COMPLIANCE · DEFINITIONS
AI Marking vs AI-Assisted Marking: What’s the Difference (and Why It Matters for Compliance)
Marketing treats these two phrases as interchangeable. They describe structurally different things, and only one fits inside Ofqual’s current rules.
By Eduface · July 2026 · 8 min read
Product marketing treats these two phrases as interchangeable, and that’s a genuine problem, because they describe two structurally different things, and only one of them fits inside Ofqual’s current rules. Learning to spot which one you’re actually looking at, past the label on the homepage, might be the single most useful thing you do before signing a contract.
The distinction in one line
AI marking releases a final grade with no human decision in between. AI-assisted marking produces a draft that an assessor reviews, can change, and has to explicitly approve before it counts. Only the second fits Ofqual’s rules, no matter how good the underlying accuracy is.
The distinction, stated plainly
AI marking, in the strict sense, means the AI produces a final grade and that grade is released without anyone making a human decision in between. The system evaluates the work, and the number that comes out is the number that counts.
AI-assisted marking means the AI produces a draft, a proposed grade with reasoning attached, that an assessor reviews, can change, and has to explicitly approve before it becomes final. The grade that counts is the assessor’s decision, informed by the AI, not the AI’s output on its own. Written out like that, the difference sounds almost too obvious to need explaining. In practice it’s constantly blurred, partly by accident and partly because AI-assisted sounds a lot less exciting on a product page than AI marking.
Why the name on the box won’t tell you which one you have
Vendors use AI marking, AI grading, automated marking, and AI-assisted marking fairly interchangeably, usually chosen for how they sound rather than what they precisely describe. A tool marketed with the more dramatic label can, underneath, have a genuinely mandatory review step built in. Equally, a tool marketed with careful, reassuring language about supporting the assessor can, in its default configuration, release a grade automatically unless someone actively intervenes to stop it.
This is why the useful test has nothing to do with the name and everything to do with the workflow. What actually happens, step by step, between the AI generating an output and a learner seeing a grade? If there’s a mandatory human decision point somewhere in that sequence, you’re looking at AI-assisted marking, whatever the label says. If there isn’t, you’re looking at AI marking in the strict sense, and that puts you on the wrong side of where Ofqual currently draws the line, regardless of the marketing.
Why this single distinction carries so much weight
Our companion pieces on Ofqual’s marking rules and on human-in-the-loop marking go into more depth, but the short version is that this distinction is, more or less, the entire compliance question in miniature. Ofqual’s concern was never that AI is involved in marking at all. It’s that the decision that actually determines a learner’s result has to involve human judgement, not just human presence somewhere nearby. AI-assisted marking, properly built, satisfies that. AI marking, in the strict sense, doesn’t, no matter how good its underlying accuracy happens to be.
That reframes this as less a matter of terminology and more a matter of what you’re legally able to deploy. A provider using a genuinely AI-assisted tool, with mandatory review built into the workflow, is on solid ground. A provider using something marketed the same way, but structurally releasing grades without that step, has a compliance problem it probably didn’t intend to create.
A few things worth knowing about how this shows up in the wild
Multiple independent AI passes checking each other’s work is not the same thing as human oversight. Some tools run several evaluation passes that flag disagreement between them and reconcile the result, which is a genuinely useful design for internal consistency. It reduces how often a human needs to step in with real force, but the final sign-off still has to be a person’s decision, not the reconciliation step’s.
Available for review is not the same as requires review. A grade a human could look at, if they chose to, is not equivalent to a grade a human must look at before it counts. That gap matters most in a busy marking window, which is precisely when an optional step is the first thing to get skipped.
Feedback and grading aren’t the same claim. A tool that generates written comments without producing a score at all sits outside this question entirely, since there’s no mark being awarded yet. That connects to a point formative assessment researchers have made for a long time: feedback that helps a learner improve and a grade that certifies attainment serve genuinely different functions, and conflating them tends to weaken both (Nicol & Macfarlane-Dick, 2006). Once a tool moves from suggesting comments to proposing a score, the AI marking versus AI-assisted marking question applies in full.
The two questions worth actually asking a vendor
Skip the label entirely and ask these directly. Can a grade reach a learner without an assessor taking an explicit approval action, under any configuration or default setting? And if the answer is genuinely no, what does the assessor actually see before approving: a single number, or reasoning they can meaningfully evaluate and disagree with? The first question tells you whether you’re looking at AI marking or AI-assisted marking. The second tells you whether the review step is real or just there for show. Neither is answered by whatever the product happens to be called.
Frequently asked questions
Is AI-assisted marking allowed, even though AI marking in the strict sense isn’t?
Yes. Ofqual’s concern is with AI acting as the sole marker. AI-assisted marking, where a human reviews and approves the final grade, is the model that fits within current rules.
How do I tell which one a tool actually does, given how loosely the terms get used?
Ask what happens between the AI’s output and the learner seeing a grade. A mandatory human approval step means AI-assisted marking. A grade that can be released without one means AI marking in the strict sense, whatever it’s called.
Does using multiple AI passes to cross-check itself count as human oversight?
No, not on its own. It can improve internal consistency, but the final decision still needs to be a human’s, not the AI reconciling with itself.
What’s the single most important question to ask a vendor?
Whether a grade can ever reach a learner without an assessor’s explicit approval, under any setting or default. If the answer isn’t a clean no, treat it as AI marking, not AI-assisted marking.
Sources
Ofqual (2026). Using AI in Marking: Why Technical Capability, Fairness, and Transparency All Matter. Ofqual blog, gov.uk.
Nicol, D. J., & Macfarlane-Dick, D. (2006). Formative Assessment and Self-Regulated Learning: A Model and Seven Principles of Good Feedback Practice. Studies in Higher Education, 31(2), 199-218.
Santoni de Sio, F., & van den Hoven, J. (2018). Meaningful Human Control over Autonomous Systems: A Philosophical Account. Frontiers in Robotics and AI, 5, Article 15.
Deploy AI-assisted marking, not AI marking
Eduface builds the mandatory review step into the workflow: every grade is a draft until an assessor approves it. Book a demo or start free.