The ApprenticeRoute

AI and apprenticeship coursework: where the line is

Rules and figures checked: 2026-07-25. Funding, tenancy and tax rules move — verify before relying on them.

Let's start with the thing everyone knows and nobody writes down: essentially all students now use generative AI for their coursework in some form.

The Higher Education Policy Institute's 2026 student survey, published in March, found 94% of full-time UK undergraduates using generative AI to help with assessed work, and 95% using it in some way. Two years earlier the equivalent figure was 53%. Twelve per cent now say they put AI-generated text directly into assessed work, up from 8% the year before. Two-thirds say assessment on their course has changed significantly in response.

So the question "should I use AI for my coursework?" is not really the live one. The live questions are how, declared or not, and what does it cost you — and for a degree apprentice there's a fourth that campus students don't have to think about at all, which is the one I'd put first.

I'm going to be honest that this is a moving target. The sector's position has shifted every year since I started, and the quality body for UK higher education has work in progress on exactly this. Check your own institution's current rules rather than trusting a web page.

Rule one: the rule is per-assessment, not per-university

The most common way people get into trouble is assuming there's a single policy.

There usually isn't, in any usable sense. Your university will have an overarching academic integrity framework, and then individual modules and individual assessments set what's permitted within it. Different lecturers on the same course routinely permit different things. That's not disorganisation, it's deliberate — a coding assignment and a reflective essay are testing different capacities, and the same AI use that's fine in one guts the other.

Lots of institutions have made this explicit with tiered systems. Several use a traffic-light scheme — typically something like no AI permitted, AI permitted for specified purposes with declaration, AI use encouraged and integral. Others have moved to a simpler two-lane split between assessments that are secured against AI and assessments that assume it. The University of Bath, for instance, announced a shift from traffic lights to a two-lane approach for 2026–27. Expect the labels where you study to be different again.

The practical instruction is boring and it is the whole ballgame: read the AI statement on each individual brief, every time. If the brief doesn't say, ask the module leader and get the answer in writing. The QAA's own work with students found that what students are most anxious about isn't the rules being strict — it's not knowing what they are, because they're communicated inconsistently. Asking is not a confession. It's the correct move.

Declaring it

Where AI use is permitted, you'll usually be required to say what you used and how, sometimes in a required declaration or a short methods note.

Do this properly, because failing to declare permitted use is itself frequently a misconduct offence. That trips people up — they reason that since the use was allowed, the paperwork is a formality. It isn't. The declaration is how the institution maintains the integrity of what it's certifying, and an undeclared permitted use looks identical, from the outside, to a concealed prohibited one.

Be specific. "Used a language model to generate practice questions on chapter 4, and to check the clarity of my introduction; all analysis and writing my own" is a real declaration. "AI was used" is not.

What actually happens with detection

Worth understanding, because the mythology runs in both directions.

Universities do use detection tools, and they generally do not treat a detection score as proof of anything. The tools produce probabilistic outputs with known false positives, and the established practice is that a flag triggers a human review — of the work itself, of the assessment's rules, of your drafting history, and of your own account of how you wrote it — rather than an automatic penalty.

That's the reassuring half. The unreassuring half is that being investigated is genuinely unpleasant even when you're innocent, it takes weeks, and there's a live fairness debate about who gets flagged: non-native English speakers and students who write in a plain, formulaic style are disproportionately caught up in it. Sector commentary through 2025 and 2026 has been increasingly blunt about that.

The defence against a false accusation is process, not indignation. Keep your drafts. Work in something with version history. Keep the notes and the reading you did. If you can show a document evolving over three weeks, with your workplace examples appearing in it before the polished sentences do, the conversation is short. If your entire assignment appeared in one sitting the night before the deadline, it's longer — which is another argument for submitting early.

The apprentice problem nobody warns you about

Now the part that's specific to us, and it is genuinely the most serious thing in this article.

Do not paste your work into a public AI tool.

Your assignments are built from your actual job — real projects, real clients, real internal processes. That's the whole design of a work-based degree. And a consumer AI service is an external third party. Putting internal material into one is, depending on your employer and your sector, a data incident: a disclosure of confidential information to an outside organisation, made by you, in writing, with a timestamp.

This is not an academic issue. It's an employment one. It sits at the intersection of the confidentiality clause and the IT policy in your contract, and it's the kind of thing that gets treated seriously in regulated sectors regardless of how innocent the intent was.

What to do instead:

I'd rate this as the highest-consequence paragraph on this entire site. An academic misconduct finding is bad. A confidentiality breach at a regulated employer is a different order of problem.

Three more reasons it works badly for us specifically

Your assignment is about a workplace the model has never seen. The standard work-based brief is "apply this framework to a situation in your organisation". A language model has no access to your organisation, so what it produces is generic — plausible, fluent, and unmoored from anything real. Markers on apprenticeship programmes read a lot of these, and the tell isn't fancy vocabulary, it's the absence of specifics. The bit that earns marks is the bit only you can supply.

Reflective writing is the one format AI cannot do for you. A reflective assignment asks what you did, what you misjudged, and what you'll change. Outsourcing that produces the exact failure mode markers are trained to spot — the smooth, general, feelings-adjacent paragraph that describes no actual event. If you're going to hand-write one thing on this degree, make it the reflection.

You have to defend it out loud. Apprenticeships end in an independent assessment against your standard, and that assessment usually includes a professional discussion — a structured conversation with an assessor about your own work and evidence. There's no version of that you can prepare a document for. Whatever you didn't learn, you find out you didn't learn there, in a room, with someone qualified asking follow-up questions.

Worth knowing that this whole end of the system is being reformed: from 2025 through 2027, end-point assessment is being replaced by "apprenticeship assessment", assessment can happen at points during the programme rather than only at the end, end-point assessment organisations become "assessment organisations", and Ofqual introduced a new regulatory framework for it in spring 2026. Existing apprentices continue on their current version until a revised one goes live. The direction of travel — assessment that is more distributed and harder to game in a single sitting — is not accidental.

Where it's genuinely useful, and defensible

I'm not making an abstinence argument. Used in the right places it's a real advantage for someone doing a degree in fifteen hours a week, and the uses students report most are mostly the legitimate ones.

Things I'd defend without hesitation, subject to the brief permitting them and to the confidentiality rules above:

The common thread: it's operating on material you've already engaged with, and the thinking stays yours.

Where it hollows you out

The uses I'd avoid even where they're technically permitted, because of what they cost rather than what they risk:

And here's the specifically apprentice version of the cost, which is why I'd take it more seriously than a campus student might. Your degree is not the point. It's a four-year proxy for a set of capabilities your employer is going to expect you to actually have, in a job you're already doing, in front of people who will notice. A campus graduate who coasted through a module can pick a career where it doesn't matter. You'll be at the same desk on Monday.

That's not a moral argument, it's a practical one. The gap between "I have a degree in this" and "I can do this" is invisible for about eighteen months and then extremely visible.

The test I'd actually use

Forget the policies for a second. One question covers most cases:

Could you defend this, out loud, right now, without the document in front of you?

If someone asked you to explain your argument, justify the framework you chose, and say what you'd do differently — could you? If yes, whatever tools you used along the way, you've done the learning and the marks are honestly yours. If no, you've produced a document rather than acquired a capability, and on this route those come apart in a way they don't elsewhere: at your professional discussion, at your next performance review, and at the point where somebody asks you to do the thing for real.

Read the rules on every brief. Declare what you use. Never put work material into a public tool. And keep the thinking.