Citing AI output is a question many students face once a term paper or thesis is written with the help of a language model. Citing, disclosing and evidencing are three different things. This guide explains what university guidelines usually require and how you stay transparent without inventing details.
Short answer: You cite AI output according to your university’s rules, usually with the tool, version, date and the prompt you used. It is rarely suitable as evidence for facts, because AI output can be wrong. Many universities require you to disclose the use of AI. Last updated: October 2026.
The most important points
- Citing and disclosing are two things you should consider separately.
- AI output is not a primary source for facts.
- Document the tool, version, date and prompt so that the statement can be checked.
- Clarify with your lecturers or supervisors which format applies.
- An appendix with a prompt log is one form of transparency that many works use.
Citing or disclosing: what is the difference?
A citation refers to someone else’s statement that you take over. A disclosure describes that you used an aid, and how, for example for your outline, language correction or research. Many universities require disclosure even when no passage has been copied word for word.
The distinction matters because the two obligations follow different rules. The basics of citation are covered in the guide Citing made easy: the most important rules at a glance (in German). Your guidelines settle whether AI output must be cited or only disclosed.
One example makes the difference clear. If you let an AI suggest your outline and adopt it almost unchanged, that is a disclosure matter, but not a citation. If you take over a sentence word for word, the duty to cite applies as well. Both can occur in the same piece of work.
Which details belong in the reference?
The table below lists the details that have proven useful in practice. Whether you need all of them depends on your university.
| Detail | Why it matters | Example |
|---|---|---|
| Tool and provider | Clear identification of the aid | ChatGPT, OpenAI |
| Version or model name | Results can differ between versions | Model name as shown in the tool |
| Date of the query | AI output changes over time | Example: 26 October 2026 |
| Prompt | Traceability of the input | Full wording in an appendix |
| Type of use | Shows what the AI actually contributed | Outline, translation, correction |
This list is a guide. Your university’s guidelines may require a different order, wording or additional details. When in doubt, use the stricter version.
How to document your use of AI
- Save the chat history or export it, if the tool allows it.
- For each use, note the date, tool, version and purpose in a table.
- Keep the prompt in its exact wording if your university requires it.
- Phrase the disclosure in your work so that a third party can follow how the AI was used.
Why AI is not evidence for facts
An AI answer is not a source in the scholarly sense. It can state things correctly, but it can also contain false information, and it can name sources that do not exist. If you take a fact from an AI answer, you must check it against a primary source and cite that source. The AI answer itself then serves at most as a hint for your search.
The copyright position is also not settled in every question. Whether and how output from generative systems is protected depends on the individual case. For the academic question of citation, it matters more that you make the use transparent and check the statement. The guidelines of the German Research Foundation (DFG) on good scientific practice describe how research deals with aids.
What role does honesty play?
Good scientific practice requires you to present your work truthfully. Concealed use of AI can be treated as deception, even if you think the text is good. Honesty here does not mean describing every bit of help in detail, but making the essential use traceable.
A rule of thumb helps: imagine your supervisor reads the disclosure and asks about the use. If you can answer without discomfort, the statement is probably sufficient.
Frequently asked questions
Do I have to cite every use of AI?
Not every use is a citation. A spell checker is disclosed differently from a paragraph you have adopted. Check your university’s rules on which uses must be stated.
How do I cite a sentence from an AI answer word for word?
Give the wording in quotation marks, state the tool, version and date, and attach the prompt in an appendix. If you use the content as a fact, you also need a reference to the primary source.
What happens if I do not disclose my use?
This can be treated as deception. The exam regulations set out the consequences. If you are unsure, ask your lecturers or the examination office early, before a problem arises.
Does this also apply to translations made with AI?
Generally yes. Translations or rewritings made with AI must be disclosed if your university requires it. Check the rules and document the use the same way you would document other aids.
Is a note in the introduction enough?
A general sentence in the introduction is rarely enough if a university requires specific details. A note at the place of use or an appendix is more precise and easier to check.
Do I need a declaration of independent work?
Many universities require one for written work. It confirms that you wrote the work yourself and stated all aids. Phrase the note on AI use so that it matches what you actually did. More on this is in the guide AI rules at German universities: why every exam regulation differs.
How to proceed now
- Check your university’s guidelines for requirements on disclosing AI.
- Start keeping a usage log during your work from today.
- Clarify the required format with your lecturers before you submit.
- Add a clear note on AI use to your declaration of independent work, if you used AI.
Related guides
- AI tools for students 2026: which offers exist for students
- AI rules at German universities: why every exam regulation differs
- Literature research with AI: tools and clean source checks
- AI at university: what is allowed and how to use tools properly (in German)
- Academic research: finding and evaluating sources (in German)
Sources: German Copyright Act (UrhG), § 51 (quotations), last updated: October 2026; German Research Foundation, guidelines on good scientific practice, last updated: October 2026; Regulation (EU) 2024/1689 (AI Act), last updated: October 2026
Last updated: October 2026. This guide is general information, not individual advice.
