What is the best AI detector for teachers?

Start with what you will do with the result

A detector gives you a number. Before choosing one, decide what that number is for. If it is a first pass over a stack of papers, to decide which ones to read more closely, most detectors will do. If you expect to act on it, to lower a grade or refer a case, no detector is enough by itself. Turnitin’s own guide says its score “should not be used as the sole basis for adverse actions against a student.”

Source: guides.turnitin.com, “Using the AI Writing Report”, checked 2026-10-22.

The main detectors, on their published facts

Sources: Turnitin guides and blog; gptzero.me pricing, technology and LMS pages; copyleaks.com; zerogpt.com; checked 2026-09-08 and 2026-10-22. All accuracy figures are the vendors’ own.

Why vendor numbers and real classrooms differ

Every number in that table comes from the vendor, measured on its own test set. Independent work finds higher error rates. Liang and colleagues at Stanford (Patterns, 2023) found GPT detectors flagged 61.3 percent of TOEFL essays by non-native English writers as AI-generated. A 2025 arXiv study found GPTZero flagged 8 of 50 human essays. Decrypt reported ZeroGPT rated part of the Declaration of Independence 97.93 percent AI-written. And the University of Waterloo turned off Turnitin’s indicator in September 2025 after an internal test flagged human text as 100 percent AI.

A small error rate also adds up. Across hundreds of essays a year, a teacher will meet some honest students with high scores. The false positive rates page collects every published number.

How to choose, by situation

• Your school licenses Turnitin. Use it, read the score as a prompt to look closer, and learn your school’s policy on how it may be used. See is Turnitin AI detection accurate.
• You are on your own and want a quick check. A free tier such as GPTZero’s covers occasional use. Do not pay for several detectors to compare scores; disagreement between them tells you little.
• You need to act on what you find. You need evidence about how the essay was written, which no detector provides. That means setting up the assignment so the writing is recorded.
• Your LMS shows an AI score already. Find out which tool produced it; the Canvas, Google Classroom and Schoology page explains where those scores come from.

When a writing record serves you better

For essays you assign, you can skip the guessing. When students write in a recorded editor, you see what was typed and what was pasted, where the student paused, and how the argument changed across sittings. That is evidence a student can explain and a reviewer can check, which a percentage is not.

EssayGuard is built for that. You create an assignment, set the rules for what may be pasted, and share one link; students write in the browser with nothing to install. Every keystroke, revision, pause and paste is recorded with server-side timestamps, and snapshots cross-check the record so a fabricated history does not read as writing. Each essay is marked Approved, Needs Review or Flagged with a plain summary, so you know which replays to open first. EssayGuard gives no AI percentage and does not judge whether prose sounds machine-made; you make every decision.

It is free for the rest of this semester for every course you teach, then $99 a year. Set up one assignment and compare.

Limits on both sides

A detector estimates something about the text and can be wrong about any single paper. A record shows how text entered the editor, not whose ideas it carries, and a patient student can retype generated text. Neither replaces a conversation with the student about the work.