What JCQ Actually Says About AI Detection in Coursework: The Checklist

JCQ does not require you to use an AI detector, does not approve any particular one, and does not treat a detector result as proof of anything. Its published guidance puts detection tools in a supporting role, as one source of evidence inside a wider judgement made by the teacher who knows the student. The obligations that actually carry weight sit elsewhere: on the centre to have a policy, on the student to acknowledge AI use and keep the evidence, and on the teacher to authenticate only work they are confident is the student's own.

Everything below is quoted from the current guidance, AI Use in Assessments: Your role in protecting the integrity of qualifications, published by JCQ on 26 April 2023 and now at revision two, dated 30 April 2025. The hub page for it, with the teacher and student support materials, is on the JCQ malpractice section.

Which assessments this applies to

The guidance is aimed mainly at the assessments where students can reach the internet rather than at examinations, though exam-room device setup is still in scope wherever a candidate could reach an AI tool. In JCQ's words:

There are some assessments in which access to the internet is permitted in the preparatory, research or production stages. The majority of these assessments will be Non-Examined Assessments (NEAs), coursework and internal assessments for General Qualifications (GQs) and Vocational and Technical Qualifications (VTQs).

For exams, the position is simply that students must not be able to use the tools at all, with a specific warning about laptops:

care must be taken when a student is allowed to use a laptop or similar device for exams, to ensure they have no access to AI tools

What counts as AI misuse

The definition is narrower than most staffroom conversations assume. It turns on two things together: whether the use was acknowledged, and whether the work submitted is genuinely the student's own.

AI misuse is where a student has used one or more AI tools but has not appropriately acknowledged this use and has submitted work for assessment when it is not their own.

The examples JCQ gives include copying or paraphrasing AI content so the work is no longer the student's own, using AI to complete parts of the assessment, "Failing to acknowledge use of AI tools when they have been used as a source of information", and "Incomplete or poor acknowledgement of AI tools".

The consequences are set out plainly:

The malpractice sanctions available for the offences of 'making a false declaration of authenticity' and 'plagiarism' include disqualification and debarment from taking qualifications for a number of years.

What students must do when they use AI

Acknowledgement has a specific form, and it comes with an evidence requirement that most schools underuse:

Where AI tools have been used as a source of information, student acknowledgement must show the name of the AI source used and the date the content was generated.

The student must retain a copy of the question(s) and computer-generated content for reference and authentication purposes, in a non-editable format (such as a screenshot) and provide a brief explanation of how it has been used.

JCQ is also clear that a vague reference is not enough:

it would be unacceptable to simply reference 'AI' or 'ChatGPT', just as it would be unacceptable to state 'Google' rather than the specific website and webpages which have been consulted

And that where AI produced the content, that content earns no credit against the marking criteria, even when it is properly acknowledged:

Students are also reminded if they use AI they have not independently met the marking criteria therefore they will not be rewarded.

What JCQ says about AI detection tools

This is the section most schools have not read closely. JCQ names four programs as examples: Copyleaks, GPTZero, Sapling and Turnitin AI writing detection. It then says what the list is and is not:

The list of certain suppliers of AI-related products is for information purposes only and does not constitute an endorsement by JCQ.

On what the tools can and cannot do:

it should be noted that the above tools will give lower scores for AI-generated content which has been subsequently amended by students, as they base their scores on the predictability of words

AI detection tools, including those listed above, employ a range of detection models which vary in accuracy depending on the AI tool and version used, the proportion of AI to human content, prompt types and other factors (such as an individual's English language competency)

JCQ is right about amended content, and we can put numbers on it. In August 2026 we
tested our own detector against AI essays written for real assignment prompts, then against
the same essays after a student had edited them. On unedited copy-paste output we flagged
about 70 per cent. After light rewording, about 67 per cent. With typos and contractions
added, half. On a submission that was half AI and half the student's own writing, under a
third. And when the model was simply asked to write like a 15-year-old, about one in five.
The false-positive rate on real student writing was held constant at 2.4 per cent throughout,
so those are like-for-like figures.

We publish that because a school cannot make a fair decision on a number it has not been
given, and because it is the specific failure mode JCQ warns about. It also means the
sentence "the detector said 20 per cent AI" carries almost no information on its own. What
carries information is the pattern across a student's work over time. JCQ puts the teacher's
own knowledge of the pupil first and the tool second, and that ordering is correct.

That last clause matters more than its length suggests. JCQ is acknowledging that a student's English language competency changes the result. Independent research has found the same effect and the reason a flag on an EAL student's work deserves more scrutiny, not less. Our own note on what false positives mean for teachers covers which groups are affected and why.

On how a result should be used:

The use of detection tools, where used, should form part of a holistic approach to considering the authenticity of students' work; all available information must be considered when reviewing any malpractice concerns.

Teachers will know their students best and so are best placed to assess the authenticity of work submitted to them for assessment

AI detection tools can be a useful part of the evidence they can consider.

And where a concern already exists:

In instances where misuse of AI is suspected it may be helpful to use more than one detection tool to provide an additional source of evidence about the authenticity of student work.

Read together, that is a description of evidence gathering, not proof. A detector output is one item you put on the table next to the drafts, the classwork and the conversation.

What teachers must not do

Two hard limits. The first is on marking, quoted in the guidance from the Instructions for Conducting Coursework:

Teachers must not use artificial intelligence as the sole means of marking candidates' work

an AI tool cannot be the sole marker. A human assessor must review all the work in its entirety and determine the mark it warrants, regardless of the outcomes of an AI tool.

The second is on authentication:

Teachers must confirm that all of the work submitted for assessment was completed under the required conditions and that they are satisfied the work is solely that of the individual candidate concerned. If they are unable to do so, the work must not be accepted for assessment.

When to report, and when not to

The dividing line is the declaration of authentication, and getting this wrong creates unnecessary awarding body referrals:

If a student has not signed the declaration of authentication, centres do not have to report the incident to the appropriate awarding organisation.

If AI misuse is detected or suspected by the centre and the declaration of authentication has been signed by the student, the case must be reported to the relevant awarding organisation.

Before the signature, it is yours to handle under your own policy. JCQ says so directly: the centre is responsible for determining next steps and "a teacher/assessor should not refer the work to the awarding organisation for a decision".

The checklist

Work through this once a year with the malpractice policy open, and once per cohort with the teaching team.

• Check your malpractice policy names AI. JCQ requires a policy covering "what it is, when it may be used and how it should be acknowledged, the risks of using AI, what AI misuse is and how this will be treated as malpractice". If AI appears nowhere in your policy document, that is the first gap to close.

• Write down how teachers will authenticate work. The guidance asks centres to review the policy to acknowledge "how teachers will authenticate work". A named method beats an assumption when a case is challenged.

• Put referencing rules in writing, including AI. The policy needs "clear guidance on how students must reference appropriately (including websites)" and separately on how students acknowledge AI use.

• Issue the Information for Candidates document and check it landed. Centres must ensure each student "is issued with a copy of, and understands" it, which means a register entry, not a pile on a desk.

• Reinforce what the declaration means. JCQ asks centres to reinforce "the significance of their declaration where they confirm the work they submit is their own, the consequences of a false declaration". A two minute explanation before signing is worth more than a paragraph in a handbook.

• Tell students the evidence rule before they start, not after. Name of the AI tool, date generated, a non-editable copy of the prompts and output, and a short note on how it was used, submitted with the work. Students who learn this after the deadline cannot comply retrospectively.

• Brief staff on AI tools and detection tools. Centres must "Ensure teachers and assessors are familiar with AI tools, their risks and AI detection tools". Familiarity includes knowing the failure modes, not just the login.

• Lock down devices where access is prohibited. Staff should know "how to disable improper internet/AI access where this is prohibited", and JCQ flags exam laptops specifically.

• Design tasks that are harder to generate. The guidance suggests assignments that are "topical, current and specific, and require the creation of content which is less likely to be accessible to AI models trained using historic data".

• Build in supervised checkpoints. Allocate time for portions of work in class under direct supervision, and "Examine intermediate stages in the production of work" so the final piece is a visible continuation of earlier drafts.

• Plan for private candidates separately. JCQ says verifying their work "can be more challenging", and asks centres to decide before accepting the entry what supervision, portfolio review or discussion will make authentication possible.

• Compare against previous work first, before any tool. JCQ's list of what to look at is spelling and punctuation, grammatical usage, writing style and tone, vocabulary, complexity and coherency, general understanding and working level, and the mode of production.

• Use the indicator list as prompts, not verdicts. It includes "A default use of American spelling, currency, terms and other localisations", references that cannot be verified, a lack of local or topical knowledge, and "Overly verbose or hyperbolic language that may not be in keeping with the candidate's usual style". Each one has innocent explanations, so treat them as questions to ask.

• If you use a detector, use it as one input and record it as one input. Note what the tool returned, what the tool is known to miss, and what else you weighed. If the concern is serious, JCQ suggests a second tool as an additional source of evidence.

• Have the conversation. A short verbal discussion about the work, which JCQ lists as a preventive measure, is usually the most informative thing you can do and the easiest to record fairly.

• Check whether you are allowed to paste the work into that tool at all. Student work is personal data and your centre is the controller, so retention and model training terms matter before the scan, not after. We cover the questions to ask a supplier in what happens to text you paste into an AI detector.

• Apply the declaration test before escalating. Not signed, handle it internally under your policy and record it. Signed, report it to the awarding organisation.

• Keep records where AI use affected the mark. JCQ asks for clear records where acknowledged AI use has influenced marks, because that is what protects you at moderation or internal appeal.

Being honest about the limits

JCQ does not endorse a detector and neither should anyone selling one. Detectors produce false positives, and they do it unevenly: formal, careful, heavily edited prose is flagged more often than casual writing, and English language competency shifts the result, which JCQ itself acknowledges.

No tool on JCQ's list of examples has been approved by JCQ, because JCQ approves none of them. Whatever you use is one input a teacher weighs alongside the drafts, the classwork and the conversation.

The uncomfortable part of the JCQ position is also the fair part: the judgement stays with the person who knows the student. A tool cannot carry that for you. If a case is challenged, the tool is not what holds up.

---

Who wrote this, and what we sell. Is It AI is an AI-writing detector, so read the section above with that in mind. We are not on JCQ's list of example programs, and being on it would not mean approval. Three pieces of our own work sit behind what is written here, and you are welcome to check them: what we found testing British English for false positives, how often detectors flag human writing, where our current model wrongly flagged 4 of 400 real student essays, and a comparison of detectors for schools if you are running a procurement. If what you actually need next is the policy itself, we have a school AI-use policy template built around the JCQ wording above.

Frequently asked questions

Does JCQ require schools to use an AI detector?

No. The guidance says detection tools "may be used as a check on student work and/or to verify concerns about the authenticity of student work", and refers to their use "where used". What JCQ does require is that teachers and assessors are familiar with AI detection tools, that the centre has a malpractice policy covering AI, and that work is authenticated before it is accepted.

Does JCQ approve or recommend any AI detector?

No. It names Copyleaks, GPTZero, Sapling and Turnitin AI writing detection as examples, then states that the list "is for information purposes only and does not constitute an endorsement by JCQ" and that verifying any supplier is the centre's responsibility. Any vendor claiming JCQ approval is misrepresenting the guidance.

Can an AI detection score be used as proof of malpractice?

No. JCQ places detection inside "a holistic approach to considering the authenticity of students' work" where "all available information must be considered". A score is evidence to weigh, alongside previous work, draft history, supervised classwork and a discussion with the student. The teacher who knows the student is described as best placed to judge.

Does a student have to declare AI use in coursework?

Yes, where AI has been used as a source of information. The acknowledgement must show the name of the AI source and the date the content was generated, and the student must retain the prompts and output in a non-editable format with a brief explanation of how it was used. A generic reference to "AI" or "ChatGPT" is not sufficient. The student-facing version of this is covered in is it safe to use AI for my essay.

What if I suspect AI misuse before the student has signed the declaration?

Handle it in the centre. JCQ says centres do not have to report incidents to the awarding organisation where the declaration of authentication has not been signed, and that the centre determines next steps rather than referring the work for a decision. Once the declaration has been signed, a suspected case must be reported.

Can teachers use AI to help mark coursework?

Only with a human assessor doing the real marking, and only after the centre has considered data privacy. JCQ says an AI tool cannot be the sole marker, that a human assessor must review all the work in its entirety and determine the mark regardless of the AI output, and that the assessor remains responsible for the mark awarded.

Does this apply to private candidates?

Yes. The requirement that work is the candidate's own applies to internal and private candidates alike, and JCQ asks centres to decide in advance what supervision, portfolio review or discussion will let a teacher authenticate a private candidate's work before accepting the entry.

Sources

• Joint Council for Qualifications, AI Use in Assessments: Your role in protecting the integrity of qualifications, published 26 April 2023, revision two 30 April 2025. All quotations above are from this document, including the extracts it reproduces from the Instructions for Conducting Coursework in Appendix C.

• JCQ, AI use in assessments hub page, which carries the guidance plus the teacher information sheet and student poster.

• JCQ, Suspected Malpractice: Policies and Procedures, the reporting route referenced throughout the AI guidance.

• JCQ, Coursework and non-examination assessment documents, where the current Instructions for Conducting Coursework and NEA documents are published.