication.

This is particularly problematic in literature reviews because the generated synthesis may look efficient while erasing which evidence came from which source.

Methods and Results Can Be Reassigned Across Sources

Consider two studies of the same phenomenon. One uses a randomized experiment with 80 participants. Another surveys 600 respondents.

An AI-generated account might attach the randomized design to the larger sample, then report an outcome taken from the survey. The resulting paper never existed, but all three details have plausible origins.

This illustrates why verifying only the title or citation is insufficient when substantive details matter. You also need to determine whether the methodological details and findings belong to the same source.

A DOI From One Paper Can Make the Conflation Look Verified

One especially deceptive configuration occurs when a conflated citation contains a genuine DOI.

You resolve the DOI and arrive at a real article. That feels like decisive confirmation.

But the DOI may correspond to only one source involved in the conflation. The generated title, authors, or findings may come partly from elsewhere.

This is why verifying an AI-generated DOI requires comparing the resolved metadata with the entire claimed citation. Successful resolution alone does not establish that every surrounding detail belongs to that publication.

Similar Papers Make Conflation Harder to Detect

Research fields often contain clusters of closely related publications. The same team may publish several studies from one project. Titles may differ by only a few words. A longitudinal project may produce baseline, follow-up, and secondary analyses. A systematic review may discuss numerous papers with overlapping terminology.

These conditions make accidental conflation easier for both humans and AI systems.

When two real papers are extremely similar, a generated description may silently borrow details from one while citing the other. Unlike a wholly invented reference, searching for the source keeps producing plausible matches.

Researchers should therefore be particularly careful when authors have multiple publications on the same topic or when several papers emerge from the same dataset.

Conflation Can Happen Even When No Fake Citation Is Generated

Suppose AI correctly cites Paper A but summarizes it using findings partly drawn from Paper B. Bibliographically, Paper A is real. There is no fabricated citation.

The source representation is nevertheless conflated.

This overlaps with the problem of AI misrepresenting a real research paper. The useful distinction here is that the erroneous account may contain recognizable information associated with another genuine source rather than information invented entirely from scratch.

Source Conflation Can Create a Citation Trail That Looks Surprisingly Convincing

A fabricated source assembled from real components can be difficult to investigate because every search reveals something familiar.

The final step is the decisive one. Verification requires convergence, not familiarity.

Do Not “Repair” a Conflated Citation Without Identifying the Actual Sources

Once you suspect conflation, it may be tempting to choose whichever real paper most closely resembles the AI-generated reference and cite that instead.

That is risky.

The closest title may not contain the claimed finding. The author match may correspond to a different study. The DOI may identify a related paper whose methodology differs substantially.

Instead, decompose the generated claim. Identify which source supports each substantive statement, then cite those actual sources separately where appropriate.

Sometimes the result will be two or three genuine citations replacing one imaginary citation. That is not an inconvenience. It is a more accurate representation of the literature.

How Three Real Studies Can Become One Fake Study

The AI-generated account may have been topically excellent. Bibliographically, it was a chimera. Academic databases are less impressed by chimeras than mythology departments might be.

Common Mistakes When a Fake Paper Contains Real Information

Verify That All the Pieces Belong to the Same Paper

When an AI-generated citation contains several details that independently appear real, resist the temptation to declare victory. Instead, test whether those details converge on one publication.

The principle is provenance. Each claim should remain attached to the study that actually produced or argued it.

How to Check Whether AI Has Conflated Several Papers

Frequently Asked Questions About AI Conflating Research Papers

A Paper Can Be Fake Even When Its Pieces Are Real

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.