How Teachers Detect AI in GCSE Combined Science Write-Ups

Detecting AI in GCSE Combined Science practical write-ups has become part of the marking routine for many science teachers. The required practicals - from investigating osmosis in potato chips to measuring the resistance of a wire - produce method, results, and conclusion sections that students increasingly draft with the help of tools like ChatGPT and Claude. When a write-up reads more like a textbook than a fourteen-year-old who was in your lab last Tuesday, you need a fair, consistent way to check.

This guide looks at how teachers use AI likelihood scores to review Combined Science write-ups, where the false positives tend to hide, and how to keep professional judgment at the centre of any decision.

Why Science Write-Ups Are a Special Case

Practical write-ups have a structure that AI models reproduce very well. There is a standard shape - aim, hypothesis, variables, method, results table, analysis, conclusion, evaluation - and a bank of expected scientific vocabulary. An AI model can generate a competent-looking write-up for a required practical in seconds, complete with the language of "independent variable", "anomalous results", and "valid conclusion".

That same predictability makes science write-ups genuinely hard to judge by eye. A conscientious student who has revised the method thoroughly can legitimately produce writing that reads formally and uses precise terminology. So the signals that might look suspicious in a creative English piece - polished phrasing, textbook structure - are often exactly what you are teaching students to produce in science. This is precisely why a probabilistic tool is more useful here than gut feeling alone.

How AI Detection Actually Works

AI detection is probabilistic, not a verdict. GradeOrbit's detection, included on school accounts, compares a new piece of work with that student's own earlier work already on file and returns a score from 0 to 100 for how far it departs from their own writing, with a Low, Medium or High confidence label, the signals it noticed, and a plain-English explanation. A student with no earlier work on file is skipped rather than scored, and you are told. It is not a lie detector and it does not "prove" anything. A 90% score is a strong signal worth investigating; it is not, on its own, an allegation of malpractice.

For a Combined Science write-up, the score is most useful when you read it section by section in your head. A student's own method - copied from the board or their exercise book - may score very differently from their conclusion or evaluation, where independent reasoning is expected. A write-up where the routine sections look human but the analysis suddenly reads like a revision guide is worth a closer look. Our guide on how AI detection likelihood scores work explains the reasoning behind the number in more detail.

Where False Positives Hide in Science

Science write-ups produce more false positives than almost any other subject, and it pays to know why before you act on a score.

• Standard method wording. If your class copied a method from the same worksheet, thirty students will submit near-identical, formal text. That formality can push scores up without any AI involvement.
• Taught sentence stems. Writing frames like "The results show that... because..." are structures you actively teach. Rewarded phrasing can read as machine-like.
• EAL students. Students writing in an additional language often lean on memorised scientific phrasing, which can raise a score unfairly.

This is why a high score is a prompt to look, not a conclusion. Compare the write-up against the student's classwork, their exercise book, and what you saw them do in the lab. A student who can explain their own conclusion when you ask them about it is telling you far more than any percentage can.

Handling a High Score Fairly

When a Combined Science write-up returns a high likelihood score, treat it as the start of a professional conversation, not the end of one. Look at whether the science is actually correct - AI-generated write-ups often contain plausible-sounding but wrong reasoning, such as confusing correlation with a valid conclusion, or citing a control variable that was never mentioned in your lesson. Ask the student to talk you through their method and what their results mean. Genuine understanding is quickly obvious.

Keep a short, factual record of what you found and what you decided.

Keeping Detection Data Protected

AI detection in GradeOrbit runs on school accounts, under a signed Data Processing Agreement. Because the check compares each write-up with the student's own earlier writing, pupils' work is held with their real names for the school's chosen retention window. Scans are removed when a run archives (90 days by default, and the school can set anywhere from 7 to 365 days), pupil records are wiped at offboarding, and work is never used to train AI models. Individual teacher accounts don't include detection; there, work uploaded for marking is used to produce the marks and feedback, then discarded, and students are referred to anonymously as Student 1, Student 2, and so on.

Try GradeOrbit's AI Detection Tool

GradeOrbit gives UK science departments a fair starting point for checking GCSE Combined Science practical write-ups - a 0 to 100 score showing how far a write-up departs from the student's own earlier work, which you interpret with your own knowledge of the class, the practical, and the student in front of you. It is included in every school teacher seat, and it never replaces your judgment; it gives you evidence to apply it consistently.

See how AI detection works for schools and start a free school trial before your next set of science write-ups.