What’s behind high marks?
Using AI, Artefact analysed almost five thousand lessons and compared their quality with students’ marks
To find out what’s behind high marks, Artefact used AI to analyse recordings of 4,796 lessons in 22 classes of one school. We compared lesson quality with students’ marks and looked at what set the better lessons apart.
In classes where lessons went better on average, marks were usually higher too. Such lessons less often involved conflicts or long stretches of the same kind of work. The teacher more often set a clear goal, kept the whole class working and checked understanding.
The study covers lessons recorded between 30 January and 29 May 2026. In the chart, each lesson is one cell: the more orange it is, the more signs of a well-built lesson it showed.
4,796lessons from one school went into the study
How we compared lessons
Each lesson is assessed on 25 teaching criteria. Among them: whether students understand the goal, whether the whole class takes part, whether the teacher gets feedback during the lesson. The higher the ratings on these criteria, the higher the total score and the more orange the cell.
The median across the full record is 81.3%; half of the lessons fall between 70.5% and 88.6%. 308 lessons scored below 50%.What do the marks say?
In classes with higher lesson scores, students’ marks for the same period were usually higher as well. We set the two measures side by side to check whether the lesson score reflects anything beyond the teaching scale itself.
For 20 classes we compared the average lesson score with the average percentage result in marks over the same four months. The correlation was 0.6 on a scale from −1 to 1.But comparing, say, a fifth-grade and a tenth-grade class directly is not sound: they have different programmes and grading systems. So we repeated the check only between classes in the same grade. In of them, higher lesson scores also went with higher marks.
The comparison covered 18 classes from 8 grades: 12 pairwise comparisons in total. In 9 of them, lesson scores and marks pointed the same way. Comparisons within three-class grades are not independent of each other.What gets in the way
After checking the link with marks, we turned to the lessons with low scores and looked at what stopped them from going well.
The first obvious cause could be conflicts, or moments when the class visibly stopped taking part. Lessons with such episodes did score lower on average, but that was not enough to explain every weak lesson.
Recorded incidents occurred in 804 lessons; their average score was 64.9%. The 3,987 lessons without such episodes averaged 80.8%. Of the 308 lessons scoring below 50%, 169 passed without a conflict or a visible drop in participation.55%of weak lessons passed without conflicts or the class dropping out of the work
So we looked at how the lesson itself was built and saw another shared trait: the lesson got stuck in one format for a long time. The work never gave way to a discussion or an independent task.
Among lessons without recorded incidents, activity switching was rated zero in 89 lessons, partial in 708 and full in 2,018. With the criterion itself removed, the median scores on the remaining items were 52.4%, 78.6% and 85.7% respectively.Conflicts and monotonous work explained only part of the problem lessons. Many weak lessons had neither. So other explanations had to be tested.
At least one of the two signs appeared in 893 lessons. Among the 308 lessons scoring below 50%, they marked 183; another 125 weak lessons fell outside these two scenarios.Outside causes?
Weak lessons turned up on every day of the week, in different classes and in different subjects. We tested three theories:
The “heavy Friday”
The “difficult class”
The “hard subject”
It turned out that weak lessons occurred at roughly the same rate all week. There was no sharp Friday slump, and no link to the lesson’s position in the day.
Among lessons without recorded incidents, the share scoring below 50% ranged from 3.3% on Tuesday to 5.4% on Friday. Shares are compared, not the number of lessons on the timetable.Perhaps there is a “difficult class” that struggles with a particular subject?
But groups had both strong and weak lessons. So the class and subject combination did not explain the result on its own either.
48%of weak lessons received the maximum rating for subject knowledge
What do high-scoring lessons have in common?
High-scoring lessons more often combined three elements: a clear goal, whole-class participation and feedback during the lesson.
A clear goal
The first difference appeared right at the start of the lesson: the teacher clearly set out the learning task and the plan.
233 of the 4,783 lessons where the criterion applied received the maximum rating for goal and plan. Maximum meant the learning task was stated and the plan was clear to students. Points for goal and plan were removed from the position and colour of the cells.Whole-class participation
It mattered how many students actually took part. Lessons where the teacher kept the whole class working scored higher.
3,276 of the 4,781 lessons where the criterion applied received the maximum rating for class participation. The rating covered the whole class, not just a few active students. Points for participation were removed from the position and colour of the cells.Feedback during the lesson
The teacher gave students feedback as the lesson went on, without waiting for the end. As a result, they knew straight away what was working and what needed fixing.
1,060 of the 3,345 lessons where the criterion applied received the maximum rating for in-lesson feedback. The average score on the remaining criteria was 51.8% without feedback and 88.8% with the maximum rating for it. Points for feedback were removed from the position and colour of the cells.37points between lessons without feedback and lessons where the teacher gave it as the lesson went on
The secret is the teacher’s work
So lesson quality is linked to students’ marks, and the high-scoring lessons shared a sequence of actions: the teacher set the direction, involved the whole class and checked students’ reactions along the way.
The histogram highlights 3,484 lessons where at least one of the three criteria received the maximum rating. Points for goal, class participation and feedback were removed from the position and colour of the cells. In 7 of 8 comparable grades, the class with the highest student participation also had the highest result in marks.The whole week
A few successful or troubled lessons cannot give the full picture: in a single week the school held 632 lessons.
A coach sees five
In a week, an instructional coach usually has time to review only five lessons in detail. Depending on which lessons land in that five, the same school can look far stronger or weaker, and whatever repeats from lesson to lesson may go unnoticed altogether.
Five lessons a week is the school’s own estimate: that is how many an instructional coach usually manages to review in detail in that time. From the 632 lessons of the week there are 32,227,781,040 possible sets, one lesson from each school day. For each set we calculated the median score.Classes with high marks also had higher average lesson scores. In those lessons the teacher more often stated the goal clearly, involved the whole class and got feedback as the lesson went on. Low-scoring lessons more often involved conflicts and long stretches of work in a single format.
The usual explanations, the “heavy Friday”, the “difficult class” and the “hard subject”, did not account for the difference. The same class in the same subject had both strong and weak lessons. The difference was more often down to how the teacher built the lesson itself.
A handful of lessons gives an unreliable picture: depending on which lessons the coach happened to observe, the school could look noticeably stronger or weaker. A full record shows what repeats from lesson to lesson and points to specific moments to work on with the teacher.
What is happening in your school?
Artefact brings every lesson into one picture and helps you see what sets one lesson apart from another.
Try Artefact