Gains:
- Ability to preserve originality not with a capture technology, but with a culture of honesty and process-based evaluation
- Ability to avoid unfair accusations by knowing that AI detectors are unreliable and cannot be evidence on their own
- Ability to implement transparent AI usage disclosure and a clear usage policy
The ability of AI to produce text touches the most fragile point of literature and language education: originality. For a teacher, "did the student really write this text?", for an editor, "is this content original or produced?", for a researcher, "is this my thought or the sentence of the machine?" questions are on every table now. In this unit, we will discuss originality, plagiarism (presenting someone else's work or idea as one's own work without citing the source) and the evaluation of texts written with AI.
A clear warning from the outset: There is no reliable tool that can "definitely" detect whether it is written in AI; So-called AI detectors are mistaken and can incriminate innocent texts. That's why the solution is not based on a "capture technology" but on a culture of integrity, process-based evaluation and clear rules.
What is originality and what is not?
Originality means that a text truly conveys one's own thought and expression. This does not mean that "no AI has been used". A writer can ask the AI for input, have the spelling corrected, get feedback, and the text can still be original and honest — as long as the contribution is transparent and the original thought and voice remain personal.
Where originality is damaged is this: AI produced the idea, structure and sentences of the text, and the person presented it as his own original work, hiding the contribution. The main issue here is not technology, but honesty: clearly saying what is being done.
Hint: “Have you used AI?” The question is the wrong question. The right question: "Where and how did you use AI and is the thought yours?" Establish a culture of open disclosure regarding the use of AI in your organization; Transparency instead of prohibition is more realistic and instructive.
Why can't AI detectors be trusted?
So-called AI detection tools predict whether a text is “machine-like” from statistical clues. But:
- They give false positives: They can mark a smooth and fluent text that is actually written by a human as "AI". This means falsely accusing an innocent student.
- They are easily circumvented: With minor changes, the text can pass the detector.
- They punish those who write in a different way from their native language: Those who write in simple, formulaic ways may be unfairly viewed as suspicious.
Therefore, a detector output is not evidence and cannot be the basis for an accusation, note or sanction on its own.
Caution: Just because a detector says "this text is 90% AI" is not enough to blame a student. Such output is at best a “conversation starter”; the actual assessment should be based on evidence of process (drafts, student ability to explain the text).
Process-based evaluation
The surest way to preserve authenticity is to evaluate the process, not the product:
- Ask for a draft. Notes, intermediate versions, corrections are traces of the original process.
- Ask for verbal explanation. The person who actually wrote the text can explain their choice: "Why did you choose that ending?"
- Use in-class/observed writing. Having part of the process done under observation makes it difficult to replicate the product.
- Ask for personal connection. Tasks that depend on the student's own experience, on a particular text he or she has read, make general AI production useless.
three mini cases
Case 1 — The detector would create unfair accusation. A teacher questioned an assignment where a detectorist said "88% AI". Before blaming the student, he asked for his drafts and verbal explanation; The student defended the text in detail and had hand-written notes. The text was original; It would be unfair if the detector was trusted.
Case 2 — Transparent use accepted. One student declared in his essay, "I had AI ask questions for the idea draft, I corrected my spelling, but I wrote the text and idea," and added process notes. The teacher considered this honest and authentic; It was not a punishment but an example of good practice.
Case 3 — Hidden production was caught. A student submitted a text that he had written entirely to the AI as his own work. In the verbal explanation, he could not explain a part of the text or the meaning of a word he chose. The problem was not the detector, but the student's unfamiliarity with his own text; The process evaluation clarified the situation.
Four copyable templates
1) Transparent AI usage statement (for student):
In this study, I used artificial intelligence in the following stages:- Idea/question: [yes/no, how]- Plan/skeleton: [yes/no]- Writing: [did I write the text]- Editing (spelling/flow): [yes/no]Stages I did not use: [...]The thought and expression of this text belong to me.
2) Originality self-control:
Read my text below and ask me (answer) the following questions: Can I explain every main idea in my own words, did I find every source, do I have my own thought behind every sentence? Point out weak links.Text: [paste text]
3) Process-based assignment design (for teacher):
Your role: assessment designer. Design a process- and personal-adaptive writing task that makes it difficult for the AI to print completely for the following outcome: [achievement]. Duty; Includes outline, oral defense, and personal experience.
4) Fair evaluation control:
I doubt the originality of a text. Give me a checklist for proceeding fairly, WITHOUT BLAMING THE STUDENT: what evidence of process should I ask for, what questions should I ask? Do not treat detector output as evidence.
Weak prompt / Strong prompt
Weak: "Tell me if an AI wrote this text."
Strong: "I know you cannot determine the originality of this text alone. Instead, list what process evidence (outline, notes) I would ask for and what content questions I might ask that someone who wrote the text could answer in order to have a fair conversation with the student. Don't make an incriminating judgment."
The powerful prompt does not impose a "detection" job on the AI that it cannot do; uses it in the design of a fair process. Poor prompt invites unreliable judgment.
Approach table
Question
unreliable road
solid road
Is this text original?
AI detector
Process + verbal explanation
Has AI been used?
trying to catch
Request transparent declaration
Should I grade?
Detector percentage
Acquisition + proof of process
Is impeachment necessary?
Don't rely on a single output
Interview, evidence, justice
Establish a clear AI usage policy
Uncertainty challenges both the student and the teacher. "I wonder if this was on leave?" While hesitation unduly worries the honest student, it also overshadows abuse. The solution is an upfront and clearly articulated usage policy: defining with clear rules where AI is allowed, where it is limited, and where it is prohibited in a task. A good policy is defined in stages, not in extremes such as "everything is forbidden" or "everything is allowed".
For example, in an essay task, the policy might be: having the AI ask questions at the idea stage is free and declared; Spelling-punctuation correction is free; It is forbidden to have the AI write the text itself; Each student adds a brief statement of use. Such a framework both ensures that the student knows what he or she can do and makes the assessment fair. Sharing the policy at the beginning of the lesson prevents “I didn't know that” discussions later.
The same principle applies to academic and editorial environments: journals, publishing houses, and institutions are increasingly requiring statements regarding the use of AI. Position these statements not as an admission of guilt, but as an honest and ordinary practice of transparency.
Tip: Create the policy by discussing it with students. A student who understands the rationale for the rule (learning and honesty) is much more willing to follow it. A rule that is shared rather than imposed is stronger.
Common mistakes
- Considering the detector output as evidence. False positives incriminate innocent people.
- Killing transparency by saying "AI is prohibited". Declaration culture is more realistic and instructive.
- Evaluating the product and ignoring the process. The process is the true testament to originality.
- Normalizing contribution concealment. Honesty is important whether it is used or not.
- Making the accusation before the conversation. Evidence and fair process first.
In summary
Authenticity is protected not by a capture technology, but by a culture of integrity and process-based evaluation. AI detectors are unreliable and cannot be evidence alone. Encourage transparent disclosure rather than banning the use of AI; Evaluate the process (outline, verbal description, personal connection) not the product. The real issue is not technology, but whether the thought and expression really belong to the person.
Application task
Reimagine a writing task as process-based: draft submission, a brief verbal description, and add an element of personal experience. Then attach the “Transparent AI usage statement” template to the task. Explain in one paragraph why this design makes it difficult to print entirely in AI.
checklist
- [ ] I did not use the AI detector output as evidence.
- [ ] I encouraged transparent AI use disclosure.
- [ ] I evaluated the process (outline, description), not the product.
- [ ] I added a personal connection and process element to the tasks.
- [ ] In case of doubt, I first sought fair discussion and evidence.