How to check your dissertation for AI text: a simple guide

At German universities, a rough consensus has now taken shape: generative AI may be used as a tool, for example for research, for feedback or to polish your phrasing, as long as the work stays your own and you are open about your use of AI. Having AI write the whole thing for you, by contrast, counts as deception, much like ghostwriting. Many universities have published their own guidelines on this, for example the orientation framework from the University of Hamburg or the guidance note from the University of Tübingen. What is ultimately binding, though, is always your university’s examination regulations.

That is exactly why you should know which parts of your work look like AI. There are two reasons for this. First, pure AI text stands out to markers and AI detectors alike. Second, and many people miss this: AI text can even turn into plagiarism. We will explain why further down. We will show you how to spot AI text, how detection works, how reliable it is and how to check your term paper or dissertation properly before you hand it in.

 

📌 In short:

An AI detector checks your text for patterns that are typical of ChatGPT and the like, and flags passages that were probably written by an AI. It gives you a probability, not proof. So false alarms are possible. For you, that means checking it yourself before you submit, revising any suspicious passages and, if in doubt, being ready to show how you worked.

Fancy a quick check yourself? Simple and with no sign-up, using the GoThesis AI detector: check your work for AI text.

What is an AI detector and what can it do?

An AI detector is a tool that analyses your text to work out whether it was written by an artificial intelligence such as ChatGPT. Rather than comparing it against sources, the way a plagiarism check does, it looks for linguistic patterns that are typical of machine-generated text and estimates how likely it is that individual passages are AI-generated.

The emphasis on likelihood matters. An AI detector proves nothing; it works with statistics. That is why good tools show you not just a percentage but, passage by passage, how suspicious each one is. The GoThesis AI detector, for example, flags text in three levels: possibly, probably or very probably AI-generated. It picks up content from the common models such as ChatGPT, Gemini and Copilot.

 

How can you spot AI text?

Even without a tool, there are tell-tale signs that let you and your examiners spot AI text:

• Too smooth and uniform. AI writes in grammatically clean prose, but often in a strangely average way, with little change of rhythm and hardly any rough edges.
• Stock phrases and filler. Formulations such as ‘In today’s world…’, ‘It is important to note that…’ or ‘In conclusion, it can be said’ start to pile up.
• Surface-level statements. Plenty that is broadly true, but little depth, no examples of its own and no real argument.
• Missing or invented sources. AI likes to ‘hallucinate’ sources that do not exist at all, or leaves references out altogether.
• No personal voice. The text sounds interchangeable, as if no one in particular had written it.

People usually spot AI by gut feeling, detectors by statistics. The combination of the two is the most reliable.

 

How does AI detection work?

AI detectors take advantage of how language models write in the first place. An AI always chooses the statistically most likely next word, one word at a time. This makes its texts very predictable and even. Technically, this is called low perplexity, meaning the text is not very surprising, and low burstiness, meaning the sentences barely vary in length and structure.

People write more irregularly: a short sentence here, then a long, nested one, with an unusual word thrown in between. It is exactly this irregularity that pure AI text often lacks. The detector measures these patterns and works out the probability from them. Because these are only patterns, it can get things wrong too. More on that in a moment.

How reliable are AI detectors?

Honestly? Good, but not flawless. There are two types of error you should know about:

• False positives: a text you wrote yourself is flagged as AI. This happens especially with a very factual, smooth writing style, or if you are not a native speaker and therefore phrase things more evenly.
• False negatives: an AI text slips through, for example when it has been heavily rewritten.

So the rule is: an AI detector is a warning sign, not a verdict. Use it to find and revise suspicious passages, not as a free pass and not as proof one way or the other. That is exactly why reputable universities must never rely on a detector’s result alone.

 

AI text and plagiarism: how AI can turn into plagiarism

Many people think AI text and plagiarism are two completely separate topics. In fact, they are more closely linked than you would think, and that is the real reason you should keep an eye on both:

• AI takes phrasing from its training data, sometimes close to the original wording of existing texts. If you use that unchecked, it can turn into classic plagiarism, even though you never copied anything yourself.
• AI invents sources. It generates authors, titles or page numbers that do not exist. Made-up references like these come to light at the latest when someone double-checks them, and they count as a serious academic error.
• Passing off AI text as your own work is deception, and if it contains someone else’s content without a source, it is plagiarism on top of that.

Worth knowing: a plagiarism check and an AI detector check two different things and do not replace each other.

	Plagiarism check	AI detector
Looks for	text taken from existing sources	machine-generated passages
Question	‘Is this from somewhere else?’	‘Is this from an AI?’

 

Some university systems such as Turnitin have now built in AI detection as well, but you cannot rely on it. If you really want to be sure, you check your work for both, for example with the GoThesis plagiarism and AI check. You can read how the classic plagiarism check works in our article on the free plagiarism check and in our comparison of plagiarism software.

 

How do universities and lecturers spot AI text?

Examiners rarely go on a single tool. Most often, AI stands out through breaks in style. If your work suddenly sounds far smoother than your earlier work or your spoken style, supervisors start to prick up their ears. Add to that missing or incorrect sources, strikingly generic statements and a text with no recognisable voice of your own.

On top of that, some universities use AI detectors or Turnitin’s AI detection and, if in doubt, simply ask about it in the viva. Anyone who can explain and defend their own work has a clear advantage here. And that is the heart of the university consensus: AI as a help is fine, as long as the work and the understanding are yours.

Is AI allowed in your term paper or dissertation?

As mentioned above, this depends on your university and its particular examination regulations. In practice, though, most rules come down to the same thing:

• Allowed: AI is usually fine as support, so for research, ideas, feedback or improving your language.
• Required: many universities insist that you declare your use of AI, for example in a list of aids used or in your declaration of authorship.
• Forbidden: having the work, or substantial parts of it, generated entirely and passing it off as your own.

If in doubt: read the examination regulations and ask your supervisor. We show you how to use AI sensibly and within the rules in our article on AI tools for your studies.

 

Check your work for AI text: the GoThesis AI detector

If you want to know before you submit where your work might stand out as AI, it is best to check it with a tool that shows you every suspicious passage individually. That is exactly what the GoThesis AI detector does: you upload your work and can check your work for AI text, right in the interactive portal. There you can see which passages are considered AI-generated, flagged in three levels from ‘possibly’ to ‘very probably’. ChatGPT, Gemini and Copilot are all recognised.

 

The upload and the first look inside the portal are free, and you do not need an account. You only give an email address, which is where the full report is sent. Your document is deleted automatically, does not end up in any public database and is not passed on to universities.

Here is how it works:

1. Upload your document (PDF or Word, up to 15 MB).
2. Look at the free preview: in the portal you can see how your work is assessed.
3. Unlock the full report once you are convinced, from €6.90 for a term paper and €9.90 for bachelor’s and master’s dissertations.

 

The safest approach is to check both. The AI detector and the plagiarism check cover different risks and, as described above, AI text can even turn into plagiarism. With GoThesis you get both from a single portal: if you start with the AI detector, you can then add the plagiarism check straight away. And the other way round, you can add the AI check to the plagiarism check. That way you can also check your work for plagiarism, on the same principle: a free first check, with no account, with automatic deletion. You can read how GoThesis holds up in practice in our review.

 

Ready for the check?

Upload your work and, in just a few minutes, see which passages count as AI. Free and with no sign-up: check for AI text now.

 

What to do if your text is wrongly flagged as AI?

False positives do happen, especially with a factual writing style. If a detector or your university wrongly flags you as AI, one thing helps most of all: being able to show how you worked. In practice, that means:

• Document your writing process: drafts, notes, outlines and the version history, for example in Word or Google Docs, show that the text grew with you.
• Back up your research: a tidy collection of sources makes it clear that you really got to grips with the topic.
• Revise very smooth passages: your own examples, concrete references and your personal take lower the risk and make the work better.
• Seek out the conversation proactively: anyone who can openly explain how their work came about usually clears up an AI suspicion quickly.