[ GUIDE ]

Do AI detectors work? What schools should know

AI-writing detectors misjudge real students, especially those writing in a second language. What the research found, what a flag means, and better evidence of who did the work.

Plan a pilot Watch the demo

Facts as of Oct 18, 2026Last reviewed Oct 19, 20263 min read

61%

of essays by non-native English writers flagged as AI-written, on average across seven detectors

Source: Liang et al., Patterns, 2023

[ SUMMARY ]

The short answer

AI-writing detectors misjudge real students, especially students writing in a second language. A detector score should never be the basis for a decision about a student.

What the research found

In a widely cited 2023 study, seven AI-writing detectors flagged on average 61% of 91 TOEFL essays by non-native English writers as AI-written; 89 of the 91 essays (97.8%) were flagged by at least one detector. Essays by US eighth-graders were classified almost perfectly. The sample was small and the detectors date from early 2023, so treat the exact rate with care; the direction of the bias is the point.

How often schools rely on them

In 2023-24, 68% of teachers of grades 6-12 used AI detectors, up 30 points, and 64% reported students disciplined over generative AI, up from 48%. Students with disabilities and English learners were more exposed (CDT, via K-12 Dive, Apr 2024).

What a flag means

A flag is a guess, not a finding. Turnitin reported that about 11% of papers it checked had at least 20% AI writing and about 3% had at least 80%; those are detector flags, not confirmed use.

Better evidence of who did the work

Drafts and version history.
Work started in class.
A short conversation in which the student explains the work.
Class AI chats, where students are told the chats can be reviewed.
Unaided checks on the same skill.

A fair process when work looks AI-written

Look at the process first: drafts, version history, the class chat, notes from class.
Talk with the student. Ask them to walk you through two choices they made and to explain one key idea without notes.
If doubt remains, ask for a short redo of one part in class, with no AI.
Decide on the evidence you gathered, not on a score, and write down what you saw.
Treat a first case as a teaching moment: restate the assignment's AI rule and how to disclose AI use.

Tell students how integrity is checked

Students should know before an assignment, not after, how the school checks whose work it is: what counts as process evidence, whether teachers can review class AI chats, and that a detector score alone never decides an outcome. Put it in the syllabus AI statement and the acceptable use policy.

If your school keeps a detector

Never use a score as the sole basis for discipline.
Require a person to review process evidence and talk with the student.
Tell students and families how the tool is used.
Check whether flags fall more often on English learners or students with disabilities.

Where Nexus stands

Nexus does not detect cheating or AI-written work. It shows teachers how a student worked in class chats, and students are told when a chat can be reviewed.

[ FAQ ]

Questions and answers

Are AI detectors accurate?

Not for many students. Seven detectors flagged on average 61% of essays by non-native English writers as AI-written in a 2023 study, while US eighth-graders' essays were classified almost perfectly.