How Exam Boards Check for AI in A-Level English Coursework
A-Level English is an awkward subject in which to spot AI-written coursework. Students are expected to write in a sophisticated, analytical register - and AI tools like ChatGPT and Claude are now capable of producing exactly the kind of carefully argued, textually referenced prose that exam boards reward. The result is that the usual visual signals of AI-generated writing - overly formal phrasing, suspiciously smooth structure - can easily be mistaken for genuine student achievement, especially in a subject where good writing is the entire point.
What follows covers the signals specific to A-Level English work, how automated detection tools work and what their scores mean, and how to use them as one part of an evidence-based approach to academic integrity.
Why A-Level English Is a High-Risk Subject
NEA components - the non-examined assessments that form part of AQA, Edexcel, and OCR A-Level English specifications - are produced independently over an extended period. Students research, draft, and submit written work without the time pressure of a formal examination. This freedom is educationally valuable, but it also creates the conditions under which AI assistance is most tempting and most difficult to detect.
The stakes are high. AQA's A-Level English Language and Literature NEA, for example, requires students to produce original writing alongside a critical commentary. Edexcel's coursework components involve extended literary analysis. OCR expects students to demonstrate independent critical thought across extended essays. In every case, the skills being assessed - argumentation, textual analysis, personal voice - are precisely the skills that modern AI tools can convincingly simulate.
Teachers working with sixth form students also face the complicating factor that students at this level are often genuinely developing sophisticated writing abilities. A student who has improved significantly between Year 12 and Year 13 may produce work that seems unfamiliar, and it is important not to conflate genuine progress with AI use. The task is to distinguish between development and fabrication - and that requires care.
What AI-Generated A-Level English Writing Looks Like
There are signals that tend to appear in AI-generated English analysis, distinct from the more general patterns you might look for in other subjects.
Generic or Non-Specific Textual Evidence
A-Level English marking depends heavily on close reading. Students are expected to engage with specific words, phrases, and structural choices and explain their effect in detail. AI tools often produce analysis that gestures at textual evidence without truly embedding it - quoting a line and then making a general claim about tone or theme rather than working through the language precisely. The analysis can read as plausible without being genuinely close. If a student's work reads like a competent summary of a critic's view rather than an independent engagement with the text, that is worth noting.
Overconfident Argumentation
Human A-Level essays typically contain moments of genuine uncertainty - hedged claims, acknowledgment of alternative readings, moments where the student is working something out on the page. AI-generated analysis tends to assert with confidence throughout. Every paragraph lands cleanly. Every argument resolves neatly. This kind of unfailing coherence can be a red flag, particularly in a student whose previous work has shown the normal messiness of developing critical thought.
Formulaic Critical Framework Use
AI tools apply critical frameworks - feminist, Marxist, post-colonial - in a formulaic way that names the framework and applies it mechanically rather than interrogating how it genuinely illuminates the text. A student who can name Foucauldian discourse theory but cannot explain in conversation what it actually reveals about their chosen text may have borrowed more than they should have.
Absence of a Personal Critical Voice
Sixth form English work, particularly at A-Level, should carry a genuine critical voice - a sense of what the student actually thinks about the text and why. AI-generated analysis is generically competent. It is never surprised, never tentative, never genuinely invested. If the writing reads like a very good study guide rather than a student's own encounter with a difficult text, that quality of detachment is worth paying attention to.
How GradeOrbit's AI Detection Tool Works
GradeOrbit's AI detection is part of its school accounts, and it does not ask the generic question "does this read like AI?". Instead it compares the NEA with that student's own earlier writing already on file - essays and timed work the department has marked in GradeOrbit - and returns a score from 0 to 100 for how far the new piece departs from it. A low score means the essay is consistent with how the student usually writes; a high score means it differs in ways worth a closer look. The result also includes a confidence label (Low, Medium, or High), a list of the signals it noticed, and a short reasoning paragraph explaining how the score was reached. A student with no earlier work on file is skipped, and you are told why.
You can run a check from a student's profile or across a whole class. Typed coursework goes in as a PDF (export Word documents to PDF first) and handwritten drafts as scans or photos, which the AI reads straight from the page images. For a class set, bulk upload splits the pages by the number of pages per student and matches the name on each cover page to your class list, asking you to pick the student if a name is unclear. Unlimited AI detection is included in every teacher seat on a GradeOrbit school account, with no per-check charge and no tiers to choose between, and results appear on each student's profile.
How student data is handled depends on your account type: on a school account, work is held under your school's Data Processing Agreement (DPA) with students' real names, so detection can compare each piece against that student's earlier writing on file. Uploaded scans are removed when a run archives (90 days by default, adjustable by the school from 7 to 365 days), and student records follow the school's retention window and are wiped at offboarding. By contrast, on an individual teacher account where no school DPA is in place, uploaded student work for marking is redacted client-side in your browser, labelled anonymously, and never stored (processed and discarded immediately). Across every tier, student work is never used to train AI models.
Understanding and Using Likelihood Scores
The most important thing to understand about AI detection scores is that they are probabilistic, not definitive. A high score does not prove that AI was used; it means the text shows patterns consistent with AI generation. A low score does not clear a student of suspicion; it means the patterns are more consistent with human writing, but a student who has carefully edited AI output may produce work that scores lower than genuinely AI-generated text.
False positives are a real concern in English specifically. Some highly proficient writers - students who have genuinely developed a formal, precise analytical style - can produce text that detection tools flag at elevated levels. This is one reason why teacher knowledge of the student remains the most important input in any integrity assessment.
The most defensible approach is to treat the detection score as one strand of evidence alongside your knowledge of the student's prior work, the consistency of this piece with their usual register and ability, and the specific signals the tool has identified. When multiple strands converge - a high score, unfamiliar writing style, generic textual engagement, and a student who cannot fluently discuss their own argument - you have a basis for a proper conversation. A score alone, without corroborating evidence, is not sufficient grounds for a formal integrity concern.
What to Do When You Have Concerns
If a detection score is high and you have other reasons for concern, the most productive immediate step is a short conversation with the student. Ask them to walk you through their argument, explain a specific analytical choice, or write a short analytical paragraph on the same or a related text in class. A student who has genuinely produced the work will be able to discuss it fluently and with the same depth of understanding the essay displays. A student who has submitted AI-generated content will typically struggle to reproduce that fluency without the AI in front of them.
Before taking any formal action, check your school's academic integrity policy. Many schools are still developing their approach to AI, and the guidance may distinguish between AI as a research or drafting aid versus submitting AI output as original work. For NEA components, check the relevant exam board's guidance directly - AQA, Edexcel, and OCR have each issued updated guidance on AI use in assessed work, and the obligations on teachers and centres are evolving.
For a broader guide to handling AI detection scores across different scenarios, the post on how to interpret AI detection scores covers the full range of responses in more detail. For the general principles of AI detection in school contexts, AI detection for teachers provides a useful foundation.
How Do Exam Boards Check for AI?
This is the question teachers and students ask most, and the honest answer surprises people: the first and most important check is not an exam-board scanner, it is you. Under JCQ's guidance on AI use in assessments, which AQA, Edexcel and OCR all follow, the centre is responsible for authenticating NEA before the student and teacher sign the declaration. Concerns raised before that point can be resolved within the school; once the declaration is signed, a suspicion has to be reported to the awarding body as possible malpractice. Moderators can and do raise concerns about authenticity when they review a sample, but a school should never assume a board will catch what it did not check.
JCQ is also explicit that detection tools are one source of evidence, not proof, and that staff should not rely on them alone. In practice that means the checks that carry weight are the ones only a teacher can make: comparing the NEA with the student's timed writing, looking at drafts and planning, and asking the student to talk through their own argument.
Exam Board Context: AQA, Edexcel, and OCR
All three major A-Level English exam boards have issued guidance acknowledging that AI use in coursework is a live concern. The general position across AQA, Edexcel, and OCR is that using AI to generate assessed work and submitting it as your own constitutes malpractice, in the same category as plagiarism. Teachers are expected to authenticate student work to the best of their ability, and centres have a responsibility to report suspected malpractice to the relevant awarding body.
The practical challenge is that authentication is difficult. Unlike plagiarism, there is no database of AI-generated text to compare against. Detection tools provide supporting evidence, but the final judgment is always a professional one. Schools that have introduced clear policies on AI use - including explicit guidance to students about what is and is not permitted - are better placed to manage this, both because students understand the boundaries and because teachers have a clearer framework for their own judgments.
The Point of Checking
AI detection in A-Level English is not about catching students out. It protects the integrity of the NEA and makes sure that students who have genuinely developed their analytical abilities get the credit for it. A detection score adds one more piece of evidence to what you already know about how each student writes; the drafts, the timed work and the conversation carry more weight.
Detection is included in each teacher seat on a GradeOrbit school account, and schools can start with a free school trial.
Further reading: detecting AI in an A-Level EPQ and how do teachers check for AI?