← Design·Forum Post Detail

V1 — Clean Thread. Single column, post card with channel accent, replies as separate cards below. Most straightforward.

PA

Priya Anand

Ed.M. '26, TIE · 3 hours ago

Question

Best practices for AI-assisted formative assessment?

I’m designing a unit where students get real-time AI feedback on their writing drafts. The goal is to use AI as a formative tool — helping students improve DURING the writing process rather than just getting a grade at the end.

The challenge I’m running into: students tend to just accept whatever the AI suggests without critically evaluating it. It’s becoming a compliance exercise rather than a learning one.

Has anyone found effective guardrails or scaffolds to prevent this? I’m thinking about:

1. Requiring students to justify why they accepted or rejected each AI suggestion 2. Having the AI give contradictory feedback sometimes to force evaluation 3. Making the AI feedback intentionally vague so students have to interpret it

Would love to hear what’s worked in your classrooms, especially with middle or high school students.

AssessmentK-12
3 replies

Replies · 3

TS
Tomoko Sato2 hours ago

Your option #1 is closest to what worked for me. I have students keep an ‘AI feedback log’ where they record each suggestion, their decision (accept/reject/modify), and a brief rationale. It adds 5-10 minutes per session but the metacognitive gains are significant.

DD
David Dockterman1 hour ago

Be careful with option #2 — intentionally misleading AI feedback can erode trust in the tool and in you as the instructor. I’d lean toward making the feedback more open-ended rather than contradictory. Something like ‘Consider whether your evidence supports your claim’ rather than giving a specific correction.

SO
Sarah Okonkwo45 min ago

I’ve been experimenting with a ‘peer + AI’ model where students first get AI feedback, then discuss it with a partner before making revisions. The peer discussion forces them to articulate why the AI’s suggestions do or don’t make sense. It’s slower but the quality of revisions is much higher.