Teacher Resources · AI Integrity
Detecting AI Work
& Designing Around It
How to recognize the signals of AI-generated work and the kinds of activities that are naturally resistant to it. Detection is hard and never certain; the most durable strategy is designing assignments that make genuine learning the only path through.
Start Here
No method, and especially no automated detector, proves AI use with certainty. Treat every signal below as a reason to look closer and start a supportive conversation, never as a verdict. Assume good faith first, and build your case from human, contextual evidence.
01 · In the Writing Itself
The most reliable signal is not a single phrase, but rather a mismatch between the work and the student you know. A student who writes in short, plain sentences in class suddenly submits flawless, evenly-paced prose with sophisticated transitions and no personality.
What to Watch For
- →Polished, uniform sentence rhythm that does not sound like the student talking.
- →Vocabulary noticeably above the level the student uses in discussion or in-class writing.
- →An absence of the small quirks, opinions, and rough edges that make student writing recognizable.
A Word of Caution
Voice shifts are evidence to start a conversation, not proof. Strong students improve, and second-language learners often write more formally than they speak.
AI-generated text tends to be broad, balanced, and risk-averse. It summarizes competently but rarely commits to a specific argument, names a particular source, or includes a concrete detail that only someone who did the work would know.
What to Watch For
- →Essays that "discuss" a topic from all angles but never take a clear position.
- →No reference to class discussions, the specific edition of a text, or material covered in your room.
- →Formulaic structure: an introduction that restates the prompt, three even body paragraphs, a conclusion that restates the introduction.
A Word of Caution
Generic writing is also a hallmark of inexperienced writers. Pair this signal with others before drawing conclusions.
Language models invent quotations, citations, statistics, and historical "facts" that sound authoritative but do not exist. Checking sources is one of the most concrete ways to surface AI use.
What to Watch For
- →Quotations that do not appear anywhere in the assigned text.
- →Citations to articles, page numbers, or authors you cannot verify.
- →Plausible-sounding dates, figures, or events that turn out to be wrong when checked.
A Word of Caution
A fabricated citation is strong evidence, but you should verify it yourself before raising the issue, since students also misremember real sources.
Students who copy and paste hastily often leave behind fragments of the chatbot exchange. These are among the clearest indicators and require no special tools to spot.
What to Watch For
- →Phrases like "As an AI language model," "Certainly! Here is," or "I hope this helps."
- →A prompt accidentally pasted at the top, or a response that addresses "you" as the user.
- →Inconsistent formatting, stray markdown symbols, or hyperlinks copied from a chat interface.
A Word of Caution
These are rare but nearly conclusive. Save the document exactly as submitted before discussing it.
02 · In the Process
Authentic work leaves a trail: messy notes, abandoned ideas, revisions, and version history. Work produced in one sitting by a chatbot has none of this. Building process checkpoints into assignments makes this gap visible.
What to Watch For
- →A final draft with no corresponding notes, outline, or earlier draft.
- →Google Docs version history showing the entire document pasted in a single edit.
- →A polished product that appears with no evidence of the intermediate steps you assigned.
A Word of Caution
Some students genuinely work on personal devices or paper. Set the expectation that process artifacts are part of the grade so the trail exists by design.
The single most powerful detection method is a brief, low-stakes conversation. A student who wrote the work can explain their choices, defend their argument, and say what they would change. A student who outsourced it cannot.
What to Watch For
- →Unable to define a sophisticated word that appears in their own essay.
- →Cannot summarize their thesis or walk you through how they reached a conclusion.
- →Surprised by the content of their own submission when asked about a specific paragraph.
A Word of Caution
Keep these chats warm and curious, not interrogative. Nerves can mimic guilt, so it is better to ask open questions and give the student room to think.
A complex assignment that would take most students hours arrives fully formed within minutes of being assigned, or far exceeds what the student has ever produced before with no intermediate growth.
What to Watch For
- →Submission timestamps far earlier than peers on a demanding task.
- →A leap in quality with no corresponding change in class participation or in-class writing.
- →Mastery of a concept in writing that the student cannot demonstrate verbally.
A Word of Caution
Growth is real and worth celebrating. Use timing as a prompt to look closer, never as a verdict on its own.
03 · About Detection Tools
Automated "AI detectors" (Turnitin AI, GPTZero, and similar) produce both false positives and false negatives at rates too high to justify an academic-integrity accusation. They have been shown to disproportionately flag writing by non-native English speakers, and they can be defeated by light paraphrasing.
What to Watch For
- →Treat any detector score as a single, weak data point rather than as definitive proof.
- →Never base a grade penalty or integrity referral on a detector percentage alone.
- →Be especially skeptical of high scores on writing by multilingual or neurodivergent students.
A Word of Caution
Your professional judgment, knowledge of the student, and a conversation are far more reliable than any tool. Check your school or district policy before using detectors at all.
Because no tool is conclusive, the strongest cases are built from human, contextual evidence, which is the kind that holds up in a conversation with a student, a parent, or an administrator.
What to Watch For
- →A documented inability to explain or reproduce the work in person.
- →Concrete artifacts: pasted AI text, fabricated sources you have verified, missing version history.
- →A pattern across multiple signals rather than any single red flag.
A Word of Caution
Approach every case assuming good faith first. The goal is to redirect students toward genuine learning, not to win a forensic argument.