Why does ChatGPT make up fake citations in the first place?

I get this question almost every week from my dissertation students across the UK, US and Middle East. The honest answer is that ChatGPT and similar AI writing tools are not connected to any research database when you ask them to write your literature review. They are trained to predict what the next word should look like, not to check whether a paper actually exists. So when you ask for a citation, the tool builds one that looks correct, real author names, a real sounding journal, a proper looking DOI, but the actual paper was never published. A peer-reviewed study published in Nature’s Scientific Reports tested this directly on 636 citations across 42 topics and found that 55% of the citations generated by GPT-3.5, and 18% generated by GPT-4, were completely fabricated. This is not a small glitch. It is built into how these models work

How common are AI-generated fake references in academic work?

In my 12 years of mentoring research scholars, I have watched this problem grow every year since ChatGPT came out. Different studies show different numbers depending on the subject area, but the pattern stays worrying. One study on stem cell research found around 15% of ChatGPT’s references were fabricated, while other studies across fields report fabrication rates as high as 69% depending on the model and topic. Even lawyers are not safe from this. Courts around the world have sanctioned attorneys in more than 200 documented cases for submitting fake, AI-generated case citations in official filings. If trained legal professionals are getting caught, imagine how easily a dissertation with 60 to 80 references can have several fake ones slip through, especially if AI was used to speed up the literature review.

How to check if a citation is real before you submit

The good news is, verifying a citation does not take much time once you know what to look for. First, search the exact DOI on doi.org. If it does not resolve to a real paper, that is your first red flag. Second, search the paper title directly on Google Scholar or Crossref. A real paper will show up with the same authors, year and journal. Third, check whether the journal name and the volume or issue numbers actually match the publication year, since AI models often mix up these details even when the citation looks convincing otherwise. Doing this manually for every reference in a 70 to 80 citation dissertation can take hours, which is exactly why I built a free tool to automate the whole process.

What happens if your university finds fake citations in your dissertation?

This is the part that worries students the most, and rightly so. Universities across the UK, US and Middle East treat fabricated references as a serious academic integrity issue, sometimes even more serious than plagiarism, because it shows an attempt to mislead the examiner about your actual research process. I have personally helped students prepare for viva voce examinations where the panel specifically questioned two or three citations that turned out to be AI-hallucinated. In the worst cases, this can mean a failed viva, a mandatory resubmission, or in serious cases, disciplinary action from the university. The safest approach is always to verify every single citation before you submit, not after your supervisor or examiner points it out.

How our free AI citation checker helps you verify references in seconds

This is exactly the problem I built the citation verification tool above to solve. Instead of manually checking each reference one by one, you simply paste your entire reference list, up to 25 citations at a time, and the tool checks every single one against real scholarly records from Crossref and OpenAlex. Within seconds you will see which citations are verified, which ones need a closer look, and which ones could not be matched to any real publication at all. I built this specifically for dissertation and thesis students who use AI tools to speed up their writing but still want to submit work with genuine, verifiable references. Scroll up and try it on your own reference list before you submit.