AI detectors are not reliable enough to be treated as proof that a student used AI — the most-cited tools produce false positives, are biased against non-native English writers, and OpenAI even withdrew its own detector for being inaccurate. At Ancourage Academy, we think parents and students should understand how these tools really perform before trusting a score, so this guide explains what AI detectors can and cannot do — and what it means if your child is wrongly flagged. For the related question of when AI use actually crosses a line, see our guide on whether using ChatGPT counts as cheating.
As schools respond to ChatGPT, many have turned to "AI detectors" that claim to tell whether a piece of writing was produced by a person or a machine. The promise is reassuring. The reality, as the evidence below shows, is far messier — and a misplaced trust in these tools can punish honest students.
How Do AI Detectors Work?
AI detectors estimate the probability that text was machine-generated by looking at statistical patterns — they do not actually know who wrote something. The main signals they use are how predictable the wording is and how uniform the sentences are: AI-generated text tends to be smooth and statistically "average", while human writing tends to be more varied. A detector turns these patterns into a percentage — for example, "85% likely AI". Because the method is probabilistic, it can never deliver certainty; it can only offer a guess, and that guess is wrong often enough to matter.
Are AI Detectors Accurate?
No current AI detector is accurate enough to be treated as evidence on its own. The most telling signals come from the organisations closest to the technology:
- OpenAI withdrew its own detector. OpenAI launched an AI Text Classifier in early 2023 and shut it down within months, in July 2023, citing a "low rate of accuracy" — in its own testing it had correctly identified only about 26% of AI-written text.
- A Stanford study found systematic bias. Researchers (Liang and colleagues, published in the journal Patterns in 2023) found that popular GPT detectors flagged 61.3% of essays written by non-native English speakers as AI-generated, while flagging native-speaker essays only about 5.1% of the time.
- Universities have switched the tools off. Vanderbilt University disabled Turnitin's AI detector in August 2023, citing concerns about reliability, the opacity of its scoring, the unfairness to non-native English speakers, and the absence of any way for a falsely accused student to prove their innocence.
| Tool | What it claims | The problem |
|---|---|---|
| OpenAI AI Text Classifier | Probability a text is AI-written | Withdrawn in 2023 — caught only ~26% of AI-written text in OpenAI's own tests |
| Turnitin AI detection | An "% AI-written" score on submissions | Even a stated 1% false-positive rate mislabels honest papers at scale; disabled by Vanderbilt and other universities |
| Free detectors (e.g. GPTZero) | Instant human-or-AI verdict | Biased against non-native writers; easily bypassed by light paraphrasing |
Can AI Detectors Be Fooled?
Yes — and easily, which is the flip side of false positives. The same Stanford research that exposed bias against non-native writers also showed that simple prompting strategies can rewrite AI text to slip past the detectors — and in practice, light paraphrasing does the same. So the tools fail in both directions at once: they wrongly flag honest students while missing students who deliberately disguise AI use. A tool that produces both false accusations and false clearances cannot be a reliable basis for a cheating decision — which is exactly why a growing number of institutions have stepped back from them.
Why a False Positive Matters for Your Child
Even a "1% false-positive rate" becomes serious at the scale schools operate. Vanderbilt noted that a 1% rate applied to its roughly 75,000 annual submissions would wrongly flag around 750 papers — three-quarters of a thousand students potentially accused of cheating they did not commit. For an individual child, being wrongly flagged is stressful, hard to disprove, and can sit unfairly on their record. The bias finding makes this worse in a multilingual society like Singapore: a student who writes careful, simpler English — exactly what many bilingual learners do — is statistically more likely to be misread as a machine. A detector score should therefore never be treated as a verdict.
What This Means for Singapore Students and Parents
The practical takeaway is that detectors are, at best, a weak signal — and honest process is the real protection. Singapore schools are still working out their approach, and MOE's direction on AI in education emphasises responsible, teacher-guided use rather than reliance on any single detection tool. The strongest defence a student has is evidence of their own thinking: keeping drafts and version history, being able to explain their argument and choices, and showing the working behind a piece. Parents can help by encouraging children to write in a way that leaves a trail — and to use AI openly for hints and feedback rather than secretly for answers, which is the line we cover in our academic integrity guide.
How Ancourage Academy Approaches AI and Honesty
At Ancourage Academy we teach students to use AI openly and to keep their own thinking at the centre — which is also the best protection against a false accusation. Our AI workshops and workshops for students show learners how to use tools like ChatGPT to explain, quiz, and give feedback in small groups, while building the writing and reasoning that make work unmistakably their own. You can also browse our full AI in education guides for more.
Book a trial class (usually $18) at Bishan or Woodlands, or read our overview of AI study tools for students.
Common Questions About AI Detectors
Can a teacher prove my child used AI with a detector?
No. An AI detector only produces a probability, not proof. OpenAI withdrew its own detector for low accuracy, and major studies show these tools wrongly flag genuine human writing — especially from non-native English speakers. A responsible school treats a detector score as a prompt to ask questions, not as evidence of cheating on its own.
Why was my child's original work flagged as AI?
False positives are common. Detectors flag writing that looks statistically "average" — clear, simple, well-structured prose. A 2023 Stanford study found detectors labelled 61% of non-native English essays as AI-generated. Writing carefully in straightforward English, which many Singapore students do, can unfairly trigger a high AI score even when no AI was used.
Do Singapore schools use AI detectors?
Practice varies by school, and there is no single national rule requiring them. MOE's guidance focuses on responsible, teacher-guided AI use rather than automated detection. Where detectors are used, the fairer schools treat the result as one signal among many — alongside drafts, the student's ability to explain their work, and the teacher's own judgement.
How can my child protect themselves from a false accusation?
Keep a clear trail of their own work: save drafts, use a tool that preserves version history, and be ready to explain their reasoning and sources. Using AI openly — for hints, feedback, and checking rather than for finished answers — also helps, because honest, documented use is far easier to defend than secret use.
Explore our AI in education hub, read whether using ChatGPT counts as cheating, or see our guide to AI study tools for students.