The student has an AI. Now the instructor has one too.

OralIQ is a spoken check on a student's own Canvas submission. It asks the student about the work they turned in, proposes a score on the instructor's rubric, and the instructor signs.

For students: asked, not accused.

An illustration of the student's screen. SYNTHETIC: the essay, the course and the question are fixtures.
The OralIQ mark, in motion.

AI can now produce any artifact a course asks for. So the artifact stopped being evidence that anyone learned anything.

The night before grades are due.

An instructor who teaches online opens a folder of 140 final essays. She has never been in a room with her students, and she cannot trust the folder. She knows what the AI-writing score will say, and she knows her conduct office will not take it. She has two moves. A zero, with the offer of a recorded oral that no student has ever accepted. Or a hearing.

Down the hall, her dean's board has just renewed its exam proctoring line for five more years, $589,368 a year, locked to 2031. Every one of those essays goes unverified.

She is a composite. Every number in her night comes from a public document: the folder of 140 from a university teaching center's published account of its own oral exams, November 2025; the exam line from a Florida state college board's transmittal, June 2026; the nineteen campuses and the twenty-two documents from their own published policies of the last three years.

The villain is a structure.

The course was built to grade the page, and the only check its faculty trust is a meeting the online course was never built to hold. So the structure sends her to the hearing room.

Nineteen campuses and systems have written into their own policy that the detector's score is not evidence, and the money stayed. Twenty-two have written down what replaces it: the student explaining their own work.

The artifact is no longer evidence.
The student is.

Same instructor. Same night.

This time the check was in her syllabus. After each student submitted, they opened it from the module, consented, and answered four to six spoken questions about their own essay. Why this source. What breaks if that paragraph goes. Where is the weakest claim, and how would you fix it. Ninety seconds an answer, one follow-up grounded in what they just said. No camera, no lockdown browser, no voiceprint.

Now she opens SpeedGrader. The proposal is there, 17 of 20, each criterion with its points, and under them two short lists: shows understanding of, and still lacks, each line quoting the student's own words. She reads the glance, opens one span, and signs. She selects, publishes and posts. Canvas held the proposal out of the gradebook until she did.

OralIQ never replaces the instructor and never says who wrote the work. It asks the student about their own work so a person can sign the grade knowing the student understands what they submitted.

What the instructor sees, on Canvas's own pages.

Three captures from the open-source Canvas, run on one laptop. Every submission, student and course is a fixture, and every capture says so.

  • SpeedGrader before the instructor acts: a short synthetic essay on the left; on the right the grade field is empty and marked Hidden, a rubric is attached, and a comment from the OralIQ service user carries a transcript file and a recording file
    SYNTHETIC. SpeedGrader before the instructor acts: no grade, the rubric, and the transcript and recording filed on the submission by the OralIQ service user.
  • The Canvas moderation page for the same assignment: one grader, the OralIQ service user, with a provisional score of 17; the final grade column is empty; the Release Grades and Post to Students buttons at the right
    SYNTHETIC. The moderation page: the proposal sits as a provisional grade of 17, the final grade is empty, and only the instructor's Release and Post can move it.
  • SpeedGrader after the instructor selected, published and posted: the grade field reads 17, the header shows 17 out of 20, 85 percent, and the transcript and recording are still on the submission
    SYNTHETIC. After the instructor selected, published and posted: 17 of 20 in the grade field, placed there by Canvas's own motion and not by the tool.

The artifact is no longer evidence.
The student is.

What each of them has the next morning.

  • The college

    A grade with a record behind it: what the student said, what the instructor signed, kept one year, on a line it already budgets.

  • The instructor

    Backed instead of being a detective. Four to six minutes a record. For a folder of 140, a day of reading instead of a week of calls.

  • The student

    Asked, not accused. Later, through a priced add-on, told what they still need to learn.

$4 a paper check. You pay only for proposals nobody rejected.

Prepaid credits, bought any month and sized to your own purchasing thresholds. The card runs by what is at stake in the assignment and tops out at $20, under what you already pay for one live-proctored exam.

See the rate card

What we will publish.

Nobody has measured whether a spoken check on a student's own work measures understanding. Not the one pilot with public numbers, which measured whether three models agree with each other. Not us.

So the first measured term prints the numbers: authors against readers, two graders' agreement with each other and with the machine, and the acceptance rate with the share accepted unread. We will publish them whichever way they fall.

Run OralIQ on one course next term.

A department block on one written quote. The check attaches to the next rubric-bearing assignment, and the instructor signs every score.

The artifact is no longer evidence. The student is.