AI language models write fluently, sound convincing and often add literature references. The problem: those sources do not have to exist. This guide explains why AI invents sources, how you recognise such references and how you check every reference in a few minutes before it appears in your term paper, thesis or presentation.
Short answer: AI models can invent sources that look real, with author names, title, year and DOI, but do not exist. You avoid this by checking every reference in your library catalogue, a subject database or with the publisher before you cite it. Whether and how you may use AI is set by your university. Last updated: October 2026.
The key points
- A language model produces text from likely word sequences. It does not automatically check whether a book or article exists.
- Invented sources often look credible precisely because some of their parts may be real.
- Every reference you take over from an AI answer must be found and read in the original by you.
- A missing hit in the catalogue is a warning sign, but not proof. Check several sources before you decide.
- You remain responsible for every reference in your work, even if an AI wrote the first draft.
Why does AI invent sources?
Language models learn from very large amounts of text which words typically belong together. When you ask for a source, the model gives an answer that sounds like a source. No database lookup takes place, unless the tool is explicitly connected to a literature database. That is why hybrid errors occur: a real author with an invented title, a real title with the wrong year, or a journal that exists but with a volume and issue that do not match.
Questions about niche topics, local studies or very recent publications are especially prone to this. The more specialised the topic, the more likely the model fills gaps with plausible-sounding details. Well-known standard works are less often affected, but even there a check is worthwhile.
How do you recognise invented sources?
The table lists typical warning signs. None of them proves anything on its own, but together they form a clear picture.
| Warning sign | What you do | Why it matters |
|---|---|---|
| The title appears in no catalogue | Search your university library catalogue and Google Scholar | A missing hit is a warning sign, not proof, since some titles are hard to find |
| The DOI leads nowhere or to another publication | Enter the DOI at doi.org and compare title and author | A DOI belongs to exactly one publication, so mismatches show up immediately |
| The author has published nothing on this topic | Check the author’s publication list in Google Scholar or with the publisher | Real names are often combined with invented works |
| Volume, issue and year do not fit the journal | Open the journal in the publisher archive | Wrong combinations only become visible when you open the issue |
| The quotation does not appear in the text | Search the original full text for the passage | Invented direct quotations are especially risky because they look like evidence |
How do you check a source in ten minutes?
- Enter the title in your university catalogue or Google Scholar and note whether a matching hit appears.
- If a DOI is given, check it at doi.org. Do title and author match?
- Open the original through the library or the publisher and find the page you refer to.
- Compare the content with your claim. Does the source really support it, or did the AI only supply a fitting keyword?
- Record the result in your reference manager and mark sources you could not find, so that you do not use them.
More on research with AI and clean source checks is in our guide Literature Research with AI: Tools and Clean Source Checks (in German).
Which rules apply at your university?
Universities set different rules. Some require you to disclose AI use, others exclude certain uses. The examination regulations, the guidelines of your department and the advice of your supervisor are decisive. A starting point is the recommendations of the University of Kassel on dealing with AI (in German) and the policy of the University of Freiburg on generative AI in research (in German). Because the rules differ, there is no blanket answer.
How can you use AI without inventing sources?
The safest role for AI is that of a discussion partner, not a reference list. You can ask it to suggest search terms, counter-arguments or technical vocabulary. You then fetch the actual sources from a database. If you need a bibliography, ask for search strategies instead of taking ready-made references. Learning with AI without copying is explained in Learning with AI: Prompts That Build Understanding, Not Copying (in German).
What happens if an invented source ends up in your work?
An invented source can be treated as deception or as a serious breach of due care. What consequences apply is set by your university’s examination regulations, and the assessment in each case lies with the examiners. It helps to document how you worked: notes, research lists and drafts show that you checked each source yourself. If you spot an error, tell your supervisor before you hand in the work.
Frequently asked questions
Can I simply cite AI output with a note saying “according to AI”?
That depends on your university. Many guidelines require you to disclose the use of the tool and do not treat AI output as a scientific source. Such a note does not replace checking the facts. Read your department’s rules before you use a citation of this kind.
How do I recognise whether a source is invented?
There is no reliable single sign. Suspicious are missing hits in the catalogue, quotations without page numbers, DOIs that lead nowhere and authors whose work does not fit the topic. Only the original full text proves that a source exists and supports your claim.
Do AI detectors help find invented sources?
No, they are not designed for that. Detectors assess the style of a text, not whether a source exists. A detector result is therefore not evidence for or against a reference.
Are sources from a literature database automatically correct?
Reputable databases contain real publications, but their metadata can also contain errors. Compare the details with the original, especially page numbers, year and volume. If you use an AI feature in a database, the duty to check stays exactly the same.
What to do now
- Mark every reference in your current draft that came from an AI answer.
- Check each marked source using the steps above and note the result.
- Remove sources you cannot find and replace them with verified literature.
- Clarify with your supervisor which statements about AI use your work requires.
What happens after your studies?
The duty to check sources does not end with your degree. In professional life and research, reports, expert opinions and presentations carry the names of the people who sign or present them. Faulty references are noticed by clients, colleagues or specialists and can damage your reputation. The ability to check sources yourself is therefore a basic skill that you train now.
Related guides
- AI Use in Studies: Rules, Disclosure and Tools
- AI in Term Papers: What Is Allowed and What You Must Disclose
- AI Rules at Universities: Why Every Examination Regulation Differs
- Literature Research with AI
- Citing Made Easy: The Key Rules at a Glance (in German)
Sources: University of Kassel, Artificial Intelligence in Research (accessed October 2026) · University of Freiburg, policy on generative AI in research (accessed October 2026) · DFG statements on the influence of AI models on science (German, accessed October 2026)
Last updated: October 2026. This guide is general information, not individual advice.
