This post was written by Diana Hughes and Alice Waldron, our colleagues from Relay Graduate School of Education. Relay is one of the nation’s largest independent, fully accredited nonprofit graduate schools of education and serves more than 4,000 students, teachers, and leaders across more than 250 school systems. You can read more about the integration between Teaching Lab and Relay GSE here.

At Relay, one of our most effective teaching methods is rehearsal practice. It works like this: in class, teachers pair up, and one practices a specific teaching skill — such as setting and supporting expectations, monitoring student understanding, or redirecting off-task behavior — while the other plays the role of a student and then provides feedback. A professor circulates, observing pairs and coaching in real time. It's hands-on, iterative, and grounded in the specific skills our curriculum targets.

This kind of deliberate practice — working toward specific goals, getting immediate feedback, and having the chance to try again — is what the research says builds expertise (Ericsson, 2008). But in teacher preparation, opportunities for this kind of practice are consistently hard to come by (Grossman et al., 2009). Rehearsal happens during Relay’s synchronous class sessions, which means coordinating schedules, pairing students, and relying on professor bandwidth to coach across multiple pairs at once. For Relay's students, most of whom are already full-time teachers, getting enough high-quality practice reps within those constraints is a real challenge.

We set out to find a more flexible, accessible way to provide that same kind of deliberate practice. We're discovering that we are on the path to something that doesn’t just match paired rehearsal practice - it could be even better.

What We Built

The Relay GSE Teaching Simulator is an on-demand practice environment where teacher candidates work through realistic teaching scenarios with AI-powered virtual students. The experience is conversational and turn-based: a teacher candidate enters a simulated classroom scenario, speaks to a group of virtual students using voice input, and the students respond. The conversation unfolds like a real classroom exchange - teachers practice the way they'd actually teach by talking, scanning the “room”, evaluating students’ responses, and adapting in the moment.

After each session, the simulator provides structured feedback tied to specific teaching skills, highlighting what went well and where there's room to grow. Candidates can repeat scenarios as many times as they want, trying different approaches each time. And because it's available on demand, practice isn't limited to scheduled class sessions; candidates can fit it in when and where it works for them.

Critically, the simulator is grounded in Relay's curriculum and coaching frameworks. The skills candidates practice and the feedback they receive reflect what Relay already knows works in classrooms. The AI is the delivery mechanism; the pedagogy is the product.

Putting It to the Test

In Fall 2025, we ran pilot studies across two required courses at Relay, comparing simulator-based practice to the traditional peer rehearsal model. The question was simple: can AI-powered practice produce outcomes and teacher satisfaction comparable to what we're already doing?

The results were encouraging. Teachers who used the simulator reported that it was just as (or more) engaging as in-class rehearsal, and crucially, also felt that it was just as or more effective than in-class rehearsal. Their end-of-term grades back this up: we see equivalent performance in the simulator groups and the non-simulator groups.

We want to be clear-eyed about what this evidence shows and doesn't show. These are early results with small samples, and most teachers scored near the top of available scales regardless of condition. But the direction is clear: simulator-based practice didn't harm outcomes or satisfaction, and it may offer something peer rehearsal can't.

The feedback candidates receive in the simulator is consistent and high quality — it doesn't vary based on which instructor happens to be circulating or whether a coaching moment gets missed in a busy classroom, and it is grounded in the same research and best practices that our faculty use when coaching. And unlike rehearsal with a single classmate, the simulator mimics the cognitive load of teaching to a room full of students, each with their own responses and behaviors. Plus, users can customize their scenarios to the grade level they teach, with developmentally-appropriate student responses included. As one candidate put it:

"I could make mistakes and feel more loose trying things out because I was in private! And I think it gave better feedback than if I was getting it from classmates, since it recorded every word I spoke and was able to hone in on more specific recommendations than any human could."

Another noted:

"It was nice to get feedback from a computer. Sometimes it's difficult and awkward to give or get feedback from a classmate."

What's Next

This spring, we're expanding the pilot with new scenarios and several new features. We're adding video input so candidates can practice with their full teaching presence — gestures, facial expressions — not just their voice. We're building lesson plan uploads, so scenarios can be personalized to what a candidate is actually teaching. And we're developing AI-generated student work, which will let candidates practice giving feedback on assignments, not just leading discussions.

We're also increasing the dosage, from 2-3 in-class sessions per semester to 8-10, with 24/7 access for use outside of class.

The big idea behind all of this is simple: practice shouldn't be scarce in teacher preparation. The constraints that have historically limited rehearsal — scheduling, coordination, instructor availability — don't have to be permanent. AI can make meaningful practice abundant and accessible, without sacrificing the pedagogical rigor that makes it valuable.