How to Run a 2-Minute Oral Check

I have been teaching high school science for thirty-two years, 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, for two minutes.

It fixed it. Not partially. Completely.

This page is the whole method. Not a preview of a method — the method, including the compromises. Take it and use it Monday.

What a two-minute oral check actually is

A short, face-to-face conversation in which a student explains the reasoning behind work they have already done. You ask two or three questions. They answer out loud, without a screen. It takes about 120 seconds, and it happens at their desk, or a side table, or standing in the corner of the room while everyone else works.

It is not a presentation. There is no audience, no slides, no preparation, and no performance. Students are not being asked to be articulate. They are being asked whether the reasoning in front of them is theirs.

Universities have done this for eight hundred years and call it a viva voce — the living voice. Doctoral candidates defend dissertations this way. Medical residents defend their reasoning on rounds every single day. We abandoned it in K-12 for a defensible reason: it does not scale on paper. I am going to show you that it scales fine, but I want to name the real reason it disappeared, which is that 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.

Why it works when nothing else does

Every other defense against AI-assisted work is a detection problem, and detection is a race you lose. Detectors produce false positives. They disproportionately flag multilingual writers. They get defeated by a paraphrasing tool that costs nine dollars a month. And even when a detector is right, you are now in an accusation conversation with a sixteen-year-old and a parent, holding probabilistic evidence.

An oral check is not detection. It sidesteps the question of how the work was produced and asks a different one: can you use it? A student who genuinely understands can extend, adapt, and defend. A student who submitted a document they did not think through cannot — not because they are dishonest, but because there is nothing to retrieve. There is no model in their head to consult.

This is why I stopped caring how the homework got done. I care whether the understanding exists. The oral check measures that directly.

The other thing it does, which I did not anticipate: it fixes the relationship. I now have short individual conversations with every student in my room, several times per unit. Thirty-two years in, I know my students better than I ever have.

The three layers

This is the part most write-ups get wrong, including my own earlier version of this page. It is not one conversation. It is three, and they do different jobs.

Layer one — daily rehearsal. Three minutes, AI. Part of the pre-work for every lesson is a short oral check-in with an AI practice tool. Students watch the lesson, take handwritten notes, and then spend three minutes saying the ideas out loud to a bot before they ever walk into my room. This is practice, not assessment. Nothing is riding on it. Its only job is to make sure the first time a student articulates a concept is not the moment it counts.

Layer two — the lesson check. Two minutes, half AI and half me. When a student finishes a lesson, they get a Mastery Viva before they move on. Roughly half of these run through the AI tool and half happen with me directly. This is the workhorse layer, and the split is what makes it survivable at 150 students. More on that below, including why I think it holds up.

Layer three — the test defense. Five minutes, always me. On every major test, alongside the written portion, students defend one conceptual question out loud with me. They explain why their answer works and walk me through the logic. That defense is worth 15% of the test grade.

Layer three is never delegated. When the stakes are real, a human does the assessing.

Two ways to run the lesson check

Which version you use depends on how your class is paced, not how big it is. A self-paced class of thirty-two can run the first model easily. A lockstep class of eighteen cannot. Pacing is the variable that matters.

The Gate. Every student, every lesson. The conversation is what unlocks the next lesson. This requires self-paced work — students finish at different moments, so the checks arrive in a trickle instead of a wave, and they spread themselves across the period without you scheduling anything. This is what I do.

The Sample. If your class moves through material together, the gate creates a line at your desk. Run a rolling sample instead: twelve to fifteen unpredictable checks per period, so every student gets one every couple of weeks. Less complete, still effective — the unpredictability is what makes students prepare, because they cannot know which day is theirs.

Most teachers start with the sample. It works inside the schedule you already have. The gate is where you end up if you move toward mastery pacing, and if you get there, the check is the thing that makes it possible.

Everything below describes the gate, because that is what I run. If you are using the sample, the only real difference is who gets called and when.

The eight questions

You do not need a bank of questions. You need these eight, adapted to your content. I have used variations across physics, chemistry, and earth science, and colleagues have used them in English, history, and Spanish.

Walk me through your first move. Why did you start there instead of somewhere else?
What made you rule out the other approach? (Name a plausible wrong path.)
Where does this break? Under what conditions would your answer stop being true?
Change one thing. If I doubled this value, what happens to your result, and why?
What was the hardest part? What did you have to go back and redo?
Explain this to a ninth grader. No jargon.
You wrote this word. What does it mean? (Pick the most technical term in their submission.)
What are you still unsure about?

Question seven catches more than the rest combined, and it is the least confrontational thing you will say all day. You are simply asking someone to define a term they chose to use.

Question eight is the one I would fight for if you cut everything else. Students who did the work always have an answer. Students who did not tell you it all made sense.

How to run one

Set the rule on day one. Finishing a lesson does not mean turning something in. It means having a conversation. Students self-report when they are done, and that is what triggers the check. Nobody moves to the next lesson without it.
Go to them. I do these at the student's desk while the rest of the room keeps working. Standing over their own paper, in their own seat, is far less intimidating than being summoned to the teacher's desk — and it keeps the room's momentum intact.
Have their work in front of both of you. Point at a specific line. "Here — talk to me about this step." Concrete beats general every time.
Ask two questions. Maybe three. Two minutes. Set a timer on your phone for the first week until you calibrate; you will overrun badly at first.
Say nothing evaluative during the conversation. Nod. "Okay." "Keep going." The moment you say "not quite," they start performing for you instead of thinking out loud, and the instrument stops working.
Score immediately. In front of them, or right after. Do not carry it. If you defer, you will have eleven of these in your head by lunch and you will remember none of them.
Tell them the result and what happens next. Ten seconds. "You've got it — move on." Or: "The setup is solid, the second half isn't there yet. Go back to problems four through nine and come find me."

When two or three finish at once: tell them to start the next thing and you will get to them in order. Do not let a line form. A queue at your desk is dead class time and it makes the conversation feel like a tribunal.

How to grade it

Four points. The top two are the ones that matter.

4 — Transfer. Takes the knowledge and applies it to a brand-new, unfamiliar situation. Handles the perturbation question without flinching.
3 — Mastery. Thoroughly knows and understands the core material. Can explain the reasoning, not just the steps. This is the bar for moving on.
2 — Approaching. Right answer, thin reasoning. Can describe what they did but not why. Something specific to fix, then back to the work.
1 — Not yet. Cannot reconstruct the thinking. Returns to the lesson.

I ran this on three levels for a long time, because three is simpler to hold in your head while you are moving around a room. I went to four because three could not distinguish between a student who has the concept and a student who can carry it somewhere new — and that difference is the whole point. A 3 says you learned it. A 4 says it is yours.

Two rules that matter more than the scale itself:

A 1 or a 2 is repeatable at no penalty. The instant a low score becomes a permanent grade, students start preparing for the check as a performance, coaching each other on what you ask, and the whole thing collapses into a different kind of theater. It is a checkpoint, not a trap.

Send them back, and mean it. If a student lands at 1 or 2, they go back to the work. This is the part that makes the whole system real. If students learn that a shaky conversation still gets them through, you have taught them the check is decorative.

And keep the two kinds of oral separate. Lesson checks are a gate — they determine whether you move on, not what your grade is. The test defense is a grade, and it is 15% of each major test. Mixing those two confuses students and makes the formative checks feel high-stakes, which is exactly what you do not want.

Scaling it: 150 students, five sections, no aide

This is the objection, and it is the right objection. Here is what I actually do, including the part I used to be uncomfortable admitting.

Students do not finish at the same time. A lesson takes most of my students about eighty minutes of class time, but they arrive at the end of it across two or three days. So I am never doing thirty checks at once. I am doing ten or twelve in a period, in between everything else I am already doing.

AI handles about half the lesson checks. A well-built oral-practice tool can listen to a student explain their reasoning, ask a follow-up, and hand me a record of the conversation. It is not as good as I am. It misses things. But it is dramatically better than the alternative, which is that half of those students get no check at all.

I want to be honest about why I think this holds up, rather than pretend it is ideal:

It is the formative layer, not the summative one. Nothing that determines a grade is delegated.
The tool asks and records. I read the records. When something looks thin, that student goes on my list for a human conversation.
The test defense is always me. Fifteen percent of every major test grade rests on a conversation with a person who knows the student.

If you outsource all the verification, you have rebuilt the problem you were trying to solve. If you outsource none of it, you will quit by October. Half is where I landed after two years, and I would rather tell you that than sell you a purity rule I do not follow myself.

What to look for in a tool, if you use one: it should refuse to give answers, ask follow-up questions rather than accept the first response, and give you a transcript you can actually review. If it scores students without showing you the conversation, it is not helping you — it is hiding the thing you needed to see.

Two more ways to buy back the time:

Stop grading what the check replaces. If you are verifying reasoning face to face, you no longer need to write feedback on every problem set. Mark completion. The conversation is the assessment. This is not additional work — it is a substitution, and if you treat it as additional you will quit by October.

Let the length vary. Some checks take sixty seconds because the understanding is obvious in the first answer. Some take four minutes because something interesting is happening. Two minutes is the average, not the rule. Spending unequal time is what makes it useful.

A class roster taped inside a folder, with a tick and a date. That is the entire tracking system. I have tried spreadsheets. The folder wins.

What makes the check work: the room around it

Two things outside the conversation itself do most of the heavy lifting.

Handwritten notes. Students take notes by hand in a physical packet and bring it every day. Handwriting builds stronger memory connections than typing, and more practically: when a student has to talk me through their own handwriting, the thinking is already externalized. There is something to point at.

Phones away, laptops closed. The default mode in my room is paper, pencil, and conversation. Not because technology is bad — we use it deliberately for the pre-work and the practice reps — but because the check depends on a room where the shortcut is not sitting in a pocket six inches away.

You can run oral checks without either of these. They work better with them.

What will go wrong in the first month?

You will talk too much. Every teacher does. You will explain, correct, and teach during the check, and it will run six minutes. 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 did not do the work can answer that. Ask about the decision, not the fact.

Quiet kids will scare you. A student who freezes is not a student who did not understand. Give them the question in writing, let them look at their own work, wait. Anxiety is not the absence of knowledge, and if you conflate the two you will do real harm to exactly the kids who need you most. This is also where the daily three-minute rehearsal earns its keep — by the time they get to me, it is not the first time they have said it out loud.

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

Samuel

Last spring I had a student in honors physics who took the class because he liked me. I doubted he could handle the load, and for a while I was right. He had the lowest test scores in the room. I let students retake assessments — I know that is controversial — and he usually landed somewhere decent on the third attempt.

But he kept struggling.

Then he came back for attempt number two on a unit test, and he got a hundred percent. He cried. I cried.

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

That last sentence is the one I keep coming back to. He knew the button was there. He knew I could not have proven he used it. And he told me about it unprompted, because in a room where you have to explain your thinking out loud every single lesson, not pushing the button becomes the thing worth being proud of.

That is what the checks actually build. Not surveillance. Not detection. A culture where the reasoning is the currency — and the slightly uncomfortable, very old fact that it is nearly impossible to explain something out loud to another person that you never actually understood.

Common questions

How long does an oral check take? About two minutes for a lesson check, five for a test defense. In a self-paced room the lesson checks arrive gradually, so you are doing ten to fifteen in a period rather than thirty at once.

Doesn't this just punish shy students? No, if you build it correctly. There is no audience, no preparation, and no penalty for a poor first attempt. Students look at their own work while they talk, and they have already rehearsed the ideas with a practice tool before they get to you. You are assessing reasoning, not public speaking.

Isn't using AI to run half the checks hypocritical? It is the obvious objection and I am not going to dodge it. The distinction I hold to is formative versus summative. AI handles the practice reps and roughly half the low-stakes lesson checks, and I read those transcripts. Every graded oral assessment is a conversation with me. If a tool ever scored my students without showing me the conversation, I would stop using it.

What if a student memorizes an AI-generated explanation? Memorized explanations survive question one and die on question four. Ask them to change a variable and predict the result. Memorization cannot adapt.

How much should the oral count toward a grade? In my room, lesson checks are a gate rather than a grade — they determine whether you move on. The oral defense on each major test is worth 15%. That is enough to signal it matters and not so much that one nervous day sinks a student.

What if my class isn't self-paced? Use the rolling sample instead — twelve to fifteen unpredictable checks per period, every student every couple of weeks. Most teachers start here.

Can I use this in English, history, or elementary grades? Yes. The questions change; the structure does not. Elementary teachers typically run them shorter — sixty to ninety seconds — and more often.

Do I have to stop giving written work? No. Keep the written work. Change what it is for. It becomes the thing you talk about rather than the thing you grade.

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

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.