What AI Detection Tools Can and Cannot Tell You

AI detection tools have become a fixture in conversations about academic integrity. Many schools are using one, considering one, or working out whether they should. But the conversation about which tool to use often moves faster than the conversation about what these tools actually measure - and what happens when they get it wrong.

Before committing to any AI detection approach, it is worth understanding what a detection score represents, why false positives happen, and what the professional implications are for a teacher who acts on a result without fully understanding its limitations. Getting this right is not just about choosing the best tool. It is about using any tool in a way that is fair to students and defensible to parents and governors.

What a Detection Score Actually Means

Every AI detection tool returns some version of a probability score. A result of 75% does not mean that 75% of a document was written by AI. It means that, based on statistical patterns in the text - sentence structure, vocabulary choices, predictability of word sequences - the tool calculates a 75% probability that AI was significantly involved in producing it.

This distinction matters enormously in practice. A probability is not a verdict. It is information - useful information, but information that requires interpretation. A teacher who treats a high detection score as proof of AI use is misapplying the tool. A teacher who understands it as a signal that warrants closer scrutiny is using it correctly.

The statistical patterns that detection tools look for are based on how AI language models generate text: they predict the most probable next word given the preceding context, which produces text that is statistically smooth in a way that human writing often is not. Human writing has more variation, more unexpected choices, more idiosyncrasy. AI output tends to be more predictable at the word-sequence level, even when it reads naturally to a human.

Why False Positives Happen - and Why They Matter in Schools

False positives - high detection scores on work that was genuinely written by a student - occur for identifiable reasons, and several of those reasons are common in the writing secondary students are taught to produce.

Students who have been taught to use specific academic phrases, evaluative connectives, or structured argument frameworks will often produce text that scores higher on detection tools than their peers who write more informally. A Year 12 Sociology student who has practised writing in an elevated academic register is not writing suspiciously - but a detection tool trained primarily on general internet text may not distinguish between learned academic formality and AI-generated prose.

Similarly, second-language learners, students who have received significant tutoring, and students who closely follow a teacher-provided structure may all generate writing that detection tools flag as higher probability. None of these are evidence of AI use. They are evidence of teaching.

This is why the framing of detection results matters as much as the score itself. A tool that returns a number without contextual framing leaves the teacher to interpret it alone. A tool that returns a number alongside guidance on what the score means, what factors might inflate it, and what an appropriate professional response looks like is a significantly more responsible product for classroom use.

How GradeOrbit Returns Detection Results

GradeOrbit's detection scores are always accompanied by framing that reinforces professional judgment. The tool does not label work as "AI-generated" - it returns a score from 0 to 100 for how far a piece departs from the pupil's own earlier writing, with a Low, Medium or High confidence label, the signals it noticed and a plain-English reasoning paragraph, and makes clear that the teacher's contextual knowledge of the student, the submission conditions, and the assessment history is an essential part of the interpretation.

Detection is available on GradeOrbit's school accounts, where it is unlimited and included with every teacher seat, with no tiers to weigh up. Because a school account holds pupils' marked work under a signed Data Processing Agreement, detection compares each submission against that pupil's own earlier writing on file - a stronger signal for coursework or controlled assessment than a generic one-off score would be - and a pupil with no earlier work on file is skipped rather than scored. It handles handwritten work as well as typed submissions: teachers upload scans, photos or PDFs, or use the phone QR camera link, and the AI reads the page images directly, with no manual transcription.

Uploaded scans on a school account are removed when a run archives (90 days by default, adjustable by the school), and pupil records follow the school's retention window and are wiped at offboarding. Individual teacher accounts do not include detection; there, uploaded student work for marking is redacted client-side and never stored after processing. Across every tier, student work is never used to train AI models. If you want to understand more about how to interpret detection outputs in practice, our guide on how to interpret AI detection scores covers the professional workflow in detail.

Using Detection Tools Responsibly

The most important thing any school can do before deploying an AI detection tool is establish what detection results will and will not be used for. A detection score should trigger a process - a conversation, a review, a request for a supervised rewrite - not a unilateral decision. The score is the beginning of professional judgment, not a substitute for it.

Schools that document this process, communicate it clearly to staff, and apply it consistently are in a much stronger position if a detection result is ever challenged by a student or parent. Schools that treat detection scores as verdicts are not.

Testing Detection on Real Work

Whatever tool you use, test it against your own process before relying on it: run it on work you know well and check whether the results, and the way they are framed, would lead staff to the right next step. A GradeOrbit school account can be tested this way during a free school trial, with baseline-aware checks run under a school Data Processing Agreement. Details are on the GradeOrbit for schools page.

Choosing between tools that mark and detect? Our comparisons of AI marking tools set GradeOrbit against named alternatives, with every claim about them sourced and dated.