Can ChatGPT Generate Fake Citations? (And How to Check Them)

Yes, ChatGPT can generate fake citations.

If you have ever copied references from ChatGPT, Claude, or Gemini into a draft without checking them, you are taking a real risk. The danger is not that fake citations look obviously broken. The danger is that they often look completely believable: realistic author names, plausible paper titles, familiar journal names, and even DOI strings that look correct at first glance.

That is why fake citations are so dangerous in academic writing. A student, researcher, editor, or reviewer can look at the bibliography and think everything is fine until they try to open the paper and discover that it does not exist.

If you use AI for brainstorming, outlining, summarizing, or drafting, you should assume every AI-generated citation is unverified until you check it yourself.

The 10-Second Answer

If you want the shortest possible answer:

• yes, ChatGPT can generate fake citations
• yes, Claude and Gemini can do the same
• no, you should not trust AI-generated references without verification
• the safest next step is to verify citations manually or use a citation checker before submission

What the Evidence Shows

This is not just a theoretical concern.

A 2023 Scientific Reports study examined bibliographic citations generated by ChatGPT and found both fabricated references and metadata errors. A later cross-disciplinary 2024 study also evaluated the reliability of LLM-generated citations and references in scholarly writing. University library guidance has been warning about the same issue: the USC Libraries generative AI research guide explicitly notes that LLMs can hallucinate fictitious citations and publications.

In other words, "AI-generated fake references" is no longer an anecdotal complaint. It is a documented reliability problem in academic workflows.

Why ChatGPT Can Generate Fake Citations

ChatGPT is a language model. Its main job is to generate plausible text based on patterns, not to verify every bibliographic detail against a live academic database.

That means when you ask for references, the model may generate something that looks like a citation because it has learned the shape of citations:

• author names
• publication years
• journal titles
• article titles
• volume and issue numbers
• DOI-like strings

But "looks like a citation" is not the same as "is a real published source."

This is why users often encounter three different failure modes:

1. Fully fabricated citations

Nothing in the citation is real. The title, author combination, or journal reference does not exist in the scholarly record.

2. Partially correct citations

Some elements are real, but the full citation is wrong. For example:

• the journal is real, but the article title is invented
• the paper exists, but the year is wrong
• the title is close, but the author list is different

3. Distorted real papers

The model may be "remembering" a real paper imperfectly. You end up with a citation that points in roughly the right direction but is still not accurate enough to use in a bibliography.

Why Fake Citations Matter

Many users treat citation errors as a minor formatting issue. They are not.

If you include fake or unsupported references in academic work, you create at least four risks:

Academic credibility risk

If a professor, editor, or reviewer checks a reference and cannot find it, confidence in the rest of your work drops immediately.

Integrity risk

Even if the fake citation was generated accidentally by AI, the submitted paper still contains a false reference under your name.

Workflow risk

One fake citation can create a chain reaction:

• reviewers ask for clarification
• revisions slow down
• collaborators lose trust in the draft
• you waste time re-checking the full bibliography under deadline pressure

Research quality risk

If the source does not exist, then the claim it was supposed to support may also be unsupported, misleading, or simply wrong.

What Fake Citations Usually Look Like

Fake citations are often not random nonsense. They usually have one or more of these traits:

• a polished, academic-sounding article title
• a real-sounding journal name that is slightly off
• author names that look plausible but do not match a real paper
• a DOI that follows the right pattern but does not resolve
• a combination of real metadata pulled from multiple different papers

That last one is especially dangerous. A fake citation may combine:

• a real topic
• a real author surname
• a real journal
• and a non-existent article

At a glance, it looks legitimate.

How to Verify Citations From ChatGPT, Claude, or Gemini

Here is a practical workflow that works for students, researchers, and academic writers.

Method 1: Search the full title in Google Scholar

Put the exact title in quotation marks and search it in Google Scholar.

If no matching paper appears, the citation is likely fake or significantly distorted.

This is the fastest first-pass check when you are working with just a few references.

Method 2: Verify the DOI

If the citation includes a DOI, check whether it resolves correctly.

Three common outcomes:

• it resolves to the exact paper: good sign
• it resolves to a different paper: the citation is wrong
• it does not resolve at all: the DOI is fake or malformed

Method 3: Compare metadata field by field

Even when you find a similar paper, do not stop there. Compare:

• title
• author list
• year
• journal
• DOI

If two or more core fields do not match, treat the citation as unreliable.

Method 4: Use a citation verification tool

Manual checking works, but it gets slow fast. If you have 10, 20, or 50 references, it becomes tedious and easy to miss errors.

Paste the full reference list into Citely's Citation Checker.

Citely helps verify references against academic sources and highlights:

• title mismatches
• author mismatches
• year mismatches
• suspicious or incomplete entries

This is especially useful when you are checking references generated from ChatGPT, Claude, or Gemini in batches rather than one by one.

What to Do If a Citation Is Fake

If a citation fails verification, do not just delete it and move on. You still need to decide whether the underlying claim is valid.

You have three main options:

Option 1: Find the real supporting source

Sometimes the claim is valid, but the AI invented the reference.

In that case, take the claim itself and use Citely's Source Finder to locate a real paper that supports it.

This works well when you have:

• a sentence in a draft
• a claim from an AI-generated paragraph
• a statistic with no reliable source attached

Option 2: Rewrite the claim

If you cannot find a strong source for the exact claim, rewrite it more cautiously.

Instead of overstating certainty, align the sentence with what the actual literature supports.

Option 3: Remove the claim

If the citation is fake and no solid source supports the point, remove the statement. That is better than keeping unsupported material in your paper.

Manual Checking vs Using a Citation Checker

Approach	Best for	Main downside
Google Scholar title search	1-3 citations	Slow for larger lists
DOI checking	Citations with DOI present	Misses title/author distortions
Metadata comparison	Careful manual review	Time-consuming and repetitive
Citely Citation Checker	Full reference lists, AI-generated bibliographies	Still requires human judgment on flagged items

When You Should Be Extra Careful

You should be especially cautious with AI-generated citations in these situations:

• literature reviews
• dissertations and theses
• grant proposals
• journal submissions
• medical or policy-related writing
• collaborative drafts where references came from multiple people

The more important the document, the less acceptable it is to assume the bibliography is "probably fine."

A Simple Rule to Follow

Use AI to speed up ideation if it helps you.

Do not use AI-generated references as final references unless they have been verified.

That one habit will prevent most fake-citation problems before they reach your professor, editor, reviewer, or client.

Key Takeaways

• Yes, ChatGPT can generate fake citations because it generates plausible text rather than validating every reference against a live academic database.
• Fake citations are often partially believable, which makes them more dangerous than obviously wrong references.
• The fastest manual checks are title search, DOI verification, and metadata comparison.
• For longer reference lists, a citation checker is more efficient and less error-prone than checking each reference one by one.
• If a citation is fake, either replace it with a real supporting source, rewrite the claim, or remove the claim entirely.

👉 Verify your references here: citely.ai/citation-checker