Let’s get this out of the way first, because it’s the question that brought you here: ChatGPT is not a scholarly source. You do not cite ChatGPT in a bachelor’s or master’s thesis to support a factual claim. You cite ChatGPT only when the AI output itself is your research object — in linguistic analyses, pedagogical studies, or content analyses of generated text. Anything else is academically indefensible, and at most German-speaking universities in 2026 it is impermissible without explicit disclosure.
This post explains why that is, what universities currently require, and — if your case really is an exception — how to cite ChatGPT correctly.
Why ChatGPT is not a scholarly source
A scholarly source satisfies three conditions: traceable authorship, quality assurance (typically peer review), and reproducibility. ChatGPT meets none of them.
- No author. OpenAI is the operator, but not the author in any substantive sense. The model generates statistically, not argumentatively.
- No peer review. Nobody checked the statement before it appeared on your screen.
- No reproducibility. The same prompt produces a different response tomorrow. Model versions are swapped without warning. Another person does not get the same text.
- Hallucination risk. Language models fabricate content and citations that sound plausible but do not exist. The topic is large enough to deserve its own post — see AI source hallucinations.
Anyone who cites ChatGPT as evidence for a factual claim (“according to ChatGPT, average life expectancy in Japan is …”) is essentially citing: a machine says that something probably sounds like this. That is not a scholarly argument.
What universities have required from 2024 to 2026
German-speaking universities — from ETH Zürich and the University of Zürich to LMU Munich, TU Munich, and HU Berlin — updated their guidelines between 2023 and 2025. The details vary; the pattern is consistent everywhere:
- Disclosure requirement. Any use of generative AI in a written academic work must be made transparent — usually in the declaration of independent work (Eigenständigkeitserklärung). Concealing it risks a finding of attempted deception, with consequences ranging up to the withdrawal of the degree.
- Prohibition on AI-generated content as your own work. You may not have a section written for you and pass it off as your own intellectual contribution. Not even if “reformulated”.
- Typically permitted: spell-checking and style editing, translation assistance, brainstorming, structural suggestions, code assistance in technical theses — provided this is disclosed in the declaration.
- Strictly prohibited: adopting references ChatGPT suggests without checking them against the original. This is exactly where most scandals originate.
The precise wording is in your programme’s examination regulations or your department’s guidelines. Read your institution’s document. A generic internet search does not substitute for it.
Citing ChatGPT as a research object — the exception
There are legitimate cases in which you cite ChatGPT: when you are studying the output itself. Examples:
- A linguistics bachelor’s thesis on gender bias in AI responses.
- A pedagogical study on how students evaluate ChatGPT-generated texts.
- A methods chapter documenting how ChatGPT solved a given task.
In these cases, the generated response is your primary material, analogous to an interview transcript. Which means: the material must be documented, dated, and verifiable.
What you must document
- The model and version used (e.g. GPT-5, as of March 2026).
- The exact prompt, entered precisely as shown.
- The date and time of the query.
- The full response — ideally in an appendix as a verbatim transcript. Without an appendix, your source cannot be verified, making it scientifically worthless.
APA 7 — reference list entry
OpenAI. (2026). ChatGPT (Version GPT-5, as of March 2026)
[Large language model]. https://chat.openai.comIn-text: (OpenAI, 2026).
Harvard (Author-Date) — reference list entry
OpenAI (2026) ChatGPT (Version GPT-5, as of March 2026)
[Large language model]. Available at:
https://chat.openai.com (Accessed: 12 March 2026).In-text: (OpenAI 2026).
Prompt documentation (example for your appendix)
Appendix A1 — AI Query
Model: GPT-5 (OpenAI, ChatGPT, as of 12 March 2026)
Date: 12 March 2026, 14:22 CET
Prompt: "Describe in three sentences how the concept of
'education' shifted in the 19th century from Humboldt
to Spencer."
Response: [complete response text, 4 paragraphs, see A1.1]Permitted — grey area — prohibited
| Category | Examples |
|---|---|
| Permitted with disclosure | Spell-checking and style editing, translation assistance, glossary checking, brainstorming for structure, having concepts explained for your own preparation |
| Grey area | Drafting argument outlines, having paragraphs reworded, generating summaries of texts, having analysis code written |
| Prohibited | Using AI-generated references without verification, passing entire sections off as your own writing, using ChatGPT to support factual claims, concealing AI use |
The grey area is exactly that: grey. What one department considers harmless writing assistance is a violation at the next. When in doubt, ask your supervisor before you start writing — and keep the answer in writing.
Template for the declaration of independent work in 2026
Many universities now add an AI paragraph to the standard declaration. A typical formulation:
I declare that I wrote this thesis independently. I used generative AI tools as follows: ChatGPT (GPT-5, OpenAI) for spell-checking and style editing in Chapter 3 and for translating English technical quotations in footnotes 14, 22, and 31. All substantive claims, arguments, and conclusions are my own. AI-generated source suggestions I have verified in full against the originals and included only after verification.
This is a template, not a legal statement. Check the exact wording required by your institution and follow it strictly — deviations regularly prompt follow-up questions.
The most common mistakes
- Adopting ChatGPT-generated reference lists without checking them. A classic: “Give me ten sources on topic X.” Seven are fabricated. Three exist but don’t say what you claim. Result: false citations in your thesis — see again AI source hallucinations.
- Including AI-generated text without disclosure. Even if no one notices because the style “sounds right”: if it is discovered (and detection tools are improving), it is deception.
- Using ChatGPT output as a factual reference. “According to ChatGPT” is not a citation. It is an open admission that you found no real source.
- Forgetting the model version and date. Anyone who cites ChatGPT legitimately but omits the model and date has an unverifiable source. That is poor scholarship, even if the citation looks formally correct.
Conclusion and final check
In 2026, ChatGPT is a tool for students, not an evidence medium. Used correctly, it saves time on proofreading and structuring. Used incorrectly, it produces false citations, puts your grade at risk, and in serious cases your degree.
The most important rule is straightforward: every source ChatGPT suggests must be checked against the original — before it enters your thesis. That is exactly what Acurio is built for. You upload your thesis and the source PDFs; we check every single citation against the actual passage in the source and report back: supported, partially supported, not supported. What ChatGPT promises, we verify. One hour of effort can, in the worst case, save your degree.