Picture Frank Bell at 7:12 p.m. on a Tuesday. He has a stack of research papers on his kitchen table and a cold cup of coffee. Twenty-three of them read like polished encyclopedia entries. He wonders how many were written by a student and how many were pasted from a general assistant such as ChatGPT, Claude, Gemini or Microsoft Copilot. He considers running the batch through a detector, but he knows a score cannot tell him what his students actually did.
AI and academic integrity becomes a classroom problem when polished work gives you no view of how a student made it. A teacher stares at work that feels off. The first instinct is to police it. That instinct leads to a dead end.
Why policing fails
A detector score is weak evidence. It cannot establish how a student wrote a paper. It might flag a nervous English-language learner who wrote every word herself, or miss a submission stitched together from three different sources. A score cannot establish misconduct on its own. The conversation that follows a suspicious paper matters far more than any number on a dashboard.
OpenAI announced an AI-text classifier and later updated the announcement to say it was withdrawn. The update appears in OpenAI, New AI classifier for indicating AI-written text. The piece on Do AI detectors work? What teachers should know first explains why detector scores stay weak evidence. A score can be wrong in both directions, and it never replaces a teacher who looks at process and talks to the student.
The cost of policing shows up in the next class period. A teacher who spends forty minutes investigating a score is not spending those forty minutes discussing a student’s draft. A class built around suspicion leaves less room for students to show you their drafts.
This piece takes a different route to AI and academic integrity: design instead of detection. You shift the focus from catching misuse to making students’ reasoning easier to examine without relying on a detector.
Require visible process
Ask students to turn in one dated, in-class draft and a short set of notes from the work. If a student cannot produce a handwritten draft, a version history from a shared document can help. If your school’s document system keeps one, check what it actually records and treat it as a starting point for a conversation. These artifacts give you something concrete to discuss.
Picture Maya Rodriguez in eighth-grade science. She tells her class that the lab notebook counts for more than the final diagram. The notebook shows how the thinking changed. A student who relied on a general assistant still needs to explain every step in that notebook. The explanation becomes the evidence.
The prompt is the instruction, not the product. A student who uses a tool must still own the final choices. Process evidence makes that ownership visible.
Bring drafting into class
Set aside twenty minutes of class time for students to begin the work under your eye. They open notebooks while you walk the rows. The in-class draft becomes the anchor document. Everything submitted later must connect to it. When you sit with a student afterward, you can ask about the choices they made in that draft.
Say Jordan Kim uses this with the twenty-seven waiting responses. The first ten minutes of writing happen at desks in the classroom. Students who feel tempted to paste a full draft now face a simple problem. The in-class version already exists. They must show how they moved from that version to the final one.
Reserve drafting time and choose one piece of work students can explain later. A history paragraph, a science claim, a math justification. All of them start on paper in your room.
Add short oral check-ins
A brief conversation gives you another piece of evidence to consider alongside drafts and notes. After a student turns in an essay, pull them aside and ask them to walk you through one paragraph. What made them choose that evidence? How would they say the same idea in their own spoken words?
Hesitation alone proves nothing. A student may hesitate for many reasons. The goal is not to trap them. The goal is to hear their thinking in real time. A student who can explain the work in their own voice gives you another piece of evidence to weigh alongside the draft and the student’s account.
A five-minute check-in lets the teacher ask about a specific choice in the student’s work. Let students refer to their draft or notes unless the skill you are assessing specifically requires recall. The point is to hear how they think, not to test memory under pressure.
Anchor the work in local and personal context
A prompt that asks for an analysis of climate policy in a generic city produces generic answers. Change the assignment to “the corner of Main and Elm where the new EV charging station went in last spring.” Ask students to record their observations during the assignment and be prepared to explain them. Those records still need checking.
For Frank Bell’s social-studies class, one possible assignment is to ask students to interview someone who lives within ten blocks of school. If an interview is not accessible, a student could observe a local public space for thirty minutes and take notes. Another option is to supply a local document, such as a school newsletter or a city planning notice, and ask students to respond to it. A student can choose the option that works without disclosing private family information.
Personal context raises the bar for a copied response. A general assistant cannot visit the corner of Main and Elm. It cannot sit on the bench outside the library for thirty minutes. A student who does that work brings back details worth checking, because a tool can invent plausible local color. Ask students to keep their notes and to explain one specific observation. Check the claims rather than treating local detail as proof.
AI and academic integrity: a worked example in middle school science
Maya wants students to explain why thermal energy moves from warmer objects to cooler ones. She gives them a short in-class draft prompt on Monday. They sketch their first explanation on paper. On Wednesday she asks three students to talk her through one line from that draft.
For the final submission she sets up a simple observation in class. Students watch ice melt in a cup on the bench, record the temperatures of the ice and the room, and note which is warmer. A student who cannot attend that day gets a short video of the same setup and a set of recorded temperatures. Maya asks each student to separate what they observed from their explanation of how thermal energy moved. The draft, the observation and the explanation give her specific work to discuss with each student.
AI and academic integrity in high school social studies
Frank Bell adapts the same pattern for a persuasive-writing unit in his social-studies class. Students draft a thesis on paper during the first ten minutes of class. They later submit a short audio recording of themselves reading one body paragraph aloud.
They must also include notes from a conversation with a family member or neighbor about the local issue they chose. A student who cannot arrange that conversation can analyze a provided letter to the editor from the local paper instead. The oral check-in happens in the last week of the unit. Frank chooses students whose final drafts raised questions he could not resolve on paper, plus a few volunteers. He asks each one to talk through a claim, and they may keep their notes in front of them.
The assignment still produces a written essay. The surrounding evidence gives Frank questions to ask about the essay and a basis for judging the student’s explanation.
A contrasting example in math
A math teacher can use the same pattern without essays. Say a hypothetical algebra teacher gives a word problem about a phone plan. Students start the solution on paper in class, showing their first attempt even if it stalls. The final submission includes that in-class start, the worked solution, and three sentences explaining where the first attempt went wrong.
The teacher checks two things. First, does the final method connect to the in-class start, or did a complete solution appear from nowhere? Second, can the student explain one step in their own words during a short check-in? A student who used a tool for the final version still has to account for the gap between the first attempt and the finished work.
A complete before-and-after assignment
Here is how the pattern could change a typical essay assignment. Before: “Write a five-paragraph essay analyzing the causes of the American Revolution. Due in two weeks.” That prompt invites a generic response, and a general assistant can draft a generic response to it.
After, the same unit looks like this. “In class on Monday, draft a one-paragraph claim about which cause mattered most in [a colony or community covered by our source packet]. By Friday, find one primary source from the provided packet that supports or complicates your claim, and annotate it in the margins. In the final week, revise your paragraph into a two-page essay. Attach your in-class draft, your annotated source, and a half-page note explaining what changed between the draft and the final version.”
The process artifacts are the in-class claim paragraph, the annotated source, and the revision note. Check-in questions for a few students might be: “Why this source and not the one on the next page?” “What did you cut from your first draft, and why?” “Which sentence in the final version are you least sure about?”
A common mistake with AI and academic integrity is to treat these moves as extra work. They replace older requirements instead of adding to them. If you add an in-class draft, remove the separate outline requirement you used to collect. If you add a revision note, drop the generic rubric checklist students used to fill in. Collect the new pieces in place of the old ones, not as well as them.
A prompt that helps you redesign an assignment
If your school has approved a general assistant, you can ask it to do a first pass on a redesign. Keep student work and identifying information out of the prompt unless school policy permits their use. Then check every suggestion and decide what fits your class.
You are helping a teacher redesign an assignment so the process is visible.
Subject and grade: [subject, grade]
Current assignment: [paste the assignment text]
Class time available: [minutes or periods]
Constraints: [materials available, students without internet at home, other limits]
Propose a revised version that includes: one in-class starting task, one piece of
process evidence students submit, two or three short oral check-in questions, and
an accessible alternative for any step that requires resources outside school.
Flag any step that could exclude students and suggest a workaround.
Check the output against your students and your schedule. Decide what to keep.
Write a clear class policy on AI and academic integrity
A clear class policy turns these expectations into shared rules instead of surprises at deadline. Keep the document to one page. Use plain sentences. State the five non-negotiables that match the responsible AI norms in Responsible AI in the classroom: a short set of house rules.
Students must disclose any tool use. A human makes every final decision. All facts must be verified from trusted sources. Private data stays out of public tools. Access must stay fair. Distinguish permitted assistance from prohibited assistance. Specify what students must disclose and what they must submit. Align the wording and any response to suspected misuse with your school’s policy. Do not assume every school permits a general assistant.
AI and academic integrity depends on students knowing exactly where the lines are. A template like this gives them the lines in plain language. Fill in the brackets with your own decisions, then check the whole document against school policy before you share it.
Approved tools for this class: [list the tools your school has approved, if any]
Permitted assistance: [for example: brainstorming ideas, checking grammar, outlining]
Prohibited assistance: [for example: generating full paragraphs, writing final
drafts, completing in-class work]
Disclosure: at the end of each submission, state which tools you used and how
you used them.
Required process evidence: [for example: one handwritten in-class draft, a
version history, or a dated outline]. I will ask you to explain any section in
your own words.
Where to get help: [your office hours, a class discussion board, a peer
tutoring slot]
Share the policy with your class this week. Spend ten minutes reading it aloud together. Ask what questions come up. The conversation itself matters more than the wording.
Talk to students about it
That conversation deserves its own space. Sit down without grades in front of you. Tell students why the work matters to you. You might say something like: “I am not here to catch you. I am here to see what you understand. If the assignment and school policy allow a tool to help you brainstorm, fine. If the rules say no tool may write your paragraphs, then no tool writes your paragraphs. Show me your work. Talk to me about it. That is how I know what you actually understand.”
A student might say they use a tool when they feel stuck. Another might say they worry about falling behind if they do not. Listen first. Then restate the policy in the language they just used. The goal is to make the boundary plain and the support visible.
The same honesty applies when you build assessments. The piece on AI for assessment: better questions in less time shows how to draft targeted questions that surface specific misconceptions instead of generic recall. An in-class, on-paper response gives you useful evidence when you need to see what a student can do without assistance. A separate answer key helps you stay consistent. The tool never decides the grade.
Step-by-step routine for one assignment
Pick the research paper or the end-of-unit essay that has given you the most doubt in the past. Rewrite the prompt on Monday. Require one handwritten draft in class, one documented local observation the student can discuss, and a short check-in with a few students before the final version counts.
If your school has approved a general assistant, run the prompt through it first so you see what students might see. Then adjust. Save the new prompt in a folder labeled “revised for process.” On Tuesday, share the one-page policy with the class and read it together. Note which sentence raises the most hands. Change that sentence before Friday.
On the day students begin the in-class draft, walk the rows. Make quick notes on who seems stuck. Use those notes to decide which students need a check-in later. When final submissions arrive, read the process evidence first. Only then read the polished version. The order matters. It keeps the focus on thinking instead of surface polish.
Use part of an existing drafting period rather than adding a new one. When the final version arrives, review the draft alongside it. That is the scheduling choice this routine asks for: one block of class time, used differently.
Questions teachers ask
What if a student still submits obvious tool output?
Start with the conversation. Ask the student to explain one paragraph, with their draft in front of them. Ask to see the in-class draft. Ask how the local observation shaped their thinking. The conversation may clarify what happened. If concerns remain, follow your school’s policy exactly. The International Center for Academic Integrity is a professional association working on academic integrity.
How do I handle different school policies?
Your school may prohibit these tools or permit specified uses with disclosure; write your class rules to match its policy. The five non-negotiables from the house-rules piece give you a stable core that travels across policies.
Does this approach take more time?
It shifts where the time goes: away from detector checks and toward reading process notes and holding short conversations. If your time is tight, check one draft decision per student rather than every version. Pick the moment where the thinking changed and look at that.
What about students who need more support?
Plan accessible alternatives for each process requirement, and follow students’ accommodations and school policy. A student who struggles with writing can produce a short draft in class and explain it in a format permitted by their accommodations and school policy. The local context often gives them something concrete to talk about. Offer scaffolds inside the permitted assistance list. The policy stays the same. The support stays visible.
Can I use a document-grounded tool for any of this?
A document-grounded tool such as NotebookLM is intended to work from documents you give it. You might feed it your local planning notice and ask for a neutral summary, then check that summary against the notice before students see it. A public planning notice may fit Green under the book’s rule, but check school policy before using the tool, and do not enter private student information.
Is AI and academic integrity just about writing classes?
No. The math example earlier in this piece uses the same pattern with no essay at all. Any subject that asks students to produce reasoning can ask for an in-class start, a piece of process evidence, and a short explanation. AI and academic integrity is a design question, not an English-department question.
The chapter “Safety: Privacy, Detectors, and High-Stakes Calls” in Keep It Simple: AI for Teachers covers privacy, detectors, and high-stakes decisions. The book also presents the Green, Yellow, and Red rule for deciding what to paste into a tool. The free sample on this site contains the book’s first four chapters.
Pick one assignment you already give. Choose one process artifact, an in-class draft or a revision note, and swap it in for a requirement you currently collect. Check that change against your school’s policy before you share it with students.