Successful Mathematics
Learning that AI can't fake
For a century, homework worked the same way: practice at home, hand in the answers, get a grade. A right answer was good evidence that a student understood. Today any phone can produce a right answer in seconds, so that evidence is gone. The answer is no longer the proof.
What changed
48% to 62%
of U.S. middle school, high school and college students used AI for homework, from May to December 2025.
67%
of those students said using AI for schoolwork harmed their critical thinking, up from 54% earlier in the year.
13% to 26%
of U.S. teens used ChatGPT for schoolwork, doubling in a single year (2023 to 2024).
17% lower
exam scores for high school math students who practiced with an unrestricted chatbot, while they believed they had learned just as much. A tutor built to give hints instead of answers did not cause the harm.
Why it matters
Math is learned by doing it. When a chatbot does the practice, the student skips the part that builds the skill, and often does not notice: in the study above, students who leaned on a chatbot felt they had learned as much as everyone else, then did worse on the exam.
Teachers and parents lose something too. A page of right answers used to tell them who was ready to move on. Now it might mean that, or it might mean nothing. Banning AI does not fix this, and neither does guessing who used it. What has to change is what we check.
What COGENT does today
Your own numbers
Questions with changing numbers give every student their own version, so a classmate's answer, or one copied from a shared chat, does not fit.
A fair-play check that looks at patterns
No camera, microphone or screen recording. It notes things like leaving the page mid-question and answering the moment you come back, pasting into an answer, and right answers far faster than the student's usual pace, and it puts them in front of the teacher or parent with the innocent explanations alongside.
Check questions
One click sends a student a fresh question on the same skill, with new numbers. Explaining it again, on their own, is the clearest evidence either way.
Home against supervised
Teachers can compare how a student does working alone with how they do in supervised work. A large gap is worth a conversation.
What we are building
The bigger change is a different idea of what a grade is for: showing that the student can do the math and explain it, not that an answer box was filled. These are being designed now.
Being built
Grading the method, not only the answer
A chatbot can hand over a final answer. It cannot hand over the student's own steps as they worked. We are designing grading that looks at how a problem was solved: the steps taken, where the student corrected themselves, and where they got stuck.
Being built
Explain it out loud
A spoken check-in, like a teacher asking "how did you get that?". The student explains their reasoning in their own words, and the explanation is what earns the credit.
Being built
Practice that builds the skill
Help that gives the next hint instead of the answer, so practice still does what practice is for. That is the design the research above found does not harm learning.