
Does AI Invent Sources? How to Verify a Citation
AI chatbots can fabricate an author, title and page number. Learn how to spot a hallucination and verify whether a citation really exists before you submit.
More and more students ask ChatGPT for sources for their thesis and get a tidy list of books with authors, titles and page numbers. The catch is that some of those entries were never published. This is called a hallucination, and it can sink a defense.
Why AI chatbots make up sources
A language model has no access to a library or an academic database. It predicts the next words from patterns in its training data, so when you ask for a citation it generates something that looks like a real reference: a familiar name, a plausible title, a reasonable year. The form is right, but the content is often invented.
The most common types of hallucination:
- Non-existent publication — the author is real, but never wrote that book or article.
- Fabricated page number — the work exists, but the quote and page are matched on the fly.
- Broken DOI — the identifier has the right format but leads nowhere, or to a completely different paper.
- Mash-up — one publication's title stitched to another's author and year.
Important: this is not malice. The model simply cannot tell a fact from a probable string of characters.
Why page numbers are especially risky
Footnotes and reference lists usually require a page number, and that is where hallucinations are most insidious. A model can correctly name a real book yet attach a page on which the quoted sentence does not appear at all. It looks convincing precisely because the publication is genuine, so fewer people check further. A reviewer who knows the edition, however, will open the cited page and immediately notice the mismatch. Page numbers have to be read from the actual document — the model has no access to the page content, so it simply invents it.
How to spot a fabricated citation manually
Before you paste a reference into your thesis, verify it. Here is a quick path:
- Check the DOI — paste the identifier into
doi.org. If it does not resolve to a specific article, the citation is suspect. - Search Crossref or OpenAlex — these are metadata databases of millions of publications. No match on title and author is a red flag.
- Look in Google Scholar and Google Books — if the entry appears nowhere except the chatbot's answer, it probably does not exist.
- Verify the page number — open the PDF or scan and confirm the quote actually appears on the page cited.
Illustrative example (FABRICATED, deliberately untrue): "Smith, J. (2018). Methodology of Digital Research, p. 142." It looks clean, but if no database knows it, that is a sign it came from a generator, not a publisher.
Why supervisors catch it
A reviewer has read in their field for years and knows the key works. When the bibliography lists a title no one has ever cited, or a page where a given sentence simply is not, it triggers immediate suspicion. More universities now spot-check references, and a single fabricated citation undermines the credibility of the whole thesis — even if the rest of the sources are solid. What is at stake is not just the grade, but an accusation of academic dishonesty.
How cytado checks your whole bibliography in seconds
Manually verifying dozens of references takes hours. That is why we built a source checker that does it automatically. You paste your bibliography, and the tool:
- queries Crossref, OpenAlex and Google Books to confirm each entry really exists;
- compares author, title and year against the database metadata;
- flags references that cannot be found anywhere as needing review.
Unlike a chatbot, cytado does not fill in missing data from imagination. If a source is in no database, you get a clear signal instead of nicely formatted fiction. And when you look for new material, cytado fetches real PDFs and reads page numbers straight from the document.
Summary
AI chatbots can speed up research, but they are not a source of bibliographic truth — they invent entries, pages and identifiers in ways that are hard to catch by eye. Before you submit, verify every reference in Crossref, OpenAlex and Google Books, or do it in one move. Check your bibliography in cytado and be sure every citation really exists.
Frequently asked
- Does ChatGPT really invent academic sources?
- Yes. A language model predicts plausible text, so it can generate an author, title, year and page number that never existed. The formatting looks credible, but the data must be checked against databases.
- What is the fastest way to check if a citation exists?
- Check the DOI at doi.org and search the title in Crossref, OpenAlex and Google Scholar. No match in any database is a sign the entry may be fabricated.
- Can a supervisor tell a source is an AI hallucination?
- Often yes. Reviewers know the key works in their field, and a quick check of a DOI or page number quickly exposes a non-existent entry and undermines the whole thesis.
Read next
- How to Spot AI-Fabricated Sources in Student WorkA practical guide for supervisors: how to recognize AI-hallucinated citations, verify them by hand, and why 30 theses demand batch verification.
- How to Use AI for Research Safely (Without Fake Sources)One rule: the model may search, but it must never retype your metadata. A five-step workflow grounded in measured citation fabrication rates.
- Can Professors Tell Your Sources Came From AI? (And How They Catch It)Markers do not detect an AI writing style — they check whether your references exist. Here are the measured hallucination rates and the five signals that give a fabricated bibliography away.