Main categories
Style and structure
Text that is too symmetrical and well-organised to be spontaneous.
Vocabulary
Formal and elevated, but generic. No personal voice or errors.
Content
Generalities without depth. Vague or made-up references.
Student context
Abrupt changes compared to previous work or known level.
Specific signals — ordered by reliability
- Abrupt change compared to previous work. If a student who normally writes with errors suddenly submits a flawless academic text, this discontinuity is the most reliable signal. high signal
- Zero spelling, typographical or agreement errors. Every student makes some. A 100% flawless text is statistically unusual. high signal
- Overuse of em dashes — AI uses them to mimic a reflective pause. In academic writing in English they are unusual in excess; their abundance is a digital fingerprint. high signal
- "It's not X, it's Y" structure. Clever but hollow contrasts: "It's not about working harder, but working smarter." AI overuses this device. high signal
- Grandiose opening sentences. "In a world where…", "When it comes to…" — they sound solemn but say nothing. high signal
- Text that is too polished and generic. Uses "the situation" instead of "the crisis", empty adjectives like "interesting", "relevant". medium signal
- Very uniform paragraph length. All exactly 4–5 lines, mathematical symmetry. Humans vary their rhythm. medium signal
- Constant hedging. "it may be that", "in some ways", "generally" non-stop. AI avoids categorical statements. medium signal
- No authorial voice. Absence of first person, irony, personal anecdotes. Uniform tone. medium signal
- A conclusion that doesn't conclude. Literal summary, no new perspective. Phrases like "it depends on the context". medium signal
- Literally translated technical terms. "learnings", "performance", "leverage" used as if natural. medium signal
- Excessive use of icons, bold text, boxes and visual emphasis. AI tends to overload the text to make it "more attractive" or "professional", something a real student rarely does with such deliberateness. medium signal
- Vague or hallucinated references. Citations with real authors but incorrect titles or years, or invented studies. complementary
- Contradictions between distant sections. AI has no real memory of the argumentative thread. complementary
- Mismatch between the student's level and the result. The text reads like an expert wrote it, but the student normally does not: they make mistakes, write simpler sentences... complementary
- Describes the surface well but avoids value judgements typical of someone who masters the subject. Lots of general data, but no opinion, no criticism, no confident comparison. complementary
Phrases and words AI repeats frequently
In a world where…It is essential to highlightIt is important to noteIn the current landscapeNowadaysIt is worth mentioning thatIn this sense…When it comes to…
crucialessentialrelevanton the other handin summaryin conclusionundoubtedlyfurthermorelikewise
Human writing vs. AI-generated text
✓ Human writing
- Occasional typing or style errors
- Variable emotions and tone
- Personal anecdotes and context
- Personal and debatable opinions
- Minor stylistic inconsistencies
- Humour, irony, shifts in register
- Acknowledges its own limitations
✗ AI-generated text
- Technically flawless
- Uniform and controlled tone
- No personal context
- Neutral and "safe" writing
- Consistently formal style
- No real humour or irony
- Generalises without concrete details
How to address / verify it — practical steps
1
Ask for specific sources
A human can cite the book, page or exact context. AI gives generic or invented references.
2
Ask in person
"Explain this paragraph to me in your own words." If they hesitate or cannot develop it, that is the definitive signal.
3
Review drafts
Ask for version history or staged submissions. A text that is "born" finished has no visible process.
4
Analyse depth
AI describes the surface well but avoids value judgements and productive contradictions.
Practical rule: detector + evidence + conversation. This combination reduces false positives far more than relying on an automated tool alone. No single signal is definitive — context always matters.
In the classroom — prevention
- Explain from the outset which uses of AI are permitted and which are not. Ambiguity creates avoidable misunderstandings.
- Assess part of the process, not just the final product: a short oral defence or staged submissions makes fraudulent use much harder.