How to spot a fake AI citation before your professor does

You finish a paper at 1:42 a.m. ChatGPT gave you ten sources. They look fine: real journal names, plausible authors, APA formatting, DOIs included. You paste them into the bibliography, fix the spacing, and submit.

A few days later, your professor emails you:

Can you send me the PDF for source #4?

You search for it. Nothing comes up.

That is the problem with fake AI citations. They do not usually look fake. They look like normal academic references, right up until someone tries to find them.

Why fake AI citations are hard to spot

A bad website often looks bad. A fake AI citation usually does not.

AI systems are very good at producing the shape of a citation. Author names, journal titles, years, page ranges, DOI-like strings, all in the right order. The reference looks academic because the model has seen millions of academic references.

But looking like a citation is not the same as being one.

The source may never have existed. The DOI may go nowhere. The journal may be real, but the article is not. Sometimes the paper exists, but it does not say what the AI claims it says.

That last version is the hardest to catch, because the first check passes. You find a real paper and feel safe. But the citation is still wrong if the paper does not support the sentence you used it for.

The 30-second fake citation smell test

Before you open five new tabs, start with a quick smell test. These signs do not prove a citation is fake. They tell you which ones deserve checking first.

Page numbers that look too neat. Real journal articles land wherever the issue puts them. Page ranges like pp. 1-20 or pp. 100-120 are not impossible, but they are worth a second look. Real publishing is often messier.

A title that matches your essay too perfectly. AI-generated titles often sound like they were written to answer your prompt. They are clean, direct, and exactly on-topic. Real academic titles tend to have narrower wording, awkward subtitles, method details, or field-specific phrasing.

Author and journal combinations that feel too convenient. A real scholar may exist. A real journal may exist. That does not mean that scholar published that article in that journal. If the pairing feels too perfectly matched to your topic, verify it.

A DOI that looks correct but feels generic. Fake DOIs often follow the general pattern of real DOIs. The problem is that real publishers have their own quirks. A clean-looking DOI is not proof of anything.

If two or three of these show up in the same citation, do not use it until you verify it.

After the smell test: prove the source exists

The first check is simple.

Paste the DOI directly into doi.org. If the DOI is real, it should redirect you to a paper or publisher page. If it goes nowhere, or lands on something unrelated, treat the citation as suspect.

Then search the exact title in quotation marks on Google Scholar. A real paper should usually appear through the journal, a database, or a preprint page. If the exact title returns nothing, that is a strong warning sign.

But do not stop too early. A DOI that resolves only proves that something exists. It does not prove the AI cited the right title, the right authors, or the right claim.

For any source you actually rely on, open the paper and check the sentence around the claim. Ask one question:

Does this source actually say what I am using it to say?

Why students skip this step

Most students do not skip citation checks because they do not care. They skip them because the process is annoying.

Copy the DOI. Open a tab. Paste it. Go back. Search the title. Open Scholar. Go back again. Repeat.

At midnight, that feels like too much work. So the bibliography gets treated as formatting, not evidence.

That is exactly where fake AI citations slip through. They look polished enough to survive the moment when everyone is tired.

Use the same checks your professor will use

Professors do not need special software to catch many fake citations. They can use the same checks you can.

They may click the DOI. They may search the title. They may ask for the PDF. They may check whether the author has actually published in that field. They may ask what the source argues.

That last question matters most. If you cannot explain what a source says, it should not be in your paper.

A simple habit helps: save the PDF, article page, or database record for every source you cite. Even if your instructor does not require it, you will have proof that the source exists and a way to revisit what it actually says.

When you have a lot of sources

Checking five citations by hand is manageable. Checking twenty is where people start cutting corners.

This is where automation helps, but only for the first layer.

AI FactScan is a browser extension for Chrome and Edge that checks AI-generated source lists while you are still on the page inside ChatGPT, Claude, Gemini, or Perplexity. It can help flag dead links, weak source signals, and suspicious citation patterns faster than checking each one manually.

It does not replace reading the source. It does not prove that a paper supports your argument. It helps you catch the obvious failures before they end up in your bibliography.

The rule to remember

You do not need to become a librarian at 2 a.m. You just need to catch the failures that should never make it into a submitted paper.

A citation is not decoration. It is a claim that a source exists and supports what you wrote.

Before your professor checks that claim, check it yourself.