Formative assessment happens during learning to improve it; summative assessment happens at the end to judge it. A draft you mark with feedback so the learner can revise is formative. The final submission that decides whether they pass is summative. Same piece of work can play either role - what changes is the purpose: one feeds back into learning, the other counts toward a result. Get the two confused and you either grade people on things they were still learning, or you give feedback on a decision that is already locked.
The core difference
The clearest way to keep them straight is a single question: does this assessment change what happens next for the learner, or does it record where they landed? Formative assessment is forward-looking. Its output is information the learner and teacher act on - a misconception spotted, a skill not yet solid, a next step. Summative assessment is backward-looking. Its output is a judgement - a grade, a competency decision, a pass or fail - that goes on the record.
The metaphor people use is the cook tasting the soup (formative) versus the diner tasting the soup (summative). The cook tastes to adjust; the diner tastes to judge. Both matter, but you cannot fix the soup once it is on the table.
What formative assessment looks like
- Low or no stakes. A quick check of understanding, a draft, a practice problem, a peer review. Mistakes here are the point, not a penalty.
- Fast, specific feedback. The value is in what the learner does next, so the feedback has to arrive in time to act on and be precise enough to act on.
- Frequent. Little and often beats one big check. You are sampling progress continuously, not measuring it once.
Examples: exit tickets, a marked first draft, a mock interview with notes, a self-assessment against the rubric before final submission, a coach reading an early reflection and steering it.
What summative assessment looks like
- Consequential. It contributes to a grade, a qualification, a hiring decision, or a certification. The result counts.
- Held to a fixed standard. Because it is on the record, it has to be consistent and defensible - the same standard applied to everyone, with evidence behind each judgement.
- Less frequent, higher weight. End of unit, end of course, final project, certification exam.
Examples: a final case study marked against a rubric, a competency decision in vocational training, a certification assessment, an end-of-programme capstone.
When to use each
You almost always want both, and in sequence. Use formative assessment all the way through so learners arrive at the summative point ready, and so you can intervene while there is still time. Then use summative assessment to make the judgement that has to be made - cleanly, against a known standard, with the feedback work already done earlier rather than crammed onto a result that cannot change.
A common failure is treating every assessment as summative: grading practice as if it were the final, which makes learners hide their mistakes instead of surfacing them. The opposite failure is treating a high-stakes decision as casually as a practice check - no consistent standard, no evidence, no defensibility. Match the rigour to the role.
Where AI marking fits
AI marking is useful in both modes, for different reasons. Formatively, it lets you give every learner detailed, evidence-based feedback on a draft without a human reading every word - so feedback that used to be rationed becomes routine. Summatively, it applies the same rubric consistently across a whole cohort and shows its working, which is exactly what a defensible final judgement needs - provided a qualified person reviews and signs off, because a consequential result should never be solely automated.
The thing both modes share is the rubric. A clear rubric is what lets the same criteria drive a helpful draft comment and a defensible final grade. If you are setting one up, start with how to write an assessment rubric, then see how AI rubric marking works. Scorafy marks open-ended answers against your rubric with cited evidence and a human sign-off step, which works whether you are returning a formative draft or finalising a summative result. Try it on a real submission.