How to Run a 2-Minute Oral Check

I've spent four-plus decades in education, over thirty years of it in the classroom — including right now — and two years ago I stopped being able to tell who understood my content.

Not because my students got worse. Because I could no longer read the evidence. A perfect problem set came in from a kid who had not spoken in six weeks. A lab conclusion arrived with better prose than I write. I was grading artifacts that told me nothing about the human being who submitted them.

I want to be careful here, because the easy move is to blame kids and I am not going to do that. My students are not cheating at me. They are responding rationally to a system that asks them to produce a document and then rewards the document. If you can get the document without the thinking, and the document is what gets graded, the thinking becomes optional. We built that. They just noticed.

So I went back to the oldest assessment technology there is. I started asking students to explain their reasoning out loud, to my face, for two minutes.

It fixed it. Not partially. Completely.

That was the beginning. What I want to tell you now is where it's gone since — because it hasn't stayed a single two-minute check. It became three different conversations, and I'm still finding the edges of all three.

What it actually is

A short, unscheduled, face-to-face conversation in which a student explains the reasoning behind work they've already done. No screen, no slides, no audience, no preparation. Universities have done this for eight hundred years and called it a viva voce — the living voice. We abandoned it in K-12 for a defensible reason: it doesn't scale on paper. Written work was easier to grade in a stack at 9 p.m. That tradeoff made sense for a hundred years. It stopped making sense in 2023.

Every other defense against AI-assisted work is a detection problem, and detection is a race you lose. A viva sidesteps the question of how the work was produced and asks a different one: can you use it? A student who understands can extend, adapt, defend. A student who submitted something they never thought through cannot — not because they're dishonest, but because there's nothing to retrieve. There's no model in their head to consult.

The three conversations, not one

I used to think of this as a single kind of check. It's actually become three, and they do different jobs.

The student asks me a question. After the pre-class video or reading, I circulate and have each student ask me something — anything they found genuinely confusing. They know it's coming, so most come with something written down. Half the value is just watching what they chose to ask.

I ask the student a question. Once they've finished a lesson, one conceptual question, out loud. I grade it, but I still treat it as formative — if someone's clearly not there yet, I send them back to the text and move to the next student. More often than not, the one I "left" comes back a few minutes later having worked it out.

The summative viva. This one's new, and it's the one I hesitated on longest. While students are taking a written test, I now call them up one at a time and ask them to defend one answer, live, no help. It's a required, graded piece of every major exam now — not something I do informally anymore.

I made that last one official a few weeks ago, and it surprised me more than anything else in this whole process.

Instinct beat the rubric

Within about ninety seconds of a summative viva, I usually already know: deeply struggling, solid, or somewhere murky in between. Thirty-some years of building rubrics, and right now I'm trusting my gut over one. I'll probably build a tighter scoring instrument eventually. I haven't yet, and I'm not going to pretend that's a finished system — it's an honest gap, not a hidden one.

What I can tell you is that it's revealed understanding faster and more accurately than any quiz I've given this year.

It also isn't the same everywhere in my own building. In my geology class, the checks are quick and foundational — a minute or two each, and I get through every student with room to spare. In Honors Physics, eighteen kids in a single seventy-five-minute block, and I barely finished the last student before the bell. Thirty kids in a standard fifty-minute period simply does not fit in one sitting. I'm not going to pretend I've solved that. What I'm actually trying: pulling kids before school or at lunch, and triaging — not every student needs the same four minutes, because some of them you already know have it cold.

One more thing I've started doing that helps more than I expected: I don't wait for a student to finish the whole assignment. If I can see partway through a lesson that someone's ready, I grab them early. Most of the time they do just fine — often better, because the reasoning is still fresh instead of finished and filed away. It also quietly solves the pileup problem, because I'm not waiting for thirty students to all cross the finish line in the same five minutes at the end of class. I'm catching them as they're ready, spread across the whole period instead of stacked at the bell.

The line I won't cross

The obvious fix for the "line at the end of class" problem — a few kids who never get their turn before the bell — is to let AI pick up the slack. So I built one: an AI viva, a spoken conversation students have with a tool at home if they didn't get to me in class. It's routine, low-stakes, and it works exactly the way the AI-Engines part of this framework is supposed to work.

But the summative viva — the one that determines a grade — has to be me, one-on-one, in the room. I won't hand that off. If it ever started to feel fine to outsource that particular conversation, I'd take that as a sign something had gone wrong with the whole approach, not as a sign the tool had gotten good enough.

Here's the part I didn't expect: the AI viva has turned into its own kind of motivation, just not the kind I was aiming for. Students would rather talk to me than to the bot. Nobody's told me that outright, but I can see it in what they do—kids hustle to finish in class so they don't end up stuck explaining their reasoning to a screen at home instead of to a person. I built the tool as a backstop. It's turned into evidence for the whole premise of this page: given the choice, they'd rather account for their thinking to a human being.

What I'm seeing in other people's rooms

I hear from teachers using some version of this most days now — a professor in Costa Rica built his own four-phase adaptation for a university banking-theory course; a ministry of education in Argentina is training teachers in it district-wide. None of them are running my exact model, and that's fine. This was never meant to be a formula.

But a pattern keeps showing up in what they tell me back: the checks reveal things a rubric doesn't, and they cost real class time to run at scale. One teacher told me she's grading too generously right now, simply because she's pressed for time and doesn't want a kid arguing about a grade she assigned on the spot. Another noticed her strongest students get more nervous in the conversation than her struggling ones — confidence and mastery aren't the same thing, and a viva finds the gap between them. A third told me that after doing a video-introduction assignment alongside her early vivas, she was startled by how much trust her students handed her in the first two weeks — more than in years of grading their papers.

None of that shows up in a stack of graded documents. It only shows up when you're talking to someone.

Samuel

Last spring I had a student in honors physics — a football player who had cheerfully described the course to me in September as the class he was going to survive rather than learn. He failed a unit assessment. Under the old system, that grade would have stood and we both would have moved on, and neither of us would have learned anything.

Instead, he came back for a retake, and before it, he sat down for his check. He explained the whole chain of reasoning without notes. Then he said something I have thought about ever since:

"Mr. Bergmann, I studied so hard. I basically went into my cave. I was just doing these problems and practicing, and it paid off. I didn't push the button."

He knew the button was there. He knew I couldn't have proven he'd used it. He did the work anyway, because he knew that in two minutes he'd have to sit across from a human being and account for his own thinking.

That's the whole thing. Not surveillance. Not detection. Just the ancient, slightly uncomfortable fact that it's very hard to explain something out loud to another person that you never actually understood.

Where to start Monday

You don't need all three conversations on day one. Start with one:

Pick one lesson this week and ask each student a single question about it, out loud, at their desk.
Say nothing evaluative while they talk — the moment you correct them mid-answer, they start performing instead of thinking.
Score it in your head immediately, in three buckets: got it, close, not yet. Don't carry it home.
If it's "not yet," send them back to the work, and mean it — no penalty, just try again.
Do that for two weeks before you decide whether to make anything summative or graded.

That's it. Everything else in this piece is what happened after I did that.

What will go wrong in the first month?

You will talk too much. Every teacher does. You'll explain, correct, and teach during the check, and it'll run six minutes instead of two. Bite your tongue. The silence is where the assessment happens.

Your first questions will be recall questions. "What's the formula for acceleration?" A student who didn't do the work can still answer that. Ask about the decision, not the fact — why they started where they started, what they ruled out.

Quiet kids will scare you. A student who freezes is not a student who didn't understand. Let them look at their own work while they talk. Give them a second to gather themselves. You're assessing reasoning, not composure under pressure.

Someone will say it's unfair to English learners. Take that seriously. Let students explain in their first language when you can find a way. Accept sketches, gestures, pointing. You're assessing reasoning, not fluency, and you have to build the check so those two never get confused.

You'll be tempted to grade every conversation the same way. A ninety-second gut call and a formal summative viva aren't the same instrument. Don't hold lesson-level checks to the rigor of a graded exam conversation, and don't let a summative viva get as loose as a hallway check-in. Keep the stakes attached to the right conversation.

Common questions

How long does a Mastery Viva take? About two minutes for a lesson-level check, longer for a summative one where you're asking a student to defend a full answer. A class of thirty takes roughly an hour of work time for lesson checks, spread across a couple of class periods.

Doesn't this just punish shy students? No, if you build it right. There's no audience, no preparation, and no penalty for a rough first attempt. Students can look at their own work while they talk. You're assessing reasoning, not public speaking.

What if a student memorizes an AI-generated explanation? Memorized explanations survive question one and die on question three or four. Ask them to change a variable and predict the result. Memorization can't adapt; understanding can.

Is the AI viva tool just outsourcing your job to a bot? No — it's a backstop for lesson-level checks a student didn't get to in class, not a substitute for the conversation that determines a grade. See "The line I won't cross" above. Students seem to sense the difference, too — they'd rather finish with me in class than talk to the bot at home.

Can I use this in English, history, or elementary grades? Yes. The questions change; the structure doesn't. Elementary teachers I've heard from run them shorter — sixty to ninety seconds — and more often.

Do I have to stop giving written work? No. Keep it. Change what it's for. The written work becomes the thing you talk about, not the thing you grade in isolation.

Is this the same as a presentation or a Socratic seminar? No. Those are performances, with an audience and preparation time. This is private, unrehearsed, and built on reasoning the student has already committed to paper or already worked through.

What if I fall behind and can't get to everyone? That's the real logistics problem, and I don't have it fully solved either — see the geology-versus-Physics section above. Grabbing students early, before they've finished the whole assignment, is the biggest lever I've found so far.

I teach high school science and I have been doing this for thirty-two years. Everything above is what I actually do in my room, and it is yours.

If you want the version with the video walkthroughs, that's the MasteryFlip Certification. If you just want to try the oral check on your own, you already have everything you need.