How Teachers Detect AI in A-Level Chemistry Coursework
AI use in A-Level coursework is no longer a theoretical concern for Chemistry teachers. Extended experimental write-ups, data analysis sections, and evaluations that once showed a student's individual scientific reasoning can now be produced in seconds by a language model. For teachers assessing Non-Exam Assessment (NEA) work, the question is increasingly not whether AI assistance is happening, but how to respond when a likelihood score comes back high.
GradeOrbit gives A-Level Chemistry teachers a built-in AI detection tool that returns a likelihood score between 0 and 100 per cent. That score is the starting point for a professional process — not a verdict. This guide focuses on what to do once you have a score, how to interpret it fairly, and how to handle the conversation with a student and your school's academic integrity framework.
Why A-Level Chemistry NEA Is a Particular Detection Challenge
Chemistry NEA writing has characteristics that make it harder to assess than, say, a history essay. Experimental write-ups follow a rigid structure: method, results, analysis, evaluation. That formal structure means AI-generated text and a well-drilled student's own writing can look superficially similar — both will use precise, impersonal scientific language, passive constructions, and hedged conclusions.
The sections that tend to show the clearest detection signals are evaluations and discussions. These are where students are asked to reason about sources of error, suggest improvements, and connect their findings to wider scientific principles. A student who has genuinely engaged with the experiment will typically show some inconsistency in depth, some hesitation in phrasing, and some very specific references to their own data. AI-generated evaluations, by contrast, tend to be uniformly fluent, structurally complete, and oddly generic — the errors mentioned are plausible but not specific to the actual experiment.
Understanding this distinction helps you read a likelihood score in context. A score of 75 per cent on a results table or method section is less informative than the same score on an evaluation or discussion paragraph.
What a Likelihood Score Actually Tells You
GradeOrbit's detection score is probabilistic. It reflects how closely the linguistic patterns in the submitted work match those typically found in machine-generated text. It is not a binary answer, and it is not infallible. Students who write in a very formal, structured academic register may receive elevated scores despite working independently. English as an Additional Language (EAL) students sometimes produce text that reads as unusually uniform because they are applying careful grammatical rules rather than writing with natural variation.
The difference between GradeOrbit's 1-credit and 3-credit detection models matters here. The 3-credit model runs a deeper analysis and is better suited to longer, more complex documents such as a full Chemistry NEA. If you are detecting on a substantial piece of coursework, the 3-credit model gives you a more robust signal. The 1-credit model is appropriate for shorter submissions — a single extended question response or a short lab report.
A useful mental frame: treat the score as you would a safeguarding referral threshold. A high score does not mean guilt. It means the threshold for starting an inquiry has been crossed. Everything after that point is professional process.
When to Act on a High Score
Most schools set a threshold in their academic integrity policy — often somewhere between 60 and 80 per cent — at which a teacher is expected to take a next step. If your school does not yet have a written threshold, now is a good time to raise it with your head of department or SLT. GradeOrbit's detection output can be saved and cited as part of that policy conversation.
Below threshold: note the score in your marking records and continue. A mid-range score (30–60 per cent) on its own is not grounds for action, but it may prompt you to look more carefully at the submission and compare it to the student's previous written work.
Above threshold: do not mark the work as failed or refer it to an exam board without completing an internal process first. The score is the beginning of an inquiry, not its conclusion. Document the score, the date, and the section of the submission that generated the highest signal. This documentation protects both you and the student.
How to Have the Conversation with the Student
The fairest approach is a direct, private conversation that gives the student an opportunity to explain their process. Frame it as a question about their work, not an accusation. Ask them to walk you through how they produced a specific section — the evaluation, for example. A student who wrote the work themselves will typically be able to explain their reasoning, name the sources they consulted, and describe choices they made. A student who submitted AI-generated text is more likely to struggle with specifics.
Keep a brief record of this conversation. If the student confirms they used AI assistance, refer to your school's academic integrity policy for next steps, which may include resubmission, a reduced mark, or a formal report to the exam board depending on the stage of the NEA. If the student disputes the finding, make clear that they have the right to appeal and that no final action has been taken on the basis of the score alone.
It is worth reading the JCQ guidance on malpractice before having this conversation, as it sets out the responsibilities of the centre (your school) and the candidate. GradeOrbit's detection output can be included as supporting evidence in any formal report, alongside your own professional assessment and the record of the student conversation.
Using GradeOrbit's Detection Tool on Chemistry NEA
GradeOrbit accepts uploaded PDFs, scanned images, and JPEG or PNG files. For typed Chemistry NEA submissions, a PDF export is the cleanest input. For handwritten work, a scanned page or a photo taken with a mobile device works well — GradeOrbit's document processing can read handwritten scientific text, though typed text typically produces a more reliable score.
Upload the submission, select the AI detection tool, and choose between the 1-credit and 3-credit model depending on document length. The likelihood score is returned within a couple of minutes. No student work is stored after processing — once the detection is complete, the uploaded document is not retained, which means you can run detection confidently without concerns about data protection compliance.
For further context on how to document your detection decisions and build them into a professional record, see our guide on how teachers document AI detection decisions for records.
Try GradeOrbit's AI Detection Tool
GradeOrbit gives A-Level Chemistry teachers a structured, privacy-safe way to run AI detection on NEA submissions and extended writing. The likelihood score supports your professional judgment — it does not replace it. Upload a submission, get a score, and follow your school's process with confidence.
Visit the GradeOrbit homepage to explore the detection tool and see how it fits into your marking workflow.