What I learned in a term of "AI-permitted" assessment

Last term I rewrote every major assessment in my Year 10 English class to explicitly permit AI use — provided students disclosed how they used it. I went in expecting either a productivity revolution or an integrity disaster. What I got was neither.

What worked

Disclosure as a learning artifact. The biggest win wasn't anything to do with AI itself — it was the disclosure paragraph. Asking students to write 100 words on "where I used AI, what I asked it, and what I changed about its output" forced a level of metacognition I'd never gotten from a reflection prompt. Their disclosures were often more interesting than the work.

Explicit "AI bands" on every task. A small visual badge on every rubric (Permitted / Permitted with disclosure / Not permitted) eliminated 90% of the "is this allowed?" anxiety. Students stopped asking; they just looked. Took me 20 minutes to design the system and it removed weeks of friction.

Conversations about quality, not policing. With AI use out of the shadows, the conversation shifted from "did you use it?" to "did it actually help?". Most students discovered their first-draft AI output was mediocre — and articulating why was a genuine literacy lesson.

What didn't work

My initial "trust the disclosure" approach. I assumed students would write honest disclosures. About 30% of them initially wrote vague ones ("I used ChatGPT for ideas"). I had to scaffold the disclosure paragraph far more aggressively in week 3 — give them sentence stems, examples of good disclosures vs weak ones.

Equity gap I didn't anticipate. Students with paid AI accounts produced consistently better first-draft material than students on the free tier. The disclosure made this visible, which then made it a teaching opportunity, but it took two weeks before I noticed.

What surprised me

The pace at which my own assessment design evolved. By week 6 I'd quietly stopped writing tasks that AI could trivially complete. Not because I was avoiding AI — because those tasks were genuinely uninteresting. The class moved toward authentic, contextual, performative work. The AI didn't ruin assessment; it ruined the bad assessments.

What I'd tell anyone starting

1. Build the disclosure muscle first. Don't open AI use until you've taught what good disclosure looks like.
2. Make the rules visible at the task level, not the policy level. Nobody reads a 12-page policy.
3. Watch for the equity gap. Tools vary. Be ready to talk about it.

I'm running it again next term with the disclosure scaffold built in from day one. Will write up term 2 in a few months.