The Case for AI-Assisted Written Response Feedback
A framework for LEAP assessment preparation
Multiple choice items have a ceiling. Even well-designed MC questions max out at Understand.
The cognitive work is recognition, not production.
| Bloom's Level | Cognitive Operation | MC? | Written? |
|---|---|---|---|
| Remember | Retrieve facts | ✓ | ✓ |
| Understand | Restate, summarize | ✓ | ✓ |
| Apply | Use in new context | Limited | ✓ |
| Analyze | Explain how/why, connect | ✗ | ✓ |
| Evaluate | Argue, justify with evidence | ✗ | ✓ |
| Create | Construct original synthesis | ✗ | ✓ |
Written Response Weight by Subject
| Rubric Language | Bloom's Level | Student Demonstrates |
|---|---|---|
| "Identifies" | Remember/Understand | Names without reasoning |
| "Describes" | Understand | Restates in own words |
| "Explains" | Analyze | Shows how/why with reasoning |
| "Supports with evidence" | Analyze | Connects information to claim |
A student who identifies but doesn't explain scores a 1, not a 2.
That's not a writing problem—it's a thinking problem.
The feedback bottleneck is human.
LDOE scoring annotations reveal exactly what separates score levels:
Identify vs. Explain. One example vs. two. Source-dependent vs. outside knowledge.
AI can detect these distinctions because they're structural, not stylistic.
"You need more detail."
"You've identified the compromise (Remember), but haven't yet explained how it addressed the underlying conflict (Analyze). What did each side gain?"
The second tells the student exactly what cognitive move they're missing.
AI doesn't replace the state test scorer. It helps students practice before the test.
10 minutes, not 10 days Low-stakes. Immediate. Repeatable.
The question isn't whether this is possible.
It's whether students get access to practice before the test—
or only find out what they didn't understand after the score comes back.