Getting the AI to grade exactly like you do isn’t magic—it just requires a few specific instructions. If you want to get a perfectly accurate grade on your very first try (and save your tokens!), follow these four Golden Rules of AI Prompting.
1. Make the AI “Show Its Work” (Chain of Thought)
AI models generate text one word at a time. If you ask the AI to output the final score first, it has to guess the grade before it has actually looked at the student’s errors!
- The Fix: Always tell the AI to list the missing items and deductions before it calculates the final score. If it lists the deductions first, it can accurately do the math for the final grade.
2. Use Negative Constraints
AI models can sometimes be a little too generous, especially when looking for math work. If a student needs to divide by 212, don’t just say “look for the fraction.”
- The Fix: Tell the AI exactly what incorrect answers look like. For example: “If the student divides by 254 or 94, it is incorrect.” Giving the AI specific “wrong” targets stops it from accidentally giving credit for bad math.
3. Provide Visual Landmarks
When grading handwritten worksheets or messy PDFs, the AI needs a map.
- The Fix: Tell it exactly where on the page to look. Instead of saying “check the graph,” say “check the x-axis of the segmented bar graph at the bottom of page 1.” This prevents the AI from accidentally reading the wrong section of the page.
4. Test Small Before Going Big
Never run a 35-student batch on an untested rubric!
- The Fix: Use the Test 1 Student feature to run the grader on a single paper. Spending 1 token to verify your prompt is a smart investment; burning 35 tokens on a typo is a tragedy!