Your Test Results Are Back. Here's How to Turn Them Into Next Week's Teaching.
We all know the honest version: you mark the test, enter the marks, notice the class did badly on question 7, mean to do something about it, and then Week 4 arrives and you've moved on.
The problem isn't that teachers don't care about the data. It's that proper analysis takes an evening you don't have — and it's worthless unless it turns into an actual lesson.
Here's a workflow that gets you from marks to a teaching plan in one sitting. Works with Claude, ChatGPT, or a Gemini Gem.
Before you start: what to give the AI
Three things, none of which require new work:
- The questions — the test paper, or typed-out question stems with marks available
- The scores — one row per student, marks per question, copy-pasted from your spreadsheet
- A few wrong answers — three or four real student responses to the questions that went badly
Strip the names. Use "Student 1, Student 2" and keep the key on your own sheet. Your school will have its own guidance on this — follow it.
Step 1: Find the real pattern, not the obvious one
Below is question-by-question mark data for a Year 10 [subject] test, plus the question stems. For each question, calculate the class facility (percentage of available marks earned). Identify the three weakest questions. For each, tell me what specific skill that question actually required — not just the topic label — and give me two competing hypotheses for why students struggled. Don't recommend anything yet.
"Two competing hypotheses" is the important bit. Left alone, AI will tell you students "need more practice with fractions." Forcing it to argue with itself surfaces the real distinction: did they not know the content, or did they not understand the question?
Step 2: Test the hypotheses against real answers
Here are four actual student answers to Question 7. Which of your two hypotheses do these support? Quote the specific part of each answer that gives it away.
This is the step people skip, and it's the one that makes the whole thing worth doing. Marks tell you that they got it wrong. The answers tell you why. AI is good at spotting the shared misconception buried in four differently-worded wrong answers.
Step 3: Turn it into one lesson, not a wish list
Based on the confirmed misconception, design a single 50-minute reteaching lesson. Structure: a 5-minute starter that surfaces the misconception without naming it, 15 minutes of worked examples that directly confront it, 20 minutes of practice moving from scaffolded to independent, and a 10-minute exit ticket that would show me whether it's fixed. Give me the actual questions for the starter and the exit ticket.
Ask for one lesson. If you ask for "a reteaching plan," you'll get a five-week programme you'll never run.
Step 4: The two groups you actually need
From the mark data, identify: (a) students who scored below 40% on the weak questions but above 70% overall, and (b) students who scored well on the weak questions and could be paired as explainers. List them with a one-line reason each.
Group (a) is the one that gets missed. They look fine on the overall mark, so nobody flags them — but they've got a specific hole that will cost them later.
What to check before you trust it
AI cannot see your students. It gets the what right far more often than the why, which is why Step 2 exists. Two checks every time:
Spot-check the maths. Ask it to "recalculate the facility for Question 7 and show your working," then compare one figure against your spreadsheet. Arithmetic on pasted data is where these tools slip.
Read the diagnosis against what you saw in the room. If it doesn't match what you watched students do in class, back yourself — you were there and it wasn't.
Fifteen minutes, one lesson, and the students who were quietly stuck stop being quietly stuck.
TeachSmarterAI helps New Zealand teachers use AI tools effectively. New posts every week.