AI detectors in a high school classroom: what to use and what to check first

What a school usually has already

Most US secondary teachers are not shopping for a detector from a blank page. They work inside whatever the district already pays for, and two platforms are worth understanding before anything else: Google Classroom and Turnitin.

Classroom's originality reports compare a student's Docs, Slides and Word files against web pages and books, and against earlier submissions from your school or district if an administrator has switched school matches on (Google Classroom Help). On the free Fundamentals edition a teacher can turn them on for 5 assignments per class, and a file larger than 2MB will not produce a report. Neither of the two Google help pages on the feature cited here mentions AI-generated text, which fits what the reports are: a similarity check. The gap is covered in detail in can Google Classroom detect AI?

What Classroom does bring is a published privacy position. Google states that Workspace for Education Fundamentals meets FERPA, COPPA and the Student Privacy Pledge, and that assignment files are not saved to a global repository (originality reports and privacy). That is the district's agreement doing work on your behalf, and it is exactly what a free website you found yourself usually cannot offer.

A school that pays for Turnitin does not automatically have AI writing detection. Turnitin's FAQ says the AI indicator is available only to institutions that license Turnitin Originality, and tells Similarity, Feedback Studio and Originality Check customers to ask their account manager about access (Turnitin Guides). The AI report is not visible to students, though an instructor can download it as a PDF and share it. Your IT or library team can confirm which product your district bought. How the indicator behaves once you have it is on our Turnitin page.

Some districts are also trialling Turnitin Clarity, which shows the writing process behind a submission rather than only the finished text. In a September 2025 press release Turnitin said nearly 100 US secondary schools or districts had launched pilots in its first 60 days, described it as a paid add-on for Feedback Studio customers, and said its AI writing detection would appear in Clarity's writing report only for customers who also license Turnitin Originality (Turnitin press release). Those are the company's own figures and descriptions; we have not tested Clarity.

We have not captured consumer detectors' prices for this page, so we quote and rank none. In a school the more important question about any of them is the one in the next section: whose contract the student's text travels under.

Before you paste student work into a free detector

Universities deal with FERPA too, but in a middle or high school most students are minors, so their parents usually hold the FERPA rights: those rights pass to the student at 18, or on entering a postsecondary institution at any age (What is FERPA?). This is where a well-meaning teacher can go wrong fastest. What follows summarises federal guidance. It is not legal advice, and the Department of Education's own guidance says it is always a best practice to consult legal counsel on these agreements. The FTC notes that many schools already have a process for approving online services (FTC COPPA FAQ, section N).

FERPA: officially, "it depends"

The Department's Privacy Technical Assistance Center was asked whether student information used in online educational services is protected by FERPA. Its February 2014 guidance answers, in its first two words, "It depends" (PTAC, 2014). FERPA protects education records, defined as records directly related to a student and maintained by the school or by a party acting for it. Whether an essay pasted into an outside website becomes one of those is not a question a single teacher can settle at a keyboard.

When a school does share records with a vendor, it usually relies on the "school official" exception. The Department lists what the outside party must do: perform a service the school would otherwise use its own employees for, stay under the school's direct control with respect to the records, and use the personal information only for the purpose it was disclosed for, as well as meet the school's annual-notification criteria (who is a school official?). A free site you signed up to with a personal email is under nobody's direct control at the district.

The Department has an FAQ written for this exact situation, a teacher who wants to use an online tool in a course. Its first step is plain: check with school or district administration whether the application is approved for classroom use, and talk to your IT representatives before using it (Protecting Student Privacy FAQ).

COPPA: when students under 13 use a tool themselves

COPPA covers personal information collected online from children under 13, and it is narrower than its reputation. The FTC's FAQ says plainly that COPPA only covers information collected online from children, and not information collected from adults that may pertain to children (FTC COPPA FAQ, F.4). A teacher pasting an essay is the adult supplying the text, so that paste is mainly the FERPA and district-approval question above.

COPPA comes into play when students under 13 use a service directly: typing into a detector or chatbot themselves, or working through a class account. For that case the FTC's FAQ says a school may act as the parent's agent and consent on their behalf, but only in the educational context: where the operator collects students' information for the use and benefit of the school, and for no other commercial purpose. It recommends that schools or districts, not individual teachers, decide whether a service's information practices are appropriate (FTC COPPA FAQ, section N). So setting up a sixth-grade class on a new AI tool is a decision to route through the district.

The FAQ is dated July 2020 and carries a notice that the COPPA Rule was amended on 22 April 2025. The amended Rule took effect on 23 June 2025, with most compliance obligations due by 22 April 2026. On schools, it changes nothing yet: the FTC proposed ed-tech and school-authorization amendments but did not finalise them, to avoid conflicting with expected changes to the FERPA regulations, and said it will continue to enforce COPPA in the ed-tech context consistent with its existing guidance, meaning the FAQ (90 FR 16918). State law adds its own layer: the FTC's example is California's Student Online Personal Information Protection Act, which restricts using K-12 students' information for targeted advertising, profiling or onward disclosure.

What to look for in a detector's terms

The 2014 guidance makes a point teachers rarely hear: clicking "I accept" on a click-wrap agreement enters the provider and the school into something akin to a signed contract. It advises caution about terms the provider can amend without notice, because that sits badly with FERPA's direct-control requirement, and it suggests saving a copy of whatever terms you accepted. Before a paste, the terms should answer these:

• Is the agreement with you personally, or with your district?
• What minimum age does it set, and does it mention students under 13 at all?
• What happens to submitted text: stored, used for training, shared?
• Can the provider change the terms without telling you?
• Does it describe a school or district agreement, and does your district have one?

GPTZero is a useful illustration because it publishes both arrangements. Its student-privacy post says that, consistent with the FTC's position, it relies on education-institution customers to obtain parental consent before collecting the minimum personal information from students under 13 (GPTZero & Student Privacy). Its general privacy policy, last updated 2 April 2026, says it does not knowingly collect data from children under 18 (privacy policy). Those describe different regimes, and which one covers a given paste depends on whether a district contract sits behind it. We are quoting their wording, not grading it.

Stripping the student's name first reduces what you disclose. It does not answer the approval question, and an essay about a student's own family, town or illness can identify them without a name anywhere on it. If none of this featured in your onboarding, you are in large company: in CDT's 2023 survey, nearly one-third of teachers said they had not received formal student-privacy training from their school, a share unchanged from the year before (Off Task). A longer account of what one large vendor keeps is in does Turnitin store your papers?

What surveys of teachers found

The Center for Democracy & Technology has surveyed US grade 6-12 teachers repeatedly; three of its reports are below, each with a published method, and Up in the Air sampled public-school teachers only. Question wording and samples differ between them, so each figure below stays with its own report. Reading them as one trend line would be a mistake.

Up in the Air carries the numbers most worth sitting with. Only 25% of teachers said they were very effective at telling whether an assignment was generated by AI or written by the student. Reports of students getting into trouble for using AI, or for being accused of it, rose from 48% to 64% between the two school years, and teachers who regularly used a detector reported it more often: 72%, against 48% of those who did not. Forty percent agreed that a student had got into trouble for how they reacted when an adult at school confronted them about suspected misuse, which is a discipline outcome about the confrontation rather than the essay (Up in the Air).

Distrust had eased since CDT's previous survey, but in Up in the Air 52% of teachers still said generative AI had made them more distrustful of whether students' work is actually theirs, and the share was 69% at schools that ban the technology against 48% at schools that permit it. A survey cannot tell you whether that distrust was justified. It does suggest a school's policy shapes how its teachers read the same kind of work (Up in the Air).

Hand in Hand also found the share of teachers who had never used a detection tool falling from 39% to 29% between 2023–24 and 2024–25, within that survey's own question (CDT, 2025).

Middle school, short homework and younger writers

Age terms come before any score

OpenAI's Terms of Use, effective 1 January 2026, set a minimum age of 13 (or the minimum required in the user's country) and require anyone under 18 to have a parent or legal guardian's permission (OpenAI). A twelve-year-old using ChatGPT is outside those terms. That is a matter for the school and the family before it is a detection question.

Gemini is less tidy. Google's sign-in help says users must be 13 or over with a personal or school account and 18 with a work account. A parent who manages an under-13 Family Link account can turn Gemini off for it, and the Gemini web app is not available to Family Link accounts at all (Gemini Apps Help). On school accounts, though, administrators decide. Google's education page lists the Gemini app's minimum age as "All ages" on the Workspace for Education editions, notes that some features are only available to users 18 and over, and sets 18 or over for Google AI Pro for Education (Gemini Apps Help, Education). Separately, Gemini in Classroom is on by default for users of all ages (Workspace Admin Help). What a given sixth-grader can open depends on your district's settings, so ask IT rather than assume.

Why a 200-word answer gives a detector almost nothing

Turnitin will not produce an AI writing report on fewer than 300 words of long-form prose or more than 30,000, and it lists English, Spanish, Japanese and Arabic as supported languages (file requirements, updated 22 September 2026). A paragraph response, a set of short answers or a lab conclusion falls under that line before anything else is considered. Other tools may accept shorter text, but fewer sentences means less for any model to weigh; our answer on whether text length affects AI detection sets out why a short passage makes every score weaker.

For short homework, the useful checks are ordinary teaching moves. Ask the student to explain the answer aloud at the start of the next lesson. Set it beside something short they wrote in class. None of that needs a login, a privacy review or a percentage.

English class: essays you can watch being written

An English teacher has one advantage a university lecturer rarely does. You see the same students several times a week, and a real share of their writing can happen in front of you.

Start the year with an in-class sample. A timed response or an opening paragraph written in the room gives you each student's voice on record. When a later essay sounds unfamiliar, that sample is better comparison material than any score, and it is already in your gradebook.

Know the limits of version history. Google's help page is precise. You need permission to edit a file to browse its earlier versions, revisions are sometimes merged, and the file owner can permanently delete unnamed version history (Google Docs Editors Help). If students own their documents, make sure you have edit access while the assignment is open, and treat a thin or merged history with caution rather than suspicion. Extensions that run inside the editor are a separate category again; our guide to which Docs and Word extensions score text and which replay the writing sets them side by side.

Conference over a paragraph. Five minutes at a desk asking why this quotation, or what this sentence is claiming, shows whether a student owns the argument, and it teaches at the same time. The student-side view of the same evidence is our guide on how to prove you wrote it yourself, which is worth sharing with a class before the first essay rather than after a dispute.

When a score does come back high

Treat a high score on a school essay as a reason to look at the process evidence above, not as a finding; the documented reasons human writing gets flagged apply to teenagers as much as to anyone. Thresholds, length, second opinions and how to open the conversation are all on our checklist for any teacher before raising a score with a student, so they are not repeated here.

This page is written around US federal rules. Teachers in UK secondary schools work under a different legal and exam-board framework, set out in our page on AI in GCSE and A-level coursework.

AP Seminar and AP Research run on College Board's rules as well as your own. College Board requires interim checkpoints and a teacher's affirmation that each final submission is the student's own work, and it runs those submissions through Turnitin; does College Board check for AI? sets out the 2026-27 policy.