Every secondary teacher has had that moment reading an essay that doesn't sound like the pupil. Here's an honest look at what actually works for detecting AI use — and why the software you might be considering isn't the answer.
Start with the uncomfortable truth about detectors
AI detection software is not reliable enough to accuse a pupil on. Detectors produce false positives — and they do so disproportionately for pupils writing in a second language and for pupils with certain writing styles. A tool that tells you an essay is "87% AI" is giving you a guess dressed as a statistic.
Never make a detector score the basis of an accusation. The consequences of wrongly accusing a pupil of cheating are severe, and detector output would not withstand scrutiny in an appeal. Use professional judgement and process, not software verdicts. See more on AI detectors in schools.
The signals that actually matter
A mismatch with known work
This is the strongest signal by far. You know this pupil's writing. A sudden leap in vocabulary, sentence complexity or structural sophistication is more telling than any software score.
Generic content with no specifics
AI essays are often fluent but oddly hollow — they discuss the topic in general terms without the specific details from your lessons, the text you actually studied, or the examples you gave.
Wrong or invented references
Fabricated quotations, misattributed sources, or references to editions and page numbers that don't exist. A very reliable tell.
Absence of the pupil's usual errors
If a pupil habitually confuses their homophones and this essay is immaculate but the analysis is unchanged in depth, that's a signal.
American spelling and idiom
Straightforward but surprisingly common — "color", "realize", "gotten", or American cultural framing in a British essay.
Uniform paragraph structure
Every paragraph the same length with the same rhythm, ending in a neat summary sentence.
What to do when you suspect it
- Don't accuse. Open with curiosity, not allegation.
- Ask them about their work. "Talk me through how you built this argument" — a pupil who wrote it can; one who didn't usually can't discuss their own reasoning.
- Ask them to expand a point verbally or write a short related piece in class.
- Follow your school's academic honesty policy — this needs to be a process, not an individual teacher's judgement call.
- Consider why. Often the answer is panic, workload or not understanding the task — which needs a different response from misconduct.
The better long-term answer: design
You will not win this by detection. The sustainable answer is setting work where AI is less useful, or where using it doesn't bypass the learning — in-class writing, drafts and process evidence, personal response, and tasks tied tightly to your specific lessons. See designing AI-resistant tasks.
And teach them to use it honestly
Pupils will use AI in their futures. A blanket ban teaches concealment; clear rules about when it's acceptable, how to declare it, and what constitutes their own work teaches something more useful. See teaching AI critical thinking.
Need a whole-school position on pupil AI use?
We help schools set policy and teach staff how to handle it consistently. AskColin gives schools low-cost monthly support with one-to-one help whenever you need it — so when you're stuck on a real job at 8am, there's someone to ask. Practical, jargon-free, built around your department.
Get one-to-one AI support for your schoolFrequently asked questions
Can you detect if a student used AI for an essay?
Not reliably with software. AI detectors produce false positives, disproportionately for pupils writing in a second language, and a score should never be the basis of an accusation. Professional judgement — particularly a mismatch with the pupil's known work — is far more reliable.
Are AI detectors accurate for schools?
No. They produce false positives and their output would not withstand scrutiny in an appeal. A tool reporting an essay as '87% AI' is giving a guess dressed as a statistic. Use judgement and process rather than software verdicts.
What are the signs a student used AI in an essay?
A mismatch with their known writing standard, fluent but hollow content lacking specifics from your lessons, fabricated or misattributed quotations, absence of the pupil's usual errors, American spelling and idiom, and uniformly structured paragraphs.
What should a teacher do if they suspect AI use?
Don't accuse. Ask the pupil to talk you through how they built the argument — someone who wrote it can discuss their reasoning. Ask them to expand a point verbally or write something related in class, follow your academic honesty policy, and consider whether panic or misunderstanding caused it.