AI Detection: Can Your University Tell If You Used AI?

AI detection is one of the most pressing issues in academic writing right now: more and more universities check submitted work not just for plagiarism, but for AI-generated content as well. Whether it is an essay, a bachelor’s dissertation or a seminar paper, if you use tools like ChatGPT you should know how AI-written text is spotted and what the consequences can be.

In this guide we explain how AI detection works, how reliable the methods are, and what happens if your university finds AI-generated text in your work. We also show you how to check your own text before you submit it. Mimir Mentor is an AI-powered platform for academic writing with a built-in AI detector that combines AI detection and plagiarism checking in a single step.

What is AI detection?

AI detection is a process that analyses whether a text was generated, in whole or in part, by an artificial intelligence such as ChatGPT, Gemini or Claude. It examines the statistical language patterns that are typical of machine-generated writing.

In academia the topic has become increasingly important. Since ChatGPT became publicly available at the end of 2022, universities have had to work out how to handle AI-generated text in assessed work. Many institutions have now developed their own policies setting out clearly what is permitted and what counts as academic misconduct.

For you as a student, that means one thing: watching out for plagiarism is no longer enough. Your use of AI tools has to be transparent too, and you should know whether and how your text can be checked for AI content.

How does an AI detector work?

An AI detector analyses the linguistic features of your text and estimates how likely it is that individual passages were machine-generated. Several complementary techniques are used to do this.

Statistical language analysis

AI language models produce text by predicting the statistically most likely next word. That creates patterns which differ from human writing: more uniform sentence lengths, fewer unexpected word choices and a strikingly consistent structure. AI detectors measure exactly these patterns, in particular perplexity (how predictable a text is) and burstiness (how much sentence length varies).

Trained classification models

Many AI detectors also draw on machine learning models trained on large volumes of human and AI-generated writing. These models learn to pick up subtle differences in style, word choice and sentence construction that are often invisible to the human eye.

No AI detector is perfect. Outputs vary by provider: some show scores or labels, others show the proportion flagged. None is proof of authorship. That is true of every tool on the market.

Check your document for AI content and plagiarism now

Upload your work as a PDF or DOCX. The free overview shows you the number of potential AI hits and plagiarised passages.

Which methods do AI detectors use, and how reliable are they?

AI detectors combine several approaches, because no single method identifies every AI-written text on its own, and even combined methods make mistakes. Here are the most important techniques at a glance:

Mimir Mentor takes a combined approach: the built-in AI detector analyses the linguistic patterns in your text and gives a model-based indicator for each flagged passage. On top of that, the plagiarism check flags unsupported claims, which are especially common in AI-written text because language models rarely cite genuine sources. For a detailed explanation of every method, see our article on how AI detection works.

Can Turnitin detect ChatGPT?

Turnitin can flag passages that its model considers likely to be AI-generated, but the result is not proof. Turnitin itself says its AI writing detection may misidentify human-written, AI-generated and AI-paraphrased text and should not be used as the sole basis for adverse actions against a student (Source: Turnitin Guides). According to Turnitin’s documentation, the AI writing report is currently available for English, Spanish, Japanese and Arabic texts, and whether it is used at all depends on your institution.

Detectors can also be biased. A Stanford study found that several widely used GPT detectors consistently misclassified essays by non-native English writers as AI-generated (Source: Liang et al. (2023)). If you write in a second language, keep your drafts and notes so that you can show how your text developed.

AI detection at university: what happens if AI text is found?

If assessed work contains AI-generated text that was not permitted or not disclosed, universities usually treat it as academic misconduct. The consequences range from a fail for the piece of work through to expulsion in serious or repeated cases.

A decision in Germany shows how seriously this is taken. The Administrative Court of Munich (case no. M 3 E 23.4371, November 2023) ruled on an applicant who had submitted an essay written with ChatGPT as an assessed piece in a master’s admissions procedure. The court upheld the rejection of his application and classed the use of AI as deception (Source: Administrative Court of Munich).

Examiners can raise a suspicion of AI use even without technical proof. The University of Hohenheim in Germany states plainly that technical proof of AI use is currently not possible beyond doubt and that the decision ultimately rests with the examiners (Source: University of Hohenheim). Goethe University Frankfurt stresses that AI detection software can only ever support a judgement and never stand alone as evidence (Source: Goethe University Frankfurt).

Even so, more and more institutions are drawing up concrete policies. In the UK, the Russell Group of research-intensive universities agreed principles on generative AI in July 2023, including a commitment to ensure that academic rigour and integrity is upheld (Source: Russell Group). In Germany, the University of Cologne treats undisclosed AI use as a breach of the declaration of original authorship (Source: University of Cologne). Bielefeld University addresses AI use as unauthorised assistance when it breaches the applicable assessment rules, even though that assistance is technical rather than human (Source: Bielefeld University). The exact wording differs from institution to institution, so always check your own university’s rules. For more on what those rules typically say, read our article on AI-written text at university: what is allowed and what is not.

How to check your own text for AI content

You can check your text for AI content in a few steps and run the plagiarism check at the same time. Here is how it works with Mimir Mentor:

1. Upload your document: upload your PDF or DOCX to mimir-mentor.com. The free document scan starts automatically.
2. Review the overview: you get a summary showing the number of potential AI hits, plagiarised passages and correction suggestions.
3. Analyse the AI report: the premium report shows you passage by passage where the AI detector sees patterns associated with generated text. Every flag shows the passage concerned and how it has been classified.
4. Review the flagged passages: check each one for accuracy, sources and your own contribution. If you used AI there, make sure the use was permitted and is disclosed; rewording does not replace disclosure.
5. Check your style: then use the text correction tool to check grammar, academic style and consistency. That way your final version reads clearly and consistently.

Check your text early, not the night before the deadline. That gives you enough time to rework the flagged passages properly.

Upload your document and check it before you submit

Understanding the report: what do the AI detection results mean?

The AI detection report gives you a model-based indicator for each flagged passage, showing where AI-typical patterns were found and how pronounced they are. A passage-level value is not the proven AI-written share of your whole document. Reading those results correctly matters.

What do the flags mean?

Mimir flags passages that show linguistic patterns typical of AI-generated text. A flag does not automatically mean you wrote that section with ChatGPT. It means the section is built in the way AI-written text often is. Specialist writing with highly standardised language can trigger hits too.

How should you work through the results?

Go through the flagged passages systematically. For each hit, ask yourself: did I write this section myself? If you did, and it reads with that AI-like evenness, it often helps to add your own examples, perspectives or a more personal line of argument. If you used AI assistance, verify the content and follow your course’s permission and disclosure rules. Rewriting does not remove those obligations.

Alongside the AI score, Mimir also shows unsupported claims: places where a citation is missing. That is especially common in AI-generated text, because language models can omit or invent references. With the built-in literature search you can find suitable sources from over 270 million publications without leaving the platform.

Before and after: AI text versus your own writing

Polish the wording of your own draft

AI detection and plagiarism checking: why they belong together

AI detection and plagiarism checking complement each other because they uncover different problems. A plagiarism check finds matches with existing sources. AI detection flags passages that may have been machine-generated, even when they appear in no database at all.

In practice the two overlap: AI-generated text often contains no citations, which the plagiarism check flags as unsupported claims. At the same time, an AI-written passage can contain phrasing that happens to match existing publications. Mimir Mentor checks your text for both in one step: AI content and plagiarism at the same time. You do not need a second tool.

For a detailed explanation of how Mimir Mentor’s plagiarism check works, our separate guide covers every detection method in full.

Tips: how to avoid trouble with AI-written text

You can use AI tools sensibly without landing yourself in difficulty. What matters is transparency and a considered approach to what the tools give you.

• Disclose your use of AI if your university requires it. Check the applicable AI rules at your university and record what you did in an AI use declaration.
• Use AI for brainstorming and structuring, but write the content out in your own words.
• Add your own examples, perspectives and arguments to show your own academic contribution.
• Support borrowed claims with citations and your own findings with evidence. Common knowledge does not usually need a citation.
• Check your text before you submit it with Mimir Mentor’s AI detection and review any flagged passages.

• Submitting AI-generated text unchanged as your own work.
• Swapping out individual words and assuming the text will slip through undetected.
• Relying on a single tool and believing that “not detected” is the same as “safe”.
• Staying quiet about AI use when your institution’s examination or academic misconduct regulations require you to be transparent.

Check and improve your work with Mimir Mentor’s AI detection.

And if you are worried about your data: Mimir Mentor is GDPR-compliant, with servers in the European Union. Your document is processed for the analysis and is deleted in full after 30 days at the latest. Your text is never added to a public database and is not used to check other users’ documents. More than 70,000 students already use Mimir Mentor for plagiarism checking, AI detection and text correction.

If you want to move on from the diagnosis to the rules and the paperwork, read on in AI-written text at university: what is allowed and what is not, how to disclose AI use and how AI detection works.

FAQ: common questions about AI detection