How long does the student conversation take?

About six minutes. That is the limit every assignment currently runs with, and a student who has an extended-time accommodation recorded on the assignment gets a longer one applied automatically. Students complete it on their own time, and the result reaches the professor within minutes of the conversation ending.

What if a student has a speech impediment or is a non-native English speaker?

The interviewer waits rather than filling silences, and it repeats a question word for word as many times as a student asks. Hesitation and filler words are explicitly not treated as signs of dishonesty. Students are not assessed on fluency, and professors receive the full transcript alongside both results, so they can apply their own judgment when a student’s communication style does not reflect what they know.

Does this assign grades or replace instructor judgment?

No. OralExam.ai does not assign grades. It reports two results, Understanding and Authorship, each as a category rather than a number, and each backed by the specific observations drawn from the conversation. That arrives with the full transcript. What it means for a grade is your call, and the platform never makes it for you.

Can students game the system by memorizing their paper?

Memorizing the paper is not enough. The questions are written from that student’s own submission and follow up on the answers they give, asking after their reasoning and the connections between their own claims. A student who only memorized the text runs out of runway quickly.

Is student data and course content secure?

All data is encrypted in transit and at rest. Conversations are processed through enterprise AI sub-processors bound by data processing agreements that forbid training on your data. Webcam recordings carry a ninety day retention window set in the platform, after which an administrator purges them; that purge is not yet on an automatic schedule, and we would rather tell you than imply otherwise. We are scoping SOC 2 Type I attestation as the first step toward Type II. Institutions can request a signed Data Processing Agreement covering FERPA-aligned ownership, retention, and deletion, and the current sub-processor list comes with it.

Does OralExam.ai integrate with my LMS?

We are currently integrated with Canvas. This includes roster and course imports. We are actively building additional integrations. Please let us know if there are specific integrations that you require.

Stop policing AI. Start verifying learning.

AI detection is an arms race. Conversation is the answer.

A student's work is uploaded, then they have a discussion with an AI examiner that has already read it. Two results emerge, Authorship and Understanding, each kept separate. Behind every result sits the evidence for it and the full transcript. No detector. No accusation. A conversation instead.

Request a Demo See how it works

Example result

Understanding

Strong

Authorship

Clear

Two signals, reported separately and never averaged, each backed by what the student actually said.

What happens when every student has access to AI?

Students are using AI. No instructor is going to change that. What teachers need to know has not changed either: did this student actually learn? Detection tools showed up promising an easy answer. Scan the text, flag the cheaters, move on.

It has not worked. It is not going to.

The technology is unreliable
False positives destroy trust and false negatives make the tool pointless. As the writing models improve, accuracy gets worse rather than better. That is an arms race, and it is not one a course can win.
Detection catches, it does not teach
A flag tells you nothing about what a student knows. It turns educators into investigators and students into suspects, and the question you actually cared about is still unanswered.
Text analysis misses what matters
Writing can be generated, paraphrased, and polished past recognition. The thinking behind it cannot. That is the part a conversation reaches and a scan never will.

The OralExam.ai solution

A conversation goes where a scanner cannot. Straight to the student. Someone who understands their own paper can talk through it without trouble, while someone who handed the thinking to a chatbot tends to come apart quickly.

Talking with a student to find out what they know is how examination worked for most of history, going back to Socrates, who tested people entirely through conversation some 2,400 years before AI. The written paper came along much later, once classes grew too large to sit with everyone. But that paper is also the one thing a chatbot can produce on demand.

All we have done is bring the older method back and scaled it for a room of any size.

Five steps to verified learning

The oral exam is not a new idea. Making it scalable is. What took hours of one-to-one interviews now runs for every student, on every assignment.

The professor creates an assignment
Set the submission requirements and up to three optional learning goals. It runs alongside the course workflow you already have rather than replacing it.
Work is uploaded
Students upload directly, or a professor uploads on their behalf. PDF, Word, plain text and Markdown files are accepted, up to 25MB.
The questions are written from that submission
The AI reads the student's own work and builds the interview out of it: their claims, their sources, the choices they made. Every question set is different because every submission is.
The student talks it through
A spoken conversation with an AI examiner and a live transcript beside it. It runs about six minutes, longer for a student with an extended-time accommodation on file. The examiner does not praise or pass judgment on student responses.
The professor reads the result
Two results, side by side, within minutes, each carrying the evidence behind it and the full transcript. The instructor takes it from here.

There is no number, on purpose.

A percentage only sounds precise. It hands you a figure you would then have to defend, with no honest way to say why the work landed at 82 rather than 88. Instead, you get two separate signals, each with the observations that produced it.

Understanding asks whether the student grasps the material and reads Weak, Partial, or Strong. Authorship asks whether the student could have written it and reads Concerns, Review, or Clear. They stay separate and are never averaged into one result.

Example result

Essay 2: Causes of the Financial Crisis

Student submission. Oral exam complete.

Understanding

Do they grasp the material?

Strong

Explained the work in their own words.

Authorship

Could the student have written this?

Clear

Nothing here needs your attention.

Evidence, Understanding

Thesis Understanding
Demonstrated
Restated the paper's central claim without reaching for its wording.
Evidence Knowledge
Partial
Named both sources, but could not say why one carried more of the argument.
Depth of Understanding
Demonstrated
Followed the argument into a case the paper never mentions.

Evidence, Authorship

Process Awareness
Demonstrated
Described an earlier version of the thesis and why it was cut.
Evidence Knowledge
Partial
Named both sources, but could not say why one carried more of the argument.

Transcript excerpt

Examiner
You argue that securitization moved the risk rather than removing it. Walk me through how that worked.
Student
Banks bundled the mortgages and sold slices of them on, so a default no longer sat with the lender who wrote the loan. The risk did not disappear, it moved to whoever held the slice, and nobody could see how much of it they were holding.
Examiner
Your paper leans on one ratings agency figure. Why that one?
Student
It was the only one that broke the downgrades out by year, which is what the argument needed. I should have checked it against a second source, and I did not.

Why voice, not text

Detection asks the wrong question
“Did this student cheat?” produces evidence for a punishment, and it gets harder to answer every year.
OralExam.ai asks the right one
“Did this student actually learn?” produces insight for better teaching. We verify learning, not authorship. That question is answerable today, and it will still be answerable when the writing tools are better than they are now.
Make cheating pointless, not just punishable.
When learning is verified in conversation, the shortcut stops paying. No accusation, no honor code case, and the incentive quietly disappears on its own.

Who this is for

The same platform whether you teach one section or run assessment for a district. A high school pilot is running now, and the first university courses go live this fall.

Individual educators

Verify that students understand what they submit, without adding a semester of one-to-one oral exams to your calendar.

High schools

Multiple pilots are actively running. OralExam.ai meets or exceeds all compliance and privacy requirements for high-school students.

Departments and programs

Standardize verification across sections and instructors, so two students doing the same work meet the same interview. Roster results export as CSV in the categories the professor read.

Universities

Multiple pilots are actively running, including one at Tulane University. Bring verification to courses across the institution, including foreign language classes. Role-based visibility and a compliance pack your security team can review.

Built to be checked.

Nothing here asks you to take our word for it. Every result arrives with the transcript and the observations behind it, and none of it overrides you.

The instructor always has the final say. We do not assign grades and we do not decide what a result means for one. We give you what the conversation showed, and the judgment stays yours.

AI-agnostic by design
We are AI-agnostic. The professor decides the policy. We verify the learning. A student who used AI and did not learn will demonstrate that in the conversation. Additionally, a student who used AI and did learn will show understanding but might show concerns over authorship. The instructor decides what to do with this information.
Process over output
A polished paper is not proof of learning. The research behind it is, and the reasoning, and the draft that got thrown away. A conversation reaches those; a finished document cannot show them.
Full transparency
Every conversation leaves a complete record: the transcript, both results, and the observations behind each. No result exists that you cannot trace back to something the student said.
No shortcut through it
Every question comes from the student's own submission. No prompt, no paste, and no tool carries someone through a real conversation about work they do not understand. We never have to stay ahead of anything.

Security and student data

FERPA-aligned infrastructure, AES-256 encryption at rest, TLS 1.2 and above in transit, US-based data residency, and role-based access. We are scoping SOC 2 Type I attestation as the first step toward Type II. We never use student or faculty work to train AI models, and our sub-processors are contractually barred from it too. These are the standards universities and school districts hold us to. The full picture, gaps included, is on the compliance page.

How we handle security and student data

What professors are asking

What comes up most when faculty and administrators look at this properly.

How long does the student conversation take?
About six minutes. That is the limit every assignment currently runs with, and a student who has an extended-time accommodation recorded on the assignment gets a longer one applied automatically. Students complete it on their own time, and the result reaches the professor within minutes of the conversation ending.
What if a student has a speech impediment or is a non-native English speaker?
The interviewer waits rather than filling silences, and it repeats a question word for word as many times as a student asks. Hesitation and filler words are explicitly not treated as signs of dishonesty. Students are not assessed on fluency, and professors receive the full transcript alongside both results, so they can apply their own judgment when a student’s communication style does not reflect what they know.
Does this assign grades or replace instructor judgment?
No. OralExam.ai does not assign grades. It reports two results, Understanding and Authorship, each as a category rather than a number, and each backed by the specific observations drawn from the conversation. That arrives with the full transcript. What it means for a grade is your call, and the platform never makes it for you.
Can students game the system by memorizing their paper?
Memorizing the paper is not enough. The questions are written from that student’s own submission and follow up on the answers they give, asking after their reasoning and the connections between their own claims. A student who only memorized the text runs out of runway quickly.
Is student data and course content secure?
All data is encrypted in transit and at rest. Conversations are processed through enterprise AI sub-processors bound by data processing agreements that forbid training on your data. Webcam recordings carry a ninety day retention window set in the platform, after which an administrator purges them; that purge is not yet on an automatic schedule, and we would rather tell you than imply otherwise. We are scoping SOC 2 Type I attestation as the first step toward Type II. Institutions can request a signed Data Processing Agreement covering FERPA-aligned ownership, retention, and deletion, and the current sub-processor list comes with it.
Does OralExam.ai integrate with my LMS?
We are currently integrated with Canvas. This includes roster and course imports. We are actively building additional integrations. Please let us know if there are specific integrations that you require.

Request a Demo

Whether you teach one section or run assessment for an institution, leave an address and we will set up a walkthrough with the people who built it.

Pricing is not set yet. We have not decided on tiers or seat counts, and we would rather say so than invent them.

Open to educators and institutions, in higher education and in high schools.

Stop policing AI. Start verifying learning.

AI detection is an arms race. Conversation is the answer.

The OralExam.ai solution

All we have done is bring the older method back and scaled it for a room of any size.

There is no number, on purpose.

Example result

Essay 2: Causes of the Financial Crisis

Student submission. Oral exam complete.

Understanding

Do they grasp the material?

Strong

Explained the work in their own words.

Authorship

Could the student have written this?

Clear

Nothing here needs your attention.

Evidence, Understanding

Thesis Understanding
Demonstrated
Restated the paper's central claim without reaching for its wording.
Evidence Knowledge
Partial
Named both sources, but could not say why one carried more of the argument.
Depth of Understanding
Demonstrated
Followed the argument into a case the paper never mentions.

Evidence, Authorship

Process Awareness
Demonstrated
Described an earlier version of the thesis and why it was cut.
Evidence Knowledge
Partial
Named both sources, but could not say why one carried more of the argument.

Transcript excerpt

Examiner
You argue that securitization moved the risk rather than removing it. Walk me through how that worked.
Student
Banks bundled the mortgages and sold slices of them on, so a default no longer sat with the lender who wrote the loan. The risk did not disappear, it moved to whoever held the slice, and nobody could see how much of it they were holding.
Examiner
Your paper leans on one ratings agency figure. Why that one?
Student
It was the only one that broke the downgrades out by year, which is what the argument needed. I should have checked it against a second source, and I did not.

Why voice, not text

Detection asks the wrong question

“Did this student cheat?” produces evidence for a punishment, and it gets harder to answer every year.

OralExam.ai asks the right one

“Did this student actually learn?” produces insight for better teaching. We verify learning, not authorship. That question is answerable today, and it will still be answerable when the writing tools are better than they are now.

Make cheating pointless, not just punishable.

When learning is verified in conversation, the shortcut stops paying. No accusation, no honor code case, and the incentive quietly disappears on its own.

Built to be checked.

Nothing here asks you to take our word for it. Every result arrives with the transcript and the observations behind it, and none of it overrides you.

The instructor always has the final say. We do not assign grades and we do not decide what a result means for one. We give you what the conversation showed, and the judgment stays yours.

Request a Demo

Whether you teach one section or run assessment for an institution, leave an address and we will set up a walkthrough with the people who built it.

Pricing is not set yet. We have not decided on tiers or seat counts, and we would rather say so than invent them.

Open to educators and institutions, in higher education and in high schools.

Stop policing AI. Start verifying learning.

An example of the result a professor receives

Understanding

Authorship

What happens when every student has access to AI?

The technology is unreliable

Detection catches, it does not teach

Text analysis misses what matters

The OralExam.ai solution

Five steps to verified learning

The professor creates an assignment

Work is uploaded

The questions are written from that submission

The student talks it through

The professor reads the result

There is no number, on purpose.

Understanding

Authorship

Evidence, Understanding

Evidence, Authorship

Transcript excerpt

Why voice, not text

Detection asks the wrong question

OralExam.ai asks the right one

Make cheating pointless, not just punishable.

Who this is for

Individual educators

High schools

Departments and programs

Universities

Built to be checked.

AI-agnostic by design

Process over output

Full transparency

No shortcut through it

Security and student data

What professors are asking

Request a Demo

Stop policing AI. Start verifying learning.

An example of the result a professor receives

Understanding

Authorship

What happens when every student has access to AI?

The technology is unreliable

Detection catches, it does not teach

Text analysis misses what matters

The OralExam.ai solution

Five steps to verified learning

The professor creates an assignment

Work is uploaded

The questions are written from that submission

The student talks it through

The professor reads the result

There is no number, on purpose.

Understanding

Authorship

Evidence, Understanding

Evidence, Authorship

Transcript excerpt

Why voice, not text

Detection asks the wrong question

OralExam.ai asks the right one

Make cheating pointless, not just punishable.

Who this is for

Individual educators

High schools

Departments and programs

Universities

Built to be checked.

AI-agnostic by design

Process over output

Full transparency

No shortcut through it

Security and student data

What professors are asking

Request a Demo